app: learn probe durations per device; stop claiming VPN after teardown

The ETA table was fixed at compile time, but real durations are a property
of this phone and the network it stands in - ICMPv6 answers in
milliseconds where IPv6 works and waits out its timeout where it does
not. Estimates now prefer an EMA (70/30) of what this device actually
measured, seeded by the old constants on first run.

VPN detection had two lies in it, both found on hardware: a disconnected
tunnel lingers in allNetworks while tearing down, so 'VPN active' is now
judged from the ACTIVE network only; and any bind failure counted as
'per-network blocked', so a network dying mid-run flipped a healthy run
to INCONCLUSIVE - now only a genuine EPERM refusal counts. Banner and
finding split three ways (VPN + blocked / blocked only / VPN only) so the
app never names a VPN the user just turned off.

App version 0.2.3.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

.gitea/workflows/build-server.yml

@@ -15,7 +15,13 @@
 #                        container registry (docker login → unauthorized).
 #                        Create: user Settings → Applications → Generate token.
 #   REGISTRY_USER        optional; defaults to the pushing actor's username.
-# The release job needs only the built-in GITHUB_TOKEN.
+#   RELEASE_SIGNING_KEY  base64 ed25519 seed that signs SHA256SUMS. Self-updating
+#                        servers verify the signature against the public key baked
+#                        into the binary (selfupdate.DefaultPublicKeyB64) and REFUSE
+#                        unsigned releases, so this job hard-fails without it —
+#                        a release nobody can install is better failed loudly here.
+#                        Mint a pair with: go run ./cmd/release-sign -gen
+# The release job otherwise needs only the built-in GITHUB_TOKEN.
 
 name: server-release
 on:
@@ -44,6 +50,18 @@ jobs:
           done
           (cd ../dist && sha256sum * > SHA256SUMS)
 
+      - name: Sign SHA256SUMS
+        working-directory: server
+        env:
+          RELEASE_SIGNING_KEY: ${{ secrets.RELEASE_SIGNING_KEY }}
+        run: |
+          [ -n "$RELEASE_SIGNING_KEY" ] || { echo "::error::secret RELEASE_SIGNING_KEY is missing — self-updating servers refuse unsigned releases, so publishing one would strand the fleet. Add it under Settings → Actions → Secrets."; exit 1; }
+          go run ./cmd/release-sign ../dist/SHA256SUMS
+          # Verify with the key baked into the binary we just built — catches a
+          # secret that does not match DefaultPublicKeyB64 before it ships.
+          PUB=$(grep -o 'DefaultPublicKeyB64 = "[^"]*"' internal/selfupdate/selfupdate.go | cut -d'"' -f2)
+          go run ./cmd/release-sign -verify -pub "$PUB" ../dist/SHA256SUMS
+
       - name: Create release + attach binaries
         env:
           TOKEN: ${{ secrets.GITHUB_TOKEN }}

CLAUDE.md

@@ -145,6 +145,17 @@ First build downloads AGP/Compose/Shizuku from Google Maven + Maven Central.
   Shizuku, **toggle Wireless debugging off/on** — Shizuku keeps running (separate process), a fresh
   port + mDNS record appear, and the beacon/connector recover. Plan the Shizuku-tier dev loop
   around this (or USB, if ever available).
+- **A poisoned Gradle *build cache* entry can silently drop a whole module from the APK.**
+  Symptom: the app dies with `ClassNotFoundException` for a class that plainly exists, while the
+  build is green and `./gradlew :app:dependencies` lists the module on `debugRuntimeClasspath`.
+  The module's own jar is correct; its code simply never reaches AGP's intermediates. `clean`,
+  `rm -rf */build` and `--rerun-tasks` all fail to fix it, because **none of them touch the build
+  cache** — look for `compileKotlin FROM-CACHE` in the log. Fix: rebuild with `--no-build-cache`.
+  Verify by grepping the APK's dex for a string literal that only that module defines; grepping for
+  a *class name* proves nothing, because callers carry the name as a reference whether or not the
+  class is packaged:
+  `unzip -o -q app-debug.apk "classes*.dex" && grep -a "pin-sha256:" *.dex`
+  Suspect this whenever a runtime failure contradicts a successful build.
 - **Empty-jar race with the IDE.** VSCodium's Java/Kotlin extension runs its own Gradle daemon on
   the same project; when it overlaps a CLI build, a module's `build/libs/*.jar` can end up
   containing only a manifest, and Gradle then considers `jar` up-to-date. Dependent modules fail

docs/build-status.md

@@ -211,8 +211,8 @@ Two collection-loop gotchas found while driving the phone over USB:
 2. ~~If `trace.errqueue_reachable` = PARTIAL, add a C-over-JNI errqueue shim.~~ **Retired** —
    SUPPORTED on both known devices; `traceroute.udp4` reads real hops via `Os.recvmsg` +
    `StructMsghdr` reflection, so no `:native` module is needed.
-3. Start the Go server skeleton (enrollment + profile + sessions + UDP echo with observation
-   blocks + canary-DNS reference records) per probe-protocol.md.
+3. ~~Start the Go server skeleton per probe-protocol.md.~~ **Shipped** — live on fmr since
+   v0.2.0 (2026-07-31); see the server sections below.
 4. Fold confirmed capabilities into the production `core-probe` / `core-shizuku` modules.
 
 ## Production probe server — LIVE on dedicated VM "fmr" (2026-07-31)
@@ -222,7 +222,7 @@ SSH only — verified untouched by the daemon (explicit multi-address binds, no
 Control: fmr-1:8443 (SPKI pin `zRV9qkiLnRexAeh4RrSfJzbPWO+U/2Oj2/NVM/KfXlg=`, verified
 externally over v4+v6). UDP data plane on all four service addresses :8442 — the second IP is
 the stun-5780 substrate. Daily randomized self-update timer installed (checksum-verified
-against SHA256SUMS; signature verification still TODO before treating the source as untrusted).
+against SHA256SUMS; signature verification landed 2026-08-02 — see "Release signing" below).
 Host config in `/etc/echolot-server.env`. SSH access for sessions: `ssh claude-echolot`.
 
 ## Server v0.3.0 — STUN + TCP echo + observations + actions (2026-07-31)
@@ -728,3 +728,642 @@ tenfold asymmetry that a round-trip measurement cannot see at all.
 
 10 unit tests on the arithmetic (a wrong denominator here does not crash, it produces a plausible
 number pointing at the wrong half of the network) plus the live correlation check.
+
+### frag_send: crafted IP fragments, so *ordering* is testable (server-v0.6.0, 2026-08-01)
+`big_send` with `df=false` answers one question — do fragments get through. It cannot answer the
+more interesting one, because the kernel always emits fragments in order, first one first.
+
+The classic middlebox fault is exactly about that ordering. Only the **first** fragment carries the
+UDP header, and therefore the ports; a stateful firewall or NAT that has not seen it has no flow to
+match the rest against, and many simply drop them. That is invisible to every in-order test, and in
+the field it looks like "large DNS answers fail on this network" or "the tunnel breaks when the MTU
+drops" — it works until the network reorders, then fails intermittently, which is the hardest kind
+of fault to chase.
+
+So the server builds the fragments itself (raw socket, `IP_HDRINCL`) and controls their order:
+`in_order` (baseline), `reversed` (last fragment first), `first_last` (first fragment held back
+250 ms). The datagram is assembled and **signed whole** before being cut up, so what the client
+reassembles is indistinguishable from an ordinary packet — otherwise the test would be measuring
+our sender rather than the path. New test type `mtu.frag_ordering`; findings
+`mtu.fragments_blocked` and `mtu.fragment_reorder_sensitive`.
+
+Two details that would otherwise produce confidently wrong answers:
+- **The UDP checksum is computed, not left zero.** Zero is legal in IPv4 and would be less code,
+  but zero-checksum datagrams are dropped by some middleboxes — and that drop would be recorded as
+  a fragmentation failure, which is the wrong conclusion entirely.
+- **Fragment offsets are in 8-byte units**, so non-final fragments are rounded down to a multiple
+  of 8. A 100-byte fragment is not an error; it is a datagram no host will ever reassemble.
+
+`frag-send` is advertised only when a raw socket can actually be opened — checked by opening one,
+because a permission model has more ways to say no (userns, seccomp, LSM) than a capability bit has
+to say yes. fmr runs as root with `cap_net_raw` in its bounding set, so it is available there.
+
+Fragment ordering runs only after `mtu.frag_delivery` shows fragments arrive at all; otherwise the
+three orderings would each report "not delivered" and read as three faults instead of one.
+
+The header arithmetic is unit-tested (reassembly coverage with no gaps or double-delivery, MF
+flags, shared IP ID, 8-byte offsets, checksum verification over odd and even lengths). Because the
+code is `//go:build linux`, the tests are **cross-compiled and run on fmr** — there is no Go
+toolchain there, so `go test -c` plus scp is the loop.
+
+Live against fmr: 4 fragments per burst, and all three orderings reassembled — a healthy path, and
+the baseline against which a mobile network will be interesting.
+
+### Testing state (2026-08-01)
+Six live tests against fmr, all green, no device involved: `LiveServerTest`, `LiveMeasurement`,
+`LiveGranted`, `LiveDownstream`, `LiveUpload`, `LiveCompat`, `LiveEnrollment`. Plus 74 client unit
+tests and the full Go suite. Everything in the last several entries is verified from the PC; the
+app's UI (settings, history, deep-link enrollment) and `mtu.pmtud_up` remain device-only.
+
+### throughput: a rate, plus the qualifier that makes it a measurement (server-v0.6.1 … v0.6.2)
+A throughput test reports the *smallest* limit on the path — and the sender's own ceiling is one of
+the candidates. If the server is asked for 50 Mbps and 50 Mbps arrives, the network was never the
+constraint and "50 Mbps" says nothing about it. So `perf.throughput_udp` always carries
+`limited_by` (duration | budget | rate | send_error) and `measures_network`, and a finding is
+raised only when the path is actually implicated. The live run against fmr reports 20 Mbit/s with
+`measures_network: false`, which is the correct and useful answer.
+
+Loss is computed against the **sender's own count**, fetched from the observations API, not against
+the requested rate. A receiver alone cannot tell "the network dropped it" from "the sender never
+sent it", and guessing turns a healthy server-side limit into a phantom network fault. The server
+keeps one summary per action rather than per-packet records — a ten-second run at 50 Mbps is half a
+million packets, and a struct each would turn a measurement into memory exhaustion.
+
+Sending is **paced**, on an absolute schedule. Unpaced would measure the server's NIC and the first
+queue it meets, then collapse into loss that reads as a network fault; sleep-per-packet would
+accumulate scheduler error and drift the rate down over a ten-second run.
+
+Throughput gets its own grant budget sized from the request, so every *other* action stays bounded
+at 8 MiB. When the byte cap binds before the clock does, the **duration is shortened and reported**
+rather than the run being truncated: promising thirty seconds and delivering twenty-one is the same
+information with a surprise attached, and it keeps "the clock ended the run" as the normal case —
+the only case where the rate is a clean property of the path. That behaviour came out of a test
+that failed honestly (30 s at 100 Mbps needs 375 MB against a 256 MB cap).
+
+It is **opt-in** in the run config, default off. A 5-second run at 50 Mbps moves ~30 MB; on a
+metered mobile connection that is the user's money, and a tool that spends it without being asked
+is not one people keep installed.
+
+#### The bug the live test found
+The first live run delivered 104 packets and stopped after 50 ms. The grant's rate check exempted
+the first 50 ms entirely, meaning to be lenient at startup — the effect was the opposite. A sender
+could dump an unbounded burst into that free window, and the instant the check switched on it
+compared those bytes against 50 ms worth of allowance and refused everything until real time caught
+up. **Every short test passed** (downtrain sends 50 packets, big_send seven); every sustained send
+died fifty milliseconds in.
+
+Replaced with a token bucket (`allowance = burst + rate × elapsed`), which is smooth from t=0.
+The burst is 100 ms of the allowed rate, floored at one ordinary datagram — deliberately one, since
+at 8 kbps a 64 KB floor is sixty-four seconds' worth, exactly the instant dump the ceiling exists to
+prevent. The pre-existing rate test caught that when I first tried the generous floor, and it was
+right to. Second half of the same bug: callers treated *any* refusal as terminal, so `TryAllow` now
+says why — a sender paces through a transient "too fast just now" and still stops dead on a spent
+budget or an expired grant. Both halves are pinned by regression tests.
+
+### Findings registry (2026-08-01)
+Closes open item 1 of measurement-schema.md §9. A finding code is the stable, machine-readable half
+of a result — what a dashboard groups by and what someone greps a year of archived runs for — and
+that only holds if a code means exactly one thing forever. Ad-hoc string literals at fifteen call
+sites cannot promise that, and by the time the registry was written the failure had already
+happened.
+
+**Two emitters had independently produced `connectivity.downstream_loss` and
+`connectivity.loss_downstream` for the same claim**, and nothing anywhere objected. Anyone
+aggregating either one would have silently seen half their data. Merged into
+`connectivity.loss_downstream`, paired with `loss_upstream` so the two directions read as a set.
+
+**Two codes were also renamed out of `nat.*`.** `nat.udp_unreachable` is not about NAT — it means
+no replies came back — but the prefix determines the category, and the category determines which
+verdict light the finding rolls up into (§7.3). A `nat.*` code landing under *connectivity* is not
+a naming quibble; it changes which light turns red. Cheap to fix now, a breaking change later.
+
+Codes are now declared as typed `FindingSpec`s carrying their category and default severity, and
+emitters reference the spec instead of retyping the string — so a typo is a compile error and two
+call sites cannot disagree about a finding's category.
+
+`docs/findings-registry.md` is the contract, and a test reads it: it fails when the document and
+the registry have codes the other lacks, or when a severity differs. Documentation that drifts from
+its implementation is worse than none, because it still looks authoritative. The check scopes
+itself to table rows, so the prose can keep explaining which codes were retired and why.
+
+Six tests: uniqueness, declared-vs-listed, prefix↔category agreement, naming convention, a
+word-order-anagram check (the shape the duplication actually took), and the document agreement.
+
+### A real privacy leak, found by starting on the machine-readable schema (2026-08-01)
+The intent was `measurement.schema.json` (§8's promised companion). The first step — checking
+whether the anonymizer actually covers the fields the schema declares as sensitive — found that it
+did not, so that became the work.
+
+**At the `balanced` level, five identifying values were being uploaded verbatim:**
+
+| value | field | why it matters |
+|---|---|---|
+| `2001:…::150` | `networks[].link.addresses[].addr` | the device's own global IPv6 address — a strong, geolocatable device identifier |
+| `2a02:…::1` | `networks[].link.routes[].gateway` | identifies the ISP allocation |
+| `203.0.113.77` | `networks[].link.dns.servers[]` | the configured resolver |
+| `nas.example.lan` | `private_dns_hostname` | an internal hostname |
+| `example.lan` | `search_domains[]` | the internal domain |
+
+The settings screen describes that level as pseudonymizing addresses. It was not.
+
+**Root cause:** classification keyed on field *names*, and the schema's actual names (`addr`,
+`gateway`, `dst`, `servers`, `search_domains`, `private_dns_hostname`) had never been added to the
+table. Not a subtle bug — just an unfalsifiable design. The existing tests all passed, because each
+one checked a field somebody had remembered to write a case for.
+
+**Two fixes, one of them structural:**
+1. The missing names were added.
+2. More importantly, a **shape-based backstop**: when a field name is unrecognised, the *value* is
+   inspected, and anything shaped like an IPv4/IPv6 address or a MAC is treated as one. A name
+   table can only protect fields someone thought of, which is precisely the wrong property for a
+   privacy control. Hostnames are deliberately *not* inferred by shape — `train.udp_updown` is
+   indistinguishable from a domain, and mangling a test type would corrupt the document to protect
+   nothing.
+
+`LeakTest` is the new guard and is written to fail for fields nobody has considered: it plants
+identifying values wherever one can actually occur and asserts none survive, rather than checking
+a list of known cases. It also pins that RFC1918 addresses still come through readable, so the
+test cannot pass by over-redacting everything.
+
+Route prefixes and the unspecified address needed care in the transform: `0.0.0.0/0` and `::/0`
+must stay themselves, or a routing table becomes unreadable for no privacy gain.
+
+**Still outstanding:** `measurement.schema.json` itself. Worth noting what this episode implies for
+it — much of a document's payload lives in `evidence`/`metrics`/`params`, which are per-test-type
+`JsonObject` by design and therefore *outside* any schema. A schema-driven anonymizer would have
+less coverage there than the name-plus-shape one now does, so the schema should be built for
+validation and external tooling, not as a replacement for the classifier.
+
+### ULA prefixes are pseudonymized whole (2026-08-01)
+Spotted in a real uploaded run from the phone: the server had
+`fda1:3fb1:ff92:6696::2662` for a DNS server. The general IPv6 path preserves the leading two
+groups (deliberately — for a global address that keeps the ISP allocation, which is the
+diagnostically useful part), and for a ULA that passed through **32 of the 40 random bits** of the
+global ID.
+
+ULA looks like the v6 equivalent of RFC1918 and the instinct is to treat it the same. That
+reasoning does not carry over, and the difference is the whole point: an RFC1918 prefix is shared
+by millions of networks and identifies none of them, while a ULA global ID is random and unique to
+one network by construction (RFC 4193). The prefix *is* the identifier — it is a network
+fingerprint that was surviving redaction.
+
+Now pseudonymized as a unit, so two addresses on the same ULA subnet still land on the same
+pseudonymous prefix: "these hosts are on one network" survives, "this is *that* network" does not.
+Three tests, one of which uses the exact value observed on the wire.
+
+Worth recording as a reasoning trap: I had originally raised this as "ULA should probably be kept
+verbatim, like RFC1918, for consistency". The surface analogy pointed the wrong way, and the
+correct answer was the opposite.
+
+### Registry adopted everywhere; v6 findings renamed; Back works (2026-08-01)
+The findings registry was only adopted in `core-engine`. The app module still emitted seven codes
+as raw strings, so the registry test passed while codes existed outside it — including
+`ipv6.broken`, which fired on a real network and was in no registry at all.
+
+All seven now reference registry entries for code, category and severity, so those three cannot
+disagree at a call site. A grep for `code = "…"` across the app, engine and probe modules returns
+nothing.
+
+**`ipv6.*` → `v6.*`.** The third instance of rule 1: they declared `Category.IPV6` while the prefix
+map only knows `v6`, so `TestType.category("ipv6.broken")` fell through to *connectivity* and the
+finding rolled up under the wrong verdict light. The test-type registry already used `v6.`.
+
+Two severities reconciled while merging:
+- `connectivity.captive_portal` is **medium**, not high. The registry had guessed high; the probe
+  that emits it had always said medium, and the probe was the considered value — a captive portal
+  on hotel wifi is what should be there, and logging in clears it. `connectivity.no_internet` is
+  the high one, because nothing the user does locally fixes that.
+- `v6.not_offered` is **info, and the registry says it must stay info**. Most networks still do not
+  offer IPv6 and that is not a fault; a warning here lights a yellow verdict on a healthy network,
+  which teaches people to ignore the light.
+
+Also: a `BackHandler` now returns from Settings/History to the run screen. The screen was a plain
+state variable with nothing connecting it to the back stack, so the system Back gesture left the
+app entirely. Enabled only when there is somewhere to go back to, so Back still exits from the run
+screen.
+
+### Upstream throughput (server-v0.6.3, 2026-08-01)
+The mirror of the downstream case: the client generates the traffic and the server counts it. No
+grant is involved — the client is sending its own packets, so there is nothing to amplify — but it
+does need the server's tally, because **only the far end knows how much arrived**. Without that
+number a sender measures how fast it can *transmit*, which is usually just the speed of the local
+NIC and is a different question from the one being asked.
+
+`TYPE_THROUGHPUT_UP` (0x0F) is counted and deliberately **never answered**: a reply would double
+the traffic and drag the return path into a measurement that is specifically about the outbound
+one.
+
+The tally is a counter, not a list, and short-circuits **before** the observation log. A
+five-second run at 20 Mbps is around ten thousand packets; one struct each would turn a
+measurement into an allocation storm on a shared server, and nothing needs the per-packet detail
+since the client holds the send-side record. The gap between the two counts is the loss.
+
+`direction=up` on the throughput action sends nothing — it zeroes the counter, so a second run in
+one session measures itself rather than inheriting the first one's packets. The live test asserts
+`received <= sent`, which is what catches a counter that was never reset.
+
+Live against fmr: **3125 sent, 3125 counted, 0 % loss, 10.0 Mbit/s** at a 10 Mbit/s request, with
+`measures_network: false` — correct, since what arrived matched what was offered, so the path was
+never the constraint.
+
+### Raw shell dumps leaked the whole LAN (2026-08-01)
+Found by running the Shizuku shell tier for the first time. The tier works — `tiers.shizuku: true`,
+`exec_path: UserService` (so the UserService binds on the OnePlus, as recorded), `runs_as
+shell(2000)`, 7/7 commands — and the run promptly uploaded **every MAC address on the local
+network** to fmr at the `balanced` level: router, phones, whatever else was on the wifi. Fourteen
+of them.
+
+The probes embed raw command output verbatim (`ip neigh`, `ip route`, `id`), which is genuinely
+good evidence and also a complete household device inventory. The anonymizer could not see it:
+classification is by field name and by whole-value shape, and `ip_neigh` is one long string that is
+itself neither a MAC nor an address. measurement-schema.md §9 item 2 had flagged raw dumps as "hard
+to anonymize" and proposed dropping them from exports; nothing enforced either.
+
+**Scrubbing beats dropping.** Identifiers inside any unclassified string are now replaced in place,
+using the same pseudonyms as everywhere else — so a MAC that appears both in a parsed field and in
+a raw dump still reads as one device. The dump stays readable and auditable: you can still see the
+neighbour table's shape, the host count, RFC1918 addresses and vendor prefixes. Dropping the
+evidence would have protected the same data while destroying the reason for collecting it.
+
+Two implementation notes worth keeping:
+- **One pass, not three.** Sequential passes re-process their own output: once a MAC became
+  `78:9a:18:xx:yy:zz`, the IPv6 pattern matched it — six hex groups separated by colons *is* an
+  address — and destroyed the vendor prefix the MAC rule had just preserved. Ordered alternation
+  resolves each position once, MAC first.
+- The patterns are conservative on purpose. A missed address gets caught by another rule or not at
+  all; an over-eager one mangles timestamps and version strings, corrupting evidence to protect
+  nothing.
+
+`RealDocumentTest` runs the anonymizer over a captured run when `ECHOLOT_REAL_RUN` points at one,
+and fails on any MAC that survives. It self-skips otherwise, so no one's network is committed to the
+repo. Against the actual leaked document: **14 MACs in, 0 surviving.**
+
+Also fixed: the Settings *Preview what an upload would send* button did nothing. It read
+`UiState.history`, which is empty until the History screen has been opened — the same root cause as
+the "0 run(s)" count. It now reads the archive directly, and says so when there is nothing to
+preview rather than silently ignoring the tap.
+
+### Security: the admin listener was publicly exposed for ~15 minutes (2026-08-01)
+Moving the admin listener to `[::2]:443` for the UI exposed `/admin/enroll-tokens` and
+`/admin/selftest` to the internet **with no authentication**. Anyone who could reach
+`fmr.echo-lot.app` could mint enrolment tokens.
+
+The listener was designed localhost-only — its own flag help says *"keep localhost"* — and that
+assumption travelled with it when the address changed. The compounding error: `checkAdminExposure`,
+added the same day, verifies **encryption** and says nothing about **authentication**. It passed,
+and a green light on an adjacent property is worse than no check, because it invites you to stop
+looking.
+
+Closed by returning to loopback (the TLS and ACME work is retained, just not exposed). All 68 device
+enrolments matched the timestamps of test runs, so there is no evidence of abuse — but the window
+existed on a freshly published hostname and absence cannot be proven. 39 unused enrolment tokens
+were purged, since any could have been minted by someone else and they cost nothing to replace, and
+63 test devices removed.
+
+**The admin listener does not become reachable again until it authenticates.** That reorders the UI
+work: auth on the listener first, everything else after.
+
+### Open: encrypted uploads, where the operator cannot read the data
+Not built. Recorded because the shape is decided by a few early choices, and the current design
+happens to leave the door open.
+
+The goal: hand someone an account, let them upload, and be unable to read what they uploaded.
+
+Sketch: a random per-account **master key**, generated on the first device and wrapped under a
+key derived from a passphrase (PBKDF2-HMAC-SHA256 — stdlib on both sides). The wrapped key is
+stored server-side as an opaque blob, so a new device signs in, fetches it, and unwraps locally;
+the server never sees either key. Runs are encrypted client-side with AES-256-GCM, fresh nonce per
+run. All of this is stdlib in Go and `javax.crypto` in Kotlin — no dependency either side.
+
+Four consequences that decide whether it is worth it:
+
+1. **What stays readable determines what the UI can do.** The server builds its index by *parsing*
+   the document — verdict, finding count, started_at. An opaque payload means the client supplies
+   that metadata or the index disappears, and with it retention-by-verdict and any "runs with
+   findings" view. The honest version supplies only run id, timestamp and size, and moves the rest
+   client-side.
+2. **Lose the passphrase, lose the data.** That is the feature working, and also the support
+   burden. It needs a recovery code printed at setup, not a reset flow — there is nothing to reset.
+3. **Metadata is not hidden.** The operator still sees which account uploaded, when, how often and
+   how large. "Cannot see it" is about content, not existence, and saying otherwise would oversell.
+4. **It makes `min_anonymization` unenforceable** — a server cannot check a level it cannot read.
+   That is not a conflict so much as a redundancy: the anonymization floor exists to protect the
+   user from the operator, and encryption does that better. The two should not both be demanded of
+   one upload.
+
+What keeps this possible: uploads are already stored byte-for-byte as received, and every index
+field is derived in one function (`runs.Put`). The thing to avoid is admin features that *require*
+reading content — those would have to be unbuilt later.
+
+### App sign-in, and an undisclosed dependency it surfaced (2026-08-01)
+The app can now sign in to the server's identity provider: authorization code with PKCE, a
+`Sign in` card in settings, and the `echolot://auth` redirect handled alongside the enrolment one
+(told apart by host, since one spends a token and the other completes an authorization).
+
+The detail that decides whether this works on a real phone: **the PKCE verifier is written to
+storage before the browser opens**, not held in memory. Handing control to a browser backgrounds
+the process and Android may kill it; the callback then arrives at a fresh process. An in-memory
+verifier works on a developer's device and fails under memory pressure, which is the worst way for
+a sign-in to break.
+
+Nothing from the IdP is retained. The ID token proves who is signing in, once, and the device
+credential authenticates everything after — no access tokens stored, no refresh tokens rotated.
+
+**A server remains entirely optional.** All eight probes are device-tier; `serverConfigured` gates
+only upload and the account. But answering that question exposed something worth fixing: two probes
+hardcode the reference deployment —
+
+```kotlin
+DnsCanaryProbe(canaryZone = "c.echo-lot.app", ...)   // "Hardcoded to the reference deployment"
+StunProbe(serverHost = "fmr-1.echo-lot.app")
+```
+
+so a user with no server of their own still sends DNS and STUN traffic to fmr without being told.
+For a tool that goes to this much trouble over what leaves the device, an undisclosed dependency on
+a third party's infrastructure is the wrong default. It should prefer the configured server, and be
+explicit when there is none. **Closed** — both probes take the enrolled server from settings
+(canary zone learned from the profile, cleared on re-enroll) and report themselves SKIPPED with
+the reason when none is configured.
+
+### v6.broken was a false positive waiting to happen (2026-08-01)
+A phone could not open `https://fmr.echo-lot.app` while loading the same server by IP literal
+perfectly well. Two things came out of chasing it.
+
+**The admin UI is IPv6-only, by consequence rather than intent.** `fmr.echo-lot.app` has an AAAA
+and no A record — verified identical at Cloudflare, Google and Quad9, so DNS itself is healthy.
+That follows from reserving all four measurement addresses for testing, which left only `::2` for
+management, and `::2` has no IPv4 counterpart. Any client without working IPv6 sees an unreachable
+admin interface — a poor property for the interface you reach *from the networks you are debugging*.
+
+**And the app's own `v6.broken` finding was unsound.** It fired on exactly one signal — ICMPv6 echo
+getting no reply — with `Confidence.HIGH`. ICMPv6 echo is widely filtered on networks where IPv6
+works fine, which is precisely what that phone demonstrated: no ICMPv6 replies, working IPv6 TCP.
+The finding asserted a cause it had no evidence for, which is the same class of error as the
+multi-homed `100 % downstream loss` earlier: a confident measurement of something that was not
+happening.
+
+Now `v6.no_icmp_reply`, severity low, confidence medium, and the text names *both* explanations
+instead of choosing one. It is still worth reporting, because filtered ICMPv6 breaks Path MTU
+Discovery — large packets vanish rather than being reported as too big — which is a real fault even
+when IPv6 works.
+
+The proper fix is corroboration: attempt a real IPv6 connection and only call it broken when that
+fails too. That needs a target, which runs into the hardcoded-reference-deployment issue already
+open above. **Both closed 2026-08-02** — see "Corroborated IPv6 findings" below.
+
+## Per-network probing is blocked while a VPN is up (2026-08-01)
+
+`Network.bindSocket()` fails with `EPERM` for every underlying network when a VPN holds the
+default route — verified on the OnePlus 15 with Netbird active: `Binding socket to network 101
+failed: EPERM` for both cellular and wifi. This is Android preventing VPN leaks, not a bug to work
+around, and it means the whole per-network measurement approach is unavailable to any user with a
+VPN connected. Worth deciding deliberately rather than discovering per report:
+
+- The run currently succeeds and simply measures nothing per network. Honest, but silent — the
+  document records `attempted: false` and the UI says green.
+- A user with a corporate VPN permanently on would get a green run that measured almost nothing.
+
+Options are to detect the VPN and say so plainly ("this network cannot be measured while a VPN is
+active"), to measure the tunnel itself as the network under test, or both. **Decided and built
+2026-08-02**: say so plainly, everywhere the run is read — see "Constrained runs" below.
+
+Related: `icmp.ping6` now records `attempted` alongside `ok` per network, because collapsing them
+made the app report "IPv6 is configured, but ICMPv6 gets no reply" about an interface it had never
+succeeded in sending on — a claim about the user's carrier with no evidence behind it.
+
+## Reserved measurement addresses, and the web UI on both families (2026-08-01)
+
+fmr has two IPv4 (.150/.151) and three IPv6 (::150/::151/::2) addresses. `.150`/`::150` now carry
+the services; `.151`/`::151` are reserved for measurement, declared in `ECHOLOT_RESERVED_ADDRS`.
+
+Reserved does **not** mean silent. The UDP data plane, the canary DNS and STUN's RFC 5780 alternate
+all belong there — reserving an address and then forbidding the measurements that need it would
+defeat the purpose. What must never appear is a service, and above all not ports 80 or 443: a
+handshake completing on a port known not to be listening is what proves interception, and that
+proof survives exactly as long as nothing binds those ports. `config.CheckReserved` enforces it at
+startup, refusing wildcard binds outright (every listener defaults to `:port`, so the next one added
+will claim reserved addresses without anyone deciding to).
+
+The first version of the guard was too strict and the live config caught it: it would have refused
+the existing UDP and DNS binds on `.151`. The rule is about services and web ports, not about
+listening at all.
+
+**The adb-beacon receiver was wildcard-bound to `0.0.0.0:443`**, occupying port 443 on every IPv4
+address including the reserved one — so the IPv4 interception test had been compromised for as long
+as it had been running, silently. It is now `systemctl disable --now echolot-adb-beacon`; restore
+with `systemctl enable --now`. Note what this implies: the guard covers this server's own listeners,
+and a stray process outside its config can still pollute a reserved address. A startup probe that
+*verifies* 80/443 are actually free on the reserved addresses would be a stronger guarantee than
+checking our own configuration — built 2026-08-02 (`selftest.ReservedWebPortsFree`, fatal at
+startup when anything is listening there).
+
+The admin UI and the ACME responder now take comma-separated addresses like every other listener;
+they were single-address, which is why the UI could only ever live on `::2`. It serves on `.150:443` and
+`[::150]:443`; sshd on `.150:2322` and `[::150]:2322`.
+
+`::2` is gone entirely — unbound, then removed from `/etc/systemd/network/ext.network`. The
+transition kept it bound throughout and dropped it only after the CNAME landed, because removing it
+first would have broken both the UI and ACME renewal for the very name the certificate is issued
+to. Listeners came off before the address did, in that order, or the services would have failed to
+bind on restart.
+
+Verified after a full reboot: `fmr.echo-lot.app` answers 200 over both families, `.151`/`::151` are
+closed on 80 and 443, canary DNS is still up on `.151`, and neither `::2` nor the beacon returns.
+(`echolot-server` is `After=network-online.target` with `Restart=on-failure`, which is what makes
+binding specific addresses safe across a boot — a wildcard bind would not have needed it, and that
+is the trade for the reserved addresses being meaningful.)
+
+The point of all this: `fmr.echo-lot.app` gained an A record, so the server stopped being reachable
+only over IPv6 — which is what made it unreachable from a phone with no working IPv6, presenting as
+"this host does not exist" in two different browsers.
+
+### If the beacon comes back, it belongs in the web UI
+
+Not as a separate listener. The receiver being its own Python service on `0.0.0.0:443` is exactly
+what silently compromised the reserved address, and a second process racing for a port is a
+recurring problem rather than a one-off: whoever loses the race simply fails to start, and on a
+reboot which one that is comes down to unit ordering.
+
+Folding it in costs little and settles several things at once. It would be two routes on the admin
+UI (`POST` the observed adb port, `GET /apk` for the staged build), behind the TLS the UI already
+terminates and the certificate it already renews, with no extra port and no wildcard. It also gets
+authentication for free — the current receiver accepts a port report from anyone who can reach it,
+which is tolerable for a dev tool on a trusted network and not something to keep once it lives
+beside an admin session.
+
+The one thing that changes on the device side is that the POST becomes HTTPS. That is a real
+certificate rather than a self-signed one, so it costs a URL scheme rather than any trust plumbing.
+
+## The control plane shares port 443 (2026-08-01)
+
+`fmr-1.echo-lot.app:443` is the control plane, `fmr.echo-lot.app:443` the admin UI, both on
+`.150`/`::150`, one listener, selected by SNI for the certificate and by `Host` for the handler.
+
+The reason is not tidiness, it is reachability. Captive portals, hotel wifi and corporate firewalls
+routinely permit only 80 and 443 — which is exactly the population of networks this tool exists to
+diagnose. A control plane on 8443 is unreachable precisely when it matters most, and it fails as
+"cannot reach server", which tells the user nothing.
+
+They cannot share a certificate, which is why this needs two names. The control plane is trusted by
+SPKI pin and so uses a long-lived self-signed certificate; a browser needs one a CA vouches for.
+One name on one port is one certificate, so the port can only be shared by splitting the names.
+Pinning the Let's Encrypt key instead was considered and rejected: it survives renewal only while
+key reuse holds, so a routine key rotation would brick the whole fleet.
+
+Verified per SNI on 443: `fmr.echo-lot.app` serves `issuer=Let's Encrypt`, `fmr-1.echo-lot.app`
+serves the self-signed cert whose pin is unchanged (`zRV9…Xlg=`), `/v1/profile` answers 401 on the
+control name and 303 to the login page on the UI name.
+
+**8443 stays open.** Devices enrolled before this carry that URL in their settings, and closing it
+for the sake of a port number would strand every one of them. It can go once no enrolled device
+still points at it — not before.
+
+The rule from the naming change still binds: `fmr` may be a CNAME to exactly one host and never a
+multi-address record, because a pinned client that reaches a different key does not fail over.
+
+## Constrained runs: a VPN'd run now says so, everywhere (2026-08-02, app 0.2.1)
+
+The measurement schema gained a top-level `constraints` block (§3) and the app now fills it.
+`ConstraintDetector` (core-probe) runs before any probe: one throwaway `Network.bindSocket()` per
+non-VPN network, plus a transport check for an active VPN. The result lands in three places, and
+all three are deliberate:
+
+- **`run.constraints`** — for machines. A server aggregating thousands of runs can now separate
+  "measured a healthy network" from "measured almost nothing through a tunnel"; the shapes were
+  identical before.
+- **A `measurement.vpn_constrained` finding** — for the person reading this run, naming the
+  interfaces that went unmeasured. A constrained run with a quiet findings list still reads as
+  "nothing wrong here".
+- **The §7.3 verdict** — `Verdicts.derive` takes the constraints and returns INCONCLUSIVE
+  outright for a per-network-blocked run, whatever the category lights say; the run screen shows
+  an amber "Measured through a VPN" banner above the verdict so INCONCLUSIVE reads as the OS
+  refusing, not the app failing.
+
+Detection is one bind per network rather than parsing per-test `attempted:false` breadcrumbs, so
+it cannot drift when probe evidence formats change.
+
+## Corroborated IPv6 findings: v6.broken is back, with evidence (2026-08-02, app 0.2.1)
+
+The new `V6ConnectProbe` (test type `v6.brokenness`) attempts a real TCP connection over IPv6 to
+the configured server's :443, per network that *claims* IPv6 (global address or v6 default
+route) — IPv4-only networks are not attempted, since their failure is by design and would
+manufacture the exact false positive this exists to kill. The finding derivation is now three-way:
+
+- ICMPv6 silent, TCP works → `v6.no_icmp_reply` at **high** confidence, retitled "ICMPv6 is
+  filtered here — IPv6 itself works" (still reported: filtered ICMPv6 breaks PMTUD).
+- ICMPv6 silent, TCP fails too → **`v6.broken`** (high severity, reinstated in the registry +
+  findings-registry.md): two independent transports silent on a network advertising IPv6.
+- No corroboration (no server configured, or the connect never got as far as sending) → the
+  two-explanation `v6.no_icmp_reply` at medium confidence, unchanged.
+
+Like STUN and the canary, the probe SKIPs honestly when no server is configured — corroboration
+is a benefit of enrollment, not a reason to borrow fmr.
+
+## Server: reserved 80/443 verified against the OS, and signed releases (2026-08-02)
+
+**Reserved-address startup probe.** `serve()` now proves 80/443 are actually free on every
+`ECHOLOT_RESERVED_ADDRS` address before starting: a throwaway bind per port
+(`selftest.ReservedWebPortsFree`), fatal on EADDRINUSE with the offending address named — the
+check `CheckReserved` cannot do, because a stray process outside our config (the adb-beacon
+receiver on `0.0.0.0:443` was exactly that) is invisible to configuration checks. Bind errors
+that are not "in use" (typo'd address, address not on this host) warn instead of refusing —
+they are config problems, not pollution.
+
+**Release signing.** Self-update now trusts a signature, not a host. CI signs `SHA256SUMS` with
+an ed25519 key (`relsign` package, `cmd/release-sign`) and the updater refuses any release whose
+`SHA256SUMS.sig` is missing or does not verify against the public key baked into the binary
+(`selfupdate.DefaultPublicKeyB64`; operators with their own pipeline override via
+`ECHOLOT_SELF_UPDATE_PUBKEY`). The private key exists in exactly two places: the Gitea Actions
+secret `RELEASE_SIGNING_KEY`, and the offline original on the dev PC at
+`~/.echolot/release-signing-key`. It is deliberately NOT on fmr and NOT in the repo — a
+compromised release host can withhold updates but no longer inject one. CI hard-fails when the
+secret is missing (an unsigned release would strand every verifying server) and cross-checks the
+signature against the key in the source it just built.
+
+**ACTION REQUIRED before the next `server-v*` tag:** add the Gitea repo secret
+`RELEASE_SIGNING_KEY` (Settings → Actions → Secrets) with the contents of
+`~/.echolot/release-signing-key` from the dev PC. Ordering is safe: the currently deployed
+v0.3.x updater does not verify, so it will happily install the first signed release; every
+release after that is verified. **Done 2026-08-02** — the secret is in place. (The "v0.3.x"
+above should read "the currently deployed release": deployments had moved on to v0.9.x by the
+time signing landed; the point — the deployed updater predates verification and will accept the
+first signed release — is unchanged.)
+
+## Prober fold: traceroute.udp4 and the mDNS inventory go production (2026-08-02, app 0.2.2)
+
+The two highest-value validated capabilities moved from the prober into `core-probe`:
+
+- **`traceroute.udp4`** (`TracerouteProbe`): UDP traceroute reading ICMP time-exceeded off the
+  socket error queue via `Os.recvmsg(MSG_ERRQUEUE)` through the reflection facade — no root, no
+  raw socket, no JNI, ~250 ms for six hops. Emits the schema's `TracerouteEvidence` (rtt in ns).
+  `OsAbi` came with it, including the measured fact that `Os.getsockoptInt` exists on neither
+  known device, so PMTU must always be read from the errqueue (`ee_info`), never
+  `getsockopt(IP_MTU)`. The load-bearing line survived the port: EAGAIN out of the reflected
+  `recvmsg` means "queue empty", not failure.
+- **`local.mdns_inventory`** (`MdnsInventoryProbe`): MulticastLock + NSD discovery, the service
+  inventory that doubles as the VLAN-leakage detector. Both hardware lessons kept: the
+  `_services._dns-sd._udp.` meta-query returns 0 beside live services on both devices (so the
+  concrete types are the measurement and the meta-query result is itself evidence), and the
+  listen window is 10 s because 4 s missed services.
+
+Still to fold, in order: the Shizuku dump *parsers* (the raw `link.ip_monitor` captures already
+hold two divergent vendor formats that could feed `link.ra_source` and `sec.arp_watch`);
+`multinetwork.request_and_bind` (extend ConstraintDetector to *request* transports rather than
+only probing present ones); `peer.ble_advertise` (needs three new permissions and a peer mode to
+exist first).
+
+## Server v0.9.2: trains, real TTL/DSCP/ECN, rate limits (2026-08-02)
+
+The spec-vs-implementation gap audit closed its top items; protocol_version 1.0.0 → 1.0.1
+(additive — below 1.0.0 the minor is the breaking axis, and nothing here breaks an old client):
+
+- **Upstream trains** (§3.2, types 0x03/0x04/0x05): per-train bounded columnar buffer (8192
+  rows, head kept on overflow with `Truncated` set — mirrors the schema's `evidence_truncated`
+  honesty), TRAIN_REPORT split across ≤1200-byte datagrams, grant-free with the §3.4 argument
+  spelled out (a 17-byte report row answers a ≥36-byte HMAC-valid packet). Unknown train id
+  gets a zero-row report: "nothing arrived" is an answer. Also surfaced as `udp.trains` in the
+  observations API.
+- **Real TTL/DSCP/ECN observation** (§3.3): the read loop is `ReadMsgUDPAddrPort` with
+  IP_RECVTTL/IP_RECVTOS/IPV6_RECVHOPLIMIT/IPV6_RECVTCLASS cmsgs on Linux; `0xFF` stays the
+  "not observed" sentinel elsewhere. This unblocks `sec.dscp_ecn_survival` both directions,
+  paired with the new `dscp` parameter on `downtrain` (validated 0–63, refused not clamped,
+  `dscp_applied` in the response).
+- **Rate limiting** (§2.5, was entirely absent): token buckets keyed per credential AND per
+  source IP; 429 + Retry-After on session/action creation (`/v1/profile` stays ungated), silent
+  drop on the data plane — charged after the HMAC gate so a spoofed flood cannot drain a
+  victim's budget, before the replay window so a dropped seq stays usable. UDP ceilings default
+  above the largest legitimate run (a 200 Mbps throughput test), because a rate limit that
+  clips a real measurement produces a confidently wrong number.
+- **`action_id` in every granted packet** (§5/§9): payload bytes [8:16] across all granted
+  types, so overlapping actions are attributable. Verified the deployed Kotlin client parses
+  only ECHO_RESP and MTU_ACK payloads, so the reshuffle strands nobody.
+- **Canary log retention**: the stated 24 h privacy default is now enforced
+  (`ECHOLOT_DNS_LOG_RETENTION_H`), where before the log was time-unbounded.
+- **`POST /admin/enroll-tokens`** now answers the spec's JSON shape under content negotiation;
+  the README's curl works as documented.
+- Spec §2.3 registry gained `downtrain` and `tcp-echo`, which the server had been advertising
+  as strings a conformant client must ignore.
+
+Client-side counterparts still to build: sending 0x03 trains + parsing 0x05 reports
+(`train.udp_updown`), and passing `dscp` on downtrain actions.
+
+## ⚠ Version lineage broken: fmr runs v0.11.2, the repo's tags stop at v0.9.x (2026-08-02)
+
+Discovered while preparing to self-update fmr to the freshly released server-v0.9.2:
+**fmr runs v0.11.2** (binary installed 2026-08-02 08:51), but this repo's remote has tags only
+up to `server-v0.9.1`, master fast-forwarded cleanly from this machine, there is no v0.10/v0.11
+release in Gitea, no source checkout or Go toolchain on fmr, and no deploy script in this repo
+that stamps versions. Conclusion: v0.11.2 was cross-built from a clone whose commits were never
+pushed — presumably another dev machine.
+
+Consequences until resolved:
+- **Do NOT run `--self-update` on fmr.** Gitea's `/releases/latest` is the *newest-created*
+  release, which is now `server-v0.9.2` — semantically older than the deployed binary; the
+  updater compares strings, not SemVer, and would happily "update" v0.11.2 down to it. No
+  automatic risk exists (fmr has no update timer installed, only the cert timer), but a manual
+  run would downgrade.
+- The next real release must be tagged **above v0.11.2** (e.g. `server-v0.11.3` or `v0.12.0`)
+  *after* the missing commits are pushed, so "latest" becomes truly latest again.
+- The unpushed v0.10–v0.11 work needs to be found and pushed from whichever machine built it,
+  or the deployed binary's provenance re-established some other way, before the release channel
+  can be trusted again.

docs/findings-registry.md

@@ -0,0 +1,126 @@
+<!--
+SPDX-FileCopyrightText: 2026 Echolot contributors
+SPDX-License-Identifier: CC-BY-4.0
+-->
+
+# Echolot findings registry
+
+Closes open item 1 of `measurement-schema.md` §9.
+
+A **finding code** is the stable, machine-readable half of a result. The prose around it changes
+freely; the code is what a dashboard groups by, what a diff between two runs keys on, and what
+someone greps a year of archived runs for. That only works if a code means exactly one thing,
+forever.
+
+This document is the contract. It is kept in step with
+`echolot-app/core-measurement/.../FindingRegistry.kt` by a test that fails when either side has a
+code the other does not — a registry that drifts from its documentation is worse than none,
+because it looks authoritative.
+
+## Rules
+
+1. **The prefix determines the category**, and the category determines which verdict light the
+   finding rolls up into (§7.3). A `nat.*` code appearing under *connectivity* is not a naming
+   quibble; it changes which light turns red. Two codes were renamed from `nat.*` to
+   `connectivity.*` for exactly this reason.
+2. **One code per concept.** Two emitters independently produced `connectivity.downstream_loss`
+   and `connectivity.loss_downstream` for the same claim before this registry existed. Anyone
+   aggregating either would have silently seen half their data.
+3. **Codes are declared, not typed.** Emitters reference a `FindingSpec`, so a typo is a compile
+   error and no two call sites can disagree about a finding's category or default severity.
+4. **Severity in the registry is the default.** An emitter may escalate for a specific run; it may
+   not quietly reclassify the finding in general.
+5. **Say what is ruled out**, where that is the useful half. "Loss upstream" is worth far more
+   when it also states that the return path is clean, because that halves where to look next.
+6. **Renaming a code is a breaking change** once runs are archived at scale. Before 1.0 it is
+   cheap; after, it needs an alias and a deprecation window.
+
+## Registry
+
+### connectivity
+
+| code | severity | means | rules out |
+|---|---|---|---|
+| `connectivity.udp_unreachable` | high | No UDP echo replies came back from the server at all. | — |
+| `connectivity.udp_unreachable_upstream` | high | The server received none of the probes, so traffic is dropped on the way out. | The return path: nothing arrived to be replied to. |
+| `connectivity.udp_loss` | medium | A large fraction of round-trip probes were lost, direction unknown. | — |
+| `connectivity.loss_upstream` | medium | Probes were lost on the way to the server. | The return path: replies came back for everything that arrived. |
+| `connectivity.loss_downstream` | medium | Packets were lost on the way back from the server. | The outbound path: the server received what it was answering. |
+| `connectivity.downstream_blocked` | high | Server-initiated packets never arrive, although round trips work. | Basic reachability: the path forwards replies, just not unsolicited traffic. |
+| `connectivity.downstream_reorder` | low | Downstream packets arrive in a different order than they were sent. | — |
+| `connectivity.captive_portal` | medium | A captive portal is intercepting connectivity checks. | — |
+| `connectivity.no_internet` | high | Android's own connectivity checks fail on this network. | — |
+
+### mtu
+
+| code | severity | means | rules out |
+|---|---|---|---|
+| `mtu.reduced_downstream` | low | The downstream path MTU is below the usual 1500 bytes. | — |
+| `mtu.downstream_blackhole` | medium | Datagrams above the path MTU are dropped downstream, fragmented or not. | — |
+| `mtu.fragments_blocked` | medium | IP fragments do not reach this device even when sent in order. | — |
+| `mtu.fragment_reorder_sensitive` | low | Fragments are delivered in order but dropped when reordered or delayed. | Fragmentation itself: in-order fragments arrive fine. |
+
+### nat
+
+| code | severity | means | rules out |
+|---|---|---|---|
+| `nat.udp_rebinding` | medium | A NAT remapped the UDP source port mid-flow. | — |
+| `nat.symmetric` | medium | The NAT assigns a different external port per destination. | — |
+
+### perf
+
+| code | severity | means | rules out |
+|---|---|---|---|
+| `perf.throughput_no_delivery` | high | No throughput traffic arrived, although the server sent it. | — |
+| `perf.throughput_below_offered` | low | Less throughput arrived than the server sent for the whole run. | — |
+
+### dns
+
+| code | severity | means | rules out |
+|---|---|---|---|
+| `dns.answer_rewritten` | high | A resolver returned an answer that differs from the authoritative record. | — |
+| `dns.authoritative_unreachable` | medium | The canary zone's authoritative server could not be reached. | — |
+
+### v6
+
+The prefix is `v6.`, matching the test-type registry (`v6.brokenness`, `v6.happy_eyeballs`, …).
+These were `ipv6.*` while declaring `Category.IPV6`; since the prefix map only knows `v6`, they
+rolled up under *connectivity* instead — the third occurrence of rule 1 being broken.
+
+| code | severity | means | rules out |
+|---|---|---|---|
+| `dns.search_domain_unanswered` | high | The network advertises a DNS search domain that its own server does not answer for. | A fault on this device: the same server answers ordinary names normally. |
+| `dns.system_resolver_broken` | high | The network's DNS server answers, but this device cannot resolve names through it. | A network fault: the server replied to a query sent from this device. |
+| `measurement.vpn_constrained` | info | A VPN was active, so the networks underneath it could not be measured. | Nothing — this run says little about the underlying network either way. |
+| `v6.no_default_route` | medium | The device has a global IPv6 address but no IPv6 default route. | Guesswork: this is read from the routing table, not inferred from silence. |
+| `v6.route_without_address` | medium | The network advertises an IPv6 default route but the device has no global IPv6 address. | A working IPv6 setup: SLAAC did not produce a usable address on this link. |
+| `v6.no_icmp_reply` | low | IPv6 is configured but ICMPv6 echo gets no reply. | Nothing on its own: IPv6 may work fine with ICMP filtered. |
+| `v6.broken` | high | IPv6 is advertised on this network but carries no traffic. | ICMP filtering as the benign explanation: a TCP connection over IPv6 failed too. |
+| `v6.not_offered` | info | This network does not offer IPv6. | — |
+
+`v6.no_icmp_reply` was `v6.broken` until a phone reported it while loading an IPv6-only site over
+TCP perfectly well. The only evidence behind it is ICMPv6 echo, which is widely filtered on
+networks where IPv6 works — so the finding now states what was observed and names both
+explanations instead of choosing one. It is still worth reporting: filtered ICMPv6 breaks Path MTU
+Discovery.
+
+`v6.broken` returned once that corroboration existed: the `v6.brokenness` test attempts a real TCP
+connection over IPv6 to the configured server, and only when *both* transports fail on a network
+that advertises IPv6 is the brokenness claim made — at high severity, because every dual-stack
+destination pays a timeout before falling back to IPv4. When the TCP connect *succeeds*,
+`v6.no_icmp_reply` is emitted at high confidence instead, now able to say plainly that ICMPv6 is
+filtered while IPv6 works. With no server configured there is no corroboration target and the
+two-explanation `v6.no_icmp_reply` stands unchanged.
+
+`v6.not_offered` is **info and must stay info**. Most networks still do not offer IPv6 and that is
+not a fault; reporting it as a warning lights a yellow verdict on a healthy network, which teaches
+people to ignore the light — the one thing a diagnostic must never do.
+
+## Adding a finding
+
+1. Add a `FindingSpec` to `FindingRegistry`, and to its `all` list.
+2. Add the row here, under the section its prefix names.
+3. Emit it with `finding(FindingRegistry.YOUR_CODE, …)`.
+
+The registry test checks 1 and 2 agree, that every prefix maps to the category it claims, and that
+no two entries share a code.

docs/measurement-schema.md

@@ -52,12 +52,26 @@ Export encoding: UTF-8 JSON, gzip for files (`.echolot.json.gz`), share intent u
   },
   "tiers": { "app": true, "shizuku": true, "root": false },
   "profiles_used": ["profile-uuid", ...],
+  "constraints": {
+    "vpn_active": true,
+    "per_network_blocked": true,
+    "unmeasured_networks": ["net-0", "net-1"]
+  },
   "notes": "free-text user annotation"
 }
 ```
 
 `tiers` records what was *available*; each test records what it *used*.
 
+`constraints` records what was *prevented*. A constrained run is neither a failed run nor a normal
+one, and the distinction has to survive into the data: a run taken through a VPN has the same shape
+and the same green verdict as a clean run of a healthy network, so without this a reader — or a
+server aggregating thousands of them — cannot tell that almost nothing was measured. The known case
+is `per_network_blocked`: Android refuses `Network.bindSocket()` on the underlying networks while a
+VPN holds the default route, so every per-network test measures the tunnel or nothing at all, and
+any conclusion about the link underneath is unfounded. Consumers should treat findings from a
+constrained run as scoped to what was actually reachable, and `unmeasured_networks` names the rest.
+
 ## 4. `networks[]` — one entry per Android `Network` in play
 
 A run may exercise several networks simultaneously (Wi-Fi + cellular + USB ethernet). Everything is a snapshot at run start; a `changes[]` list captures mid-run deltas.
@@ -269,7 +283,7 @@ The JSON Schema (machine-readable companion, `measurement.schema.json`, generate
 
 | type | example fields | v2 anonymizer transform |
 |---|---|---|
-| `ip4`, `ip6` | addresses, routes, hops, DNS answers | prefix-preserving pseudonymization, consistent per document; well-known/reserved ranges kept verbatim |
+| `ip4`, `ip6` | addresses, routes, hops, DNS answers | prefix-preserving pseudonymization, consistent per document; well-known/reserved ranges kept verbatim. **Exception: ULA (`fc00::/7`) has its whole prefix pseudonymized as a unit.** It resembles RFC1918 but is not analogous: a ULA global ID is 40 random bits, unique to one network by construction (RFC 4193), so the prefix *is* the identifier, whereas `192.168.0.0/16` is shared by millions of networks and identifies none. Pseudonymizing it as a unit keeps "these hosts are on one subnet" while dropping "this is that subnet". |
 | `mac`, `bssid` | wifi, arp_watch | OUI kept, NIC part pseudonymized |
 | `fqdn` | DNS names, reverse lookups | per-label pseudonyms, public-suffix kept |
 | `ssid` | wifi | pseudonym |
@@ -279,7 +293,8 @@ Free-text fields (`notes`, `error.detail`, dump excerpts from Shizuku parsers) c
 
 ## 9. Open items
 
-1. Findings registry document — start alongside the first implemented tests.
+1. ~~Findings registry document~~ — done: `findings-registry.md`, kept in step with
+   `FindingRegistry.kt` by a test that fails when the two disagree.
 2. Whether Shizuku raw-dump excerpts (dumpsys/ip output) are embedded in `evidence` verbatim (auditable, but large and hard to anonymize) or parsed-only with an optional "attach raw dumps" toggle. Proposal: toggle, default on for local archive, default off for export.
 3. Peer-mode documents: each device produces its own run; the coordinator embeds the peer's findings summary and cross-references by `run.id`. Full merge format deferred.
 4. Size guardrails: soft cap 20 MB uncompressed per run; trains beyond that downsample evidence (keep aggregates + first/last N + all anomalies) and record `"evidence_truncated": true`.

docs/probe-protocol.md

@@ -84,7 +84,12 @@ The app re-fetches the profile at the start of every run (falling back to the ca
 
 ### 2.3 Capabilities (v1 registry)
 
-`udp-probe`, `stun-basic`, `stun-5780`, `canary-dns`, `recursive-dns`, `connect-back`, `delayed-echo`, `big-send`, `frag-send`, `tls-echo`, `http-echo`, `throughput`, `ntp`. A server omits what it can't offer (e.g. `stun-5780` without a second IP degrades to `stun-basic`). Clients must skip, and record as `unsupported`, any test whose capability is absent. Unknown capability strings are ignored.
+`udp-probe`, `stun-basic`, `stun-5780`, `canary-dns`, `recursive-dns`, `connect-back`, `delayed-echo`, `big-send`, `frag-send`, `tls-echo`, `http-echo`, `throughput`, `ntp`, plus:
+
+- `downtrain` — server-sent downstream trains via the §5 `downtrain` action. Upstream trains need no capability of their own: they are plain client-sent data-plane packets and ride `udp-probe`.
+- `tcp-echo` — the plain-TCP echo endpoint (§4); `tls-echo` is its ALPN variant on the same port.
+
+A server omits what it can't offer (e.g. `stun-5780` without a second IP degrades to `stun-basic`). Clients must skip, and record as `unsupported`, any test whose capability is absent. Unknown capability strings are ignored.
 
 ### 2.4 Sessions
 
@@ -111,6 +116,31 @@ DELETE /v1/sessions/{id}
 
 Per-credential and per-source-IP token buckets on: session creation, actions, UDP packets, bytes. `429` on control plane; silent drop on data plane (probes must tolerate loss anyway). All reflected/generated traffic goes **only** to the session's observed source address (or, for connect-back, the source address of the session-creating request). Data-plane responses to unauthenticated packets are never larger than the request (§3.4).
 
+### 2.2 `GET /v1/discover` — where the control plane lives
+
+Unauthenticated, and says almost nothing: the control-plane URL and the server's display name.
+
+```json
+{ "control_url": "https://probe.example.net", "name": "example" }
+```
+
+It exists so an enrollment link can carry the name a person recognises while the app still connects
+to the name that selects the pinned certificate. When a server shares port 443 between its admin UI
+and its control plane, those must be different hostnames — one port and one name is one certificate,
+and the two need different ones (a browser-trusted certificate, and a long-lived self-signed one the
+client pins). Without discovery, the difference leaks into every enrollment link an operator hands
+out.
+
+**It hands out an address, never a pin.** The pin travels in the link itself. Serving it here would
+reduce pinning to whatever the certificate authorities are worth, and pinning exists precisely to
+survive one the operator does not control — a root injected by corporate device management, for
+instance, which is unremarkable on the networks this tool is pointed at. Because the pin is
+pre-shared, an intercepted discovery response can only send a device to the wrong host, where the
+pin will not match: an outage, not a compromise.
+
+Clients treat it as optional. A server that does not answer, or a link that already names the
+control endpoint, works unchanged — enrollment must not begin failing because a lookup did.
+
 ## 3. UDP probe protocol
 
 ### 3.1 Packet header (fixed 32 bytes, network byte order)

echolot-app/app/build.gradle.kts

@@ -15,7 +15,7 @@ plugins {
 //
 // major*1_000_000 + minor*10_000 + patch*10 leaves room for 9 patch-level rebuilds (the trailing
 // digit) without disturbing the mapping, and stays inside the 2_100_000_000 ceiling until major 2100.
-val appVersionName = "0.2.0"
+val appVersionName = "0.2.3"
 
 fun versionCodeOf(semver: String): Int {
     val (major, minor, patch) = semver.substringBefore('-').split(".").map(String::toInt)
@@ -34,8 +34,16 @@ android {
         versionName = appVersionName
         // Automation: `adb shell am start -n app.echo_lot.app/.MainActivity --ez autorun true`
         // runs a measurement immediately and POSTs the report here (dev collection endpoint).
-        buildConfigField("String", "REPORT_UPLOAD_URL", "\"http://89.185.109.150:443/report\"")
-        buildConfigField("String", "REPORT_UPLOAD_SECRET", "\"D4OmG5gGJsElqVVbtYIZbR\"")
+        // Empty: the collection endpoint this pointed at was the adb-beacon receiver, which held
+        // 0.0.0.0:443 in cleartext. That service is gone and echolot-server owns 443 with TLS, so
+        // posting plaintext there now fails as "client sent an HTTP request to an HTTPS server" —
+        // an alarming error for a debugging convenience that is no longer needed, since autorun
+        // reports are read straight off the device with `run-as cat`.
+        //
+        // Deliberately not repointed at /v1/runs. That is the consent-gated upload, and a
+        // debugging shortcut must not be able to satisfy it by accident.
+        buildConfigField("String", "REPORT_UPLOAD_URL", "\"\"")
+        buildConfigField("String", "REPORT_UPLOAD_SECRET", "\"\"")
         // The bare SemVer, without the debug build's "-dev" suffix stripped away by the server's
         // parser anyway — sent to servers so they can apply their compatibility window.
         buildConfigField("String", "APP_SEMVER", "\"$appVersionName\"")

echolot-app/app/src/main/AndroidManifest.xml

@@ -22,7 +22,8 @@
 
         <activity
             android:name=".MainActivity"
-            android:exported="true">
+            android:exported="true"
+            android:launchMode="singleTask">
             <intent-filter>
                 <action android:name="android.intent.action.MAIN" />
                 <category android:name="android.intent.category.LAUNCHER" />
@@ -39,6 +40,17 @@
                 <category android:name="android.intent.category.BROWSABLE" />
                 <data android:scheme="echolot" android:host="enroll" />
             </intent-filter>
+            <!--
+              Sign-in redirect. The browser hands the authorization code back through this, which
+              is exactly why the flow uses PKCE: any app may register this scheme, so the code
+              alone must not be enough to complete a sign-in.
+            -->
+            <intent-filter android:autoVerify="false">
+                <action android:name="android.intent.action.VIEW" />
+                <category android:name="android.intent.category.DEFAULT" />
+                <category android:name="android.intent.category.BROWSABLE" />
+                <data android:scheme="echolot" android:host="auth" />
+            </intent-filter>
         </activity>
 
         <provider

echolot-app/app/src/main/kotlin/app/echo_lot/app/Account.kt

@@ -0,0 +1,125 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package app.echo_lot.app
+
+import app.echo_lot.protocol.AuthInfo
+import app.echo_lot.protocol.ControlClient
+import app.echo_lot.protocol.OidcLogin
+import kotlinx.serialization.json.Json
+import kotlinx.serialization.json.jsonObject
+import kotlinx.serialization.json.jsonPrimitive
+
+/**
+ * Signing in to the configured server's identity provider.
+ *
+ * The awkward part of a browser-based sign-in on Android is that the app is not running while it
+ * happens. Handing control to a browser puts this process in the background, where it may be
+ * killed at any moment; the callback then arrives at a fresh process with none of the state the
+ * exchange needs. So the PKCE verifier and state are written to storage before the browser opens,
+ * not held in memory — an in-memory value works on a developer's device and fails on a phone under
+ * memory pressure, which is the worst way for this to break.
+ *
+ * Nothing from the identity provider is kept afterwards. The ID token proves who is signing in,
+ * once; the device credential authenticates everything from then on.
+ */
+class Account(private val settings: Settings) {
+
+    private val json = Json { ignoreUnknownKeys = true }
+
+    sealed interface SignInStart {
+        /** Open this in a browser. */
+        data class Browser(val url: String) : SignInStart
+        data class Unavailable(val reason: String) : SignInStart
+    }
+
+    /** Fetches the server's auth configuration and builds the authorization URL. */
+    fun begin(): SignInStart {
+        if (!settings.serverConfigured) {
+            return SignInStart.Unavailable(
+                "Enrol with a server first — sign-in belongs to the server's identity provider."
+            )
+        }
+        val auth = runCatching { client().profile(settings.serverCredential).auth }.getOrNull()
+            ?: return SignInStart.Unavailable("Could not reach the server to ask how to sign in.")
+
+        auth.discoveryError?.let {
+            // The distinction matters: "the operator configured an IdP that is not answering" is
+            // their problem to fix, and is not the same as "this server has no accounts".
+            return SignInStart.Unavailable("The server's identity provider is not responding: $it")
+        }
+        if (!auth.enabled) {
+            return SignInStart.Unavailable("This server does not offer accounts.")
+        }
+        return try {
+            val pending = OidcLogin.begin(auth)
+            // Written before the browser opens, because after that this process may not survive.
+            settings.pendingVerifier = pending.verifier
+            settings.pendingState = pending.state
+            SignInStart.Browser(pending.authorizationUrl)
+        } catch (t: Throwable) {
+            SignInStart.Unavailable(t.message ?: "Could not start sign-in.")
+        }
+    }
+
+    /**
+     * Completes sign-in from the `echolot://auth` redirect.
+     *
+     * Blocking; callers run it off the main thread.
+     */
+    fun complete(callbackUri: String): String {
+        val verifier = settings.pendingVerifier
+        val state = settings.pendingState
+        // Cleared first, whatever happens next: these are single-use, and leaving them behind
+        // would let a later callback be completed against a flow nobody started.
+        settings.clearPendingAuth()
+
+        if (verifier.isBlank() || state.isBlank()) {
+            return "That sign-in did not start on this device."
+        }
+        return try {
+            val auth = client().profile(settings.serverCredential).auth
+            val idToken = OidcLogin.complete(
+                auth, OidcLogin.Pending("", verifier, state), callbackUri,
+            )
+            val reply = client().linkAccount(settings.serverCredential, idToken)
+            val o = json.parseToJsonElement(reply).jsonObject
+            val name = o["display_name"]?.jsonPrimitive?.content ?: "signed in"
+            settings.accountName = name
+            settings.accountId = o["account_id"]?.jsonPrimitive?.content ?: ""
+            val admin = o["admin"]?.jsonPrimitive?.content == "true"
+            "Signed in as $name" + if (admin) " (administrator)" else ""
+        } catch (e: OidcLogin.LoginFailed) {
+            e.message ?: "Sign-in failed."
+        } catch (t: Throwable) {
+            "Sign-in failed: ${t.message ?: t.javaClass.simpleName}"
+        }
+    }
+
+    /** Signs out. The device stays enrolled — signing out should not cost an enrolment. */
+    fun signOut(): String = try {
+        client().unlinkAccount(settings.serverCredential)
+        settings.accountName = ""
+        settings.accountId = ""
+        "Signed out. This device is still enrolled."
+    } catch (t: Throwable) {
+        "Could not sign out: ${t.message ?: t.javaClass.simpleName}"
+    }
+
+    /** Asks the server who it thinks is signed in, so the UI is not trusting stale local state. */
+    fun refresh(): String? = runCatching {
+        val o = json.parseToJsonElement(client().accountStatus(settings.serverCredential)).jsonObject
+        val signedIn = o["signed_in"]?.jsonPrimitive?.content == "true"
+        settings.accountName = if (signedIn) {
+            o["display_name"]?.jsonPrimitive?.content ?: ""
+        } else {
+            ""
+        }
+        settings.accountName.takeIf { it.isNotBlank() }
+    }.getOrNull()
+
+    private fun client() = ControlClient(
+        settings.serverUrl, setOf(settings.serverPin), BuildConfig.APP_SEMVER,
+        fallbackAddrs = settings.serverAddrList(),
+    )
+}

echolot-app/app/src/main/kotlin/app/echo_lot/app/HistoryScreen.kt

@@ -4,6 +4,7 @@
 package app.echo_lot.app
 
 import androidx.compose.foundation.layout.Arrangement
+import androidx.compose.foundation.layout.safeDrawingPadding
 import androidx.compose.foundation.layout.Column
 import androidx.compose.foundation.layout.Row
 import androidx.compose.foundation.layout.fillMaxWidth
@@ -39,7 +40,7 @@ fun HistoryScreen(
     onDelete: (String) -> Unit,
     onBack: () -> Unit,
 ) {
-    Column(Modifier.fillMaxWidth().padding(16.dp), verticalArrangement = Arrangement.spacedBy(10.dp)) {
+    Column(Modifier.fillMaxWidth().safeDrawingPadding().padding(16.dp), verticalArrangement = Arrangement.spacedBy(10.dp)) {
         Row(verticalAlignment = Alignment.CenterVertically) {
             TextButton(onClick = onBack) { Text("‹ Back") }
             Text("History", style = MaterialTheme.typography.titleLarge)
@@ -71,12 +72,20 @@ fun HistoryScreen(
                             )
                         }
                         Text(
-                            "${r.findingCount} finding(s) · ${r.sizeBytes / 1024} kB · ${r.anonymization}",
+                            "${r.findingCount} finding(s) · ${r.sizeBytes / 1024} kB · " +
+                                "kept complete on this device",
                             style = MaterialTheme.typography.bodySmall,
                         )
+                        // The upload line names the level the upload was made at, not the
+                        // archive's. They describe different documents, and showing the archive's
+                        // level here claimed more had left the device than actually did.
                         Text(
-                            if (r.uploaded) "uploaded to ${r.uploadedTo ?: "a server"}"
-                            else "on this device only",
+                            if (r.uploaded) {
+                                "uploaded to ${r.uploadedTo ?: "a server"}" +
+                                    (r.uploadedAs?.let { " as $it" } ?: "")
+                            } else {
+                                "on this device only"
+                            },
                             style = MaterialTheme.typography.bodySmall,
                             color = if (r.uploaded) Color(0xFF7FD17F) else Color(0xFFBBBBBB),
                         )

echolot-app/app/src/main/kotlin/app/echo_lot/app/MainActivity.kt

@@ -43,9 +43,27 @@ class MainActivity : ComponentActivity() {
     private val permissionLauncher =
         registerForActivityResult(ActivityResultContracts.RequestMultiplePermissions()) { /* proceed regardless */ }
 
+    /**
+     * The intent currently being acted on, so a deep link that arrives while the app is running
+     * is seen by the screen the user is already looking at.
+     *
+     * The activity is singleTask for the same reason. As a standard activity it stacked a second
+     * instance per link, each with its own ViewModel: the enrolment then happened in a throwaway
+     * copy, and pressing back returned to the original screen showing none of it. Silent, and
+     * indistinguishable from the link simply not working.
+     */
+    private val liveIntent = mutableStateOf<android.content.Intent?>(null)
+
+    override fun onNewIntent(intent: android.content.Intent) {
+        super.onNewIntent(intent)
+        setIntent(intent)
+        liveIntent.value = intent
+    }
+
     override fun onCreate(savedInstanceState: Bundle?) {
         super.onCreate(savedInstanceState)
         requestRuntimePermissions()
+        liveIntent.value = intent
         setContent {
             MaterialTheme(colorScheme = darkColorScheme()) {
                 Surface(color = MaterialTheme.colorScheme.background) {
@@ -63,13 +81,88 @@ class MainActivity : ComponentActivity() {
                     // An echolot://enroll link (QR scan, or a link the operator sent) opens the
                     // app straight into settings with the enrollment already done, so the user
                     // sees the result rather than a form they still have to fill in.
-                    val enrollUri = intent?.takeIf { it.action == Intent.ACTION_VIEW }?.dataString
+                    // Both deep links land here. They are told apart by host, so a sign-in
+                    // redirect is never mistaken for an enrolment link — one spends a token, the
+                    // other completes an authorization, and confusing them would fail obscurely.
+                    val incoming = liveIntent.value?.takeIf { it.action == Intent.ACTION_VIEW }?.dataString
+                    val authUri = incoming?.takeIf { it.startsWith("echolot://auth") }
+                    val enrollUri = incoming?.takeIf { it.startsWith("echolot://enroll") }
                     androidx.compose.runtime.LaunchedEffect(enrollUri) {
                         if (enrollUri != null) {
                             vm.enroll(enrollUri)
                             screen = Screen.SETTINGS
                         }
                     }
+                    // Replacing an existing enrollment is asked about, never assumed. Following a
+                    // link from a web page is one tap, and the old credential does not survive it.
+                    vm.state.pendingEnroll?.let { pending ->
+                        androidx.compose.material3.AlertDialog(
+                            onDismissRequest = { vm.cancelEnroll() },
+                            title = {
+                                androidx.compose.material3.Text(
+                                    if (pending.sameServer) "Enroll again with this server?"
+                                    else "Replace this device's server?"
+                                )
+                            },
+                            text = {
+                                androidx.compose.material3.Text(
+                                    // Naming the same URL twice reads as a mistake and buries the
+                                    // one consequence that actually applies: the device is issued a
+                                    // fresh credential and shows up as a second entry.
+                                    if (pending.sameServer) {
+                                        "This device is already enrolled with " +
+                                            "${pending.currentServer}.\n\n" +
+                                            "Enrolling again replaces its credential. The old one " +
+                                            "stops working immediately, and the device appears on " +
+                                            "the server as a new entry alongside the current one — " +
+                                            "which you may want to revoke afterwards.\n\n" +
+                                            "Runs already uploaded, and runs stored on this phone, " +
+                                            "are not affected."
+                                    } else {
+                                        "This device is already enrolled with " +
+                                            "${pending.currentServer}.\n\n" +
+                                            "Enrolling with ${pending.newServer} replaces that. Runs " +
+                                            "already uploaded stay where they are, but this device " +
+                                            "stops reporting to the old server and appears on the new " +
+                                            "one as a new device.\n\n" +
+                                            "Runs stored on this phone are not affected."
+                                    }
+                                )
+                            },
+                            confirmButton = {
+                                androidx.compose.material3.TextButton(onClick = { vm.confirmEnroll() }) {
+                                    androidx.compose.material3.Text(
+                                        if (pending.sameServer) "Enroll again" else "Enroll here"
+                                    )
+                                }
+                            },
+                            dismissButton = {
+                                androidx.compose.material3.TextButton(onClick = { vm.cancelEnroll() }) {
+                                    androidx.compose.material3.Text("Keep current server")
+                                }
+                            },
+                        )
+                    }
+                    androidx.compose.runtime.LaunchedEffect(authUri) {
+                        if (authUri != null) {
+                            vm.completeSignIn(authUri)
+                            screen = Screen.SETTINGS
+                        }
+                    }
+                    // Shizuku can be started, stopped or authorised in its own app, where nothing
+                    // calls back into this process. Asking again each time this screen comes
+                    // forward is what makes the banner right after the user has been away to fix
+                    // it — which is exactly the moment they look at it.
+                    val lifecycleOwner = androidx.compose.ui.platform.LocalLifecycleOwner.current
+                    androidx.compose.runtime.DisposableEffect(lifecycleOwner) {
+                        val obs = androidx.lifecycle.LifecycleEventObserver { _, event ->
+                            if (event == androidx.lifecycle.Lifecycle.Event.ON_RESUME) {
+                                vm.refreshShizuku()
+                            }
+                        }
+                        lifecycleOwner.lifecycle.addObserver(obs)
+                        onDispose { lifecycleOwner.lifecycle.removeObserver(obs) }
+                    }
                     androidx.compose.runtime.LaunchedEffect(autorun) {
                         if (autorun) vm.run(devUpload = true)
                     }
@@ -85,23 +178,43 @@ class MainActivity : ComponentActivity() {
                             finish()
                         }
                     }
+                    // Without this, the system Back gesture leaves the activity from Settings or
+                    // History instead of returning to the run screen — the screen is a plain state
+                    // variable, so nothing connects it to the back stack. Registered only when
+                    // there is somewhere to go back to, so Back still exits from the run screen.
+                    androidx.activity.compose.BackHandler(enabled = screen != Screen.RUN) {
+                        screen = Screen.RUN
+                    }
+
                     when (screen) {
                       Screen.SETTINGS -> SettingsScreen(
                         settings = vm.settings,
-                        archivedRuns = vm.state.history.size,
+                        archivedRuns = vm.archivedRunCount(),
                         archivedBytes = vm.archivedBytes(),
                         onApplyRetention = vm::applyRetention,
                         onDeleteAll = vm::deleteAllRuns,
                         onPreviewUpload = {
-                            // Preview the newest run, since that is the one the user just made
-                            // and the one they are deciding about.
-                            vm.state.history.firstOrNull()?.let { r ->
-                                lifecycleScope.launch { preview = vm.uploadPreview(r.id) }
-                            }
+                            // Straight from the archive: the newest run is the one the user just
+                            // made and the one they are deciding about. Always shows something,
+                            // even when there is nothing to preview yet.
+                            lifecycleScope.launch { preview = vm.previewNewestRun() }
                         },
                         onCheckServer = vm::checkServer,
+                        accountName = vm.accountName,
+                        onSignIn = {
+                            vm.beginSignIn { url ->
+                                // A plain VIEW intent rather than a Custom Tab: the browser is
+                                // where the user's existing IdP session already lives, and
+                                // androidx.browser would be a dependency for a rounded corner.
+                                runCatching {
+                                    startActivity(Intent(Intent.ACTION_VIEW, android.net.Uri.parse(url)))
+                                }
+                            }
+                        },
+                        onSignOut = vm::signOut,
                         onEnroll = vm::enroll,
                         serverStatus = vm.state.archiveStatus,
+                        enrollStatus = vm.state.enrollStatus,
                         onBack = { screen = Screen.RUN },
                       )
                       Screen.HISTORY -> HistoryScreen(
@@ -300,6 +413,41 @@ private fun EcholotScreen(
 
 @Composable
 private fun Results(doc: MeasurementDocument) {
+    // A constrained run is answered before the lights are: the verdict below is INCONCLUSIVE by
+    // §7.3, and without this banner "inconclusive" reads as the app failing rather than the OS
+    // (correctly) refusing to let anything past the VPN be measured.
+    val constraints = doc.run.constraints
+    if (constraints.constrained) {
+        val blocked = constraints.unmeasuredNetworks
+            .mapNotNull { id -> doc.networks.firstOrNull { it.id == id } }
+            .joinToString(", ") { it.iface?.takeIf { s -> s.isNotBlank() } ?: it.transport.name.lowercase() }
+            .ifBlank { "the networks beneath it" }
+        // Same three-way split as the finding: saying "VPN" when the user just disconnected
+        // theirs (the wall lingers during teardown) reads as the app being wrong, not the OS.
+        val (headline, body) = when {
+            constraints.vpnActive && constraints.perNetworkBlocked ->
+                "Measured through a VPN" to
+                    ("Android does not let apps send on the networks beneath an active VPN, so " +
+                        "$blocked could not be measured — these results describe the tunnel. " +
+                        "Disconnect the VPN and run again to measure the networks themselves.")
+            constraints.perNetworkBlocked ->
+                "Some networks could not be measured" to
+                    ("Android refused sends on $blocked — the restriction a VPN leaves in place " +
+                        "while it tears down. These networks went unmeasured; wait a few " +
+                        "seconds and run again.")
+            else ->
+                "A VPN holds the default route" to
+                    ("Default-route results describe the tunnel; per-network measurements " +
+                        "reached the underlying networks.")
+        }
+        Card(colors = CardDefaults.cardColors(containerColor = Color(0xFF3A2E12))) {
+            Column(Modifier.fillMaxWidth().padding(12.dp)) {
+                Text(headline, color = Color(0xFFFFD08A), fontWeight = FontWeight.SemiBold)
+                Text(body, fontSize = 12.sp, color = Color(0xFFFFD08A))
+            }
+        }
+    }
+
     val summary = doc.summary
     if (summary != null) {
         Card(colors = CardDefaults.cardColors(containerColor = verdictColor(summary.overall))) {

echolot-app/app/src/main/kotlin/app/echo_lot/app/RunStore.kt

@@ -80,8 +80,16 @@ class RunStore(context: Context, private val settings: Settings) {
 
     private fun client() = ControlClient(
         settings.serverUrl, setOf(settings.serverPin), BuildConfig.APP_SEMVER,
+        fallbackAddrs = settings.serverAddrList(),
     )
 
+    /** Remembers where the server lives, so a later run can reach it without DNS. */
+    private fun rememberAddrs(p: app.echo_lot.protocol.Profile) {
+        val addrs = p.targets.flatMap { listOfNotNull(it.ip4, it.ip6) }
+            .filter { it.isNotBlank() }
+        if (addrs.isNotEmpty()) settings.serverAddrs = addrs.joinToString(",")
+    }
+
     /**
      * Checks the configured server without uploading anything: reachable, pinned, compatible, and
      * willing to accept runs. Lets the user find out in settings rather than from a failed run.
@@ -90,6 +98,10 @@ class RunStore(context: Context, private val settings: Settings) {
         if (!settings.serverConfigured) return "Fill in the server URL, pin and credential first."
         return try {
             val profile = client().profile(settings.serverCredential)
+            // Learned here so the next run's canary probe knows what to ask for.
+            profile.canaryZone.takeIf { it.isNotBlank() }?.let { settings.canaryZone = it }
+            settings.serverFacts = describeFacts(profile)
+            rememberAddrs(profile)
             val compat = Compat.check(profile, BuildConfig.APP_SEMVER)
             val head = "${profile.name} · server ${profile.serverVersion} · " +
                 "protocol ${profile.compat.protocolVersion.ifBlank { "unstated" }}"
@@ -115,6 +127,36 @@ class RunStore(context: Context, private val settings: Settings) {
      * no credential — fails later, somewhere else, with an error that points at the wrong thing.
      * Blocking; callers run it off the main thread.
      */
+
+    /**
+     * Renders what the server says about itself, for display.
+     *
+     * Only what a person measuring against it would want to check: which addresses the tests will
+     * actually use, on which ports, and what the server admits it can do. Addresses first, because
+     * "which address did this result come from" is the question a report leaves open.
+     */
+    private fun describeFacts(p: app.echo_lot.protocol.Profile): String {
+        val lines = ArrayList<String>()
+        // "label|value" per line, laid out as real columns by the UI rather than padded with
+        // spaces here. Space padding only lines up in a monospaced font, which makes the layout
+        // depend on a typeface choice made somewhere else entirely.
+        fun row(label: String, value: String) = lines.add("$label|$value")
+
+        row("server", "${p.name} · ${p.serverVersion}")
+        for (t in p.targets) {
+            t.ip4?.let { row("IPv4", it) }
+            t.ip6?.let { row("IPv6", it) }
+            // Marked rather than listed apart: it is the same server, and what matters is being
+            // able to tell which address a NAT-behaviour result came from.
+            t.ip4Alt?.let { row("IPv4 alt", it) }
+            t.ip6Alt?.let { row("IPv6 alt", it) }
+            row("ports", "udp ${t.udpPort} · tcp ${t.tcpPort} · stun ${t.stunPort}")
+        }
+        if (p.canaryZone.isNotBlank()) row("dns zone", p.canaryZone)
+        if (p.capabilities.isNotEmpty()) row("measures", p.capabilities.joinToString(", "))
+        return lines.joinToString(System.lineSeparator())
+    }
+
     fun enroll(link: String, deviceName: String?): String {
         val parsed = app.echo_lot.protocol.EnrollmentLink.parse(link)
             ?: return "That does not look like an Echolot enrollment link. It should start with " +
@@ -122,7 +164,11 @@ class RunStore(context: Context, private val settings: Settings) {
         return try {
             val enrolled = parsed.redeem(deviceName, BuildConfig.APP_SEMVER)
             val compat = Compat.check(enrolled.profile, BuildConfig.APP_SEMVER)
+            settings.serverFacts = describeFacts(enrolled.profile)
+            rememberAddrs(enrolled.profile)
+            enrolled.profile.canaryZone.takeIf { it.isNotBlank() }?.let { settings.canaryZone = it }
             settings.serverUrl = enrolled.controlUrl
+            settings.serverPublicUrl = enrolled.publicUrl
             settings.serverPin = enrolled.pin
             settings.serverCredential = enrolled.credential
             val head = "Enrolled with ${enrolled.profile.name} " +
@@ -150,6 +196,9 @@ class RunStore(context: Context, private val settings: Settings) {
         return try {
             val client = client()
             val profile = client.profile(settings.serverCredential)
+            profile.canaryZone.takeIf { it.isNotBlank() }?.let { settings.canaryZone = it }
+            settings.serverFacts = describeFacts(profile)
+            rememberAddrs(profile)
 
             // Compatibility before policy: an incompatible server may well advertise an upload
             // policy it would never actually apply to us.
@@ -164,8 +213,10 @@ class RunStore(context: Context, private val settings: Settings) {
             )
             val body = redactedForUpload(docJson, level)
             val reply = client.uploadRun(settings.serverCredential, body)
-            archive.markUploaded(runId, profile.name)
-            UploadOutcome.Sent(profile.name, "as $level, ${body.toByteArray().size} bytes: ${reply.take(120)}")
+            archive.markUploaded(runId, profile.name, level.wire)
+            // Deliberately not echoing `reply`: it is the server's index entry as raw JSON, and
+            // it ended up rendered verbatim in the UI. Size and level are what a person wants.
+            UploadOutcome.Sent(profile.name, "as $level, ${body.toByteArray().size} bytes")
         } catch (e: VersionRefused) {
             UploadOutcome.Incompatible(e.message ?: "the server refused this app's version")
         } catch (e: UploadRefused) {

echolot-app/app/src/main/kotlin/app/echo_lot/app/RunViewModel.kt

@@ -42,6 +42,17 @@ data class UiState(
     val archiveStatus: String? = null,
     /** History, newest first. Refreshed after every run and whenever the history screen opens. */
     val history: List<app.echo_lot.archive.ArchivedRun> = emptyList(),
+    /**
+     * Result of the last enrollment attempt, shown beside the Enroll button.
+     *
+     * Separate from [archiveStatus]: they are two different actions with two different results,
+     * and sharing one line put the answer to "did enrolling work" at the far end of the card,
+     * below three text fields — or nowhere at all on a fresh install, since that line only
+     * renders once a run exists.
+     */
+    val enrollStatus: String? = null,
+    /** An enrollment link waiting on confirmation, because this device is already enrolled. */
+    val pendingEnroll: PendingEnroll? = null,
     /** Shell-tier readiness, shown before a run; null message = say nothing (Shizuku not installed). */
     val shizukuNotice: String? = null,
     val shizukuReady: Boolean = false,
@@ -49,6 +60,18 @@ data class UiState(
     val shizukuState: ShizukuAvailability.State = ShizukuAvailability.State.NOT_INSTALLED,
 )
 
+/**
+ * An enrollment link that would replace an existing one, held until the user says so.
+ *
+ * Enrolling is not additive: the new credential replaces the old, and on the previous server this
+ * device simply stops reporting. Following a link is one tap from a web page, which is not enough
+ * deliberation to discard a working enrollment by accident.
+ */
+data class PendingEnroll(val link: String, val currentServer: String, val newServer: String) {
+    /** Re-enrolling with the server already configured, rather than moving to a different one. */
+    val sameServer: Boolean get() = currentServer.trimEnd('/') == newServer.trimEnd('/')
+}
+
 /**
  * Drives one measurement run: device-tier probes (link snapshot, per-network ICMP) always run;
  * results assemble into a MeasurementDocument with a §7.3 summary. Lives in a ViewModel so a run
@@ -80,6 +103,23 @@ class RunViewModel(app: Application) : AndroidViewModel(app) {
         }
     }
 
+    /**
+     * Re-reads the shell tier's state, for when it changed somewhere this process cannot see.
+     *
+     * Permission can be granted inside Shizuku's own app, and Shizuku can be started or stopped
+     * there too; none of that calls back here. Asking again on resume is the only way to be right
+     * after the user has been somewhere else to fix it.
+     */
+    fun refreshShizuku() {
+        val st = ShizukuAvailability.current(getApplication())
+        state = state.copy(
+            shizukuNotice = ShizukuAvailability.describe(st),
+            shizukuReady = st == ShizukuAvailability.State.READY,
+            shizukuHint = ShizukuAvailability.actionHint(st),
+            shizukuState = st,
+        )
+    }
+
     override fun onCleared() {
         stopShizukuObserver()
         super.onCleared()
@@ -90,6 +130,7 @@ class RunViewModel(app: Application) : AndroidViewModel(app) {
     private var runStartWall: String = ""
     private var runNetworks: List<app.echo_lot.measurement.Network> = emptyList()
     private var runShizukuOk = false
+    private var runConstraints = Constraints()
 
     /** Two-clock ids: UUIDs + monotonic ns relative to a per-run origin. */
     private class RunIds : ProbeIds {
@@ -108,6 +149,7 @@ class RunViewModel(app: Application) : AndroidViewModel(app) {
     fun run(devUpload: Boolean = false) {
         if (state.running) return
         collected.clear()
+        runConstraints = Constraints()
         state = state.copy(running = true, currentStep = "starting", document = null,
             uploadStatus = null, archiveStatus = null)
         runJob = viewModelScope.launch {
@@ -125,7 +167,13 @@ class RunViewModel(app: Application) : AndroidViewModel(app) {
             if (devUpload) {
                 state = state.copy(currentStep = "uploading report")
                 val r = withContext(Dispatchers.IO) { ReportUploader.upload(doc) }
-                status = if (r.ok) "uploaded ✓ ${r.detail}" else "upload failed: ${r.detail}"
+                status = when {
+                    r.ok -> "uploaded ✓ ${r.detail}"
+                    // Not a failure worth alarming about: the dev collection endpoint is simply
+                    // not configured, and the run is on the device either way.
+                    r.detail.startsWith("no upload URL") -> "run complete — read it with adb"
+                    else -> "upload failed: ${r.detail}"
+                }
             }
             if (archived != null && settings.autoUpload) {
                 state = state.copy(currentStep = "uploading to server")
@@ -194,13 +242,109 @@ class RunViewModel(app: Application) : AndroidViewModel(app) {
         store.read(id)?.let { store.redactedForUpload(it) }
     }
 
+    /**
+     * Preview of the most recent run, read from the archive rather than from [UiState.history].
+     *
+     * The history list is only populated once the History screen has been opened, so a preview
+     * driven from it did nothing at all on a freshly-opened Settings screen — a button that
+     * silently does nothing is worse than one that says why.
+     */
+    suspend fun previewNewestRun(): String = withContext(Dispatchers.IO) {
+        val newest = store.list().firstOrNull()
+            ?: return@withContext "No archived runs yet. Run a measurement first, then this will " +
+                "show exactly what an upload would send."
+        store.read(newest.id)?.let { store.redactedForUpload(it) }
+            ?: "That run could not be read back from the archive."
+    }
+
     fun archivedBytes(): Long = store.totalBytes()
 
+    /** Counted from the archive itself, not from [UiState.history], which is empty until the
+     *  history screen has been opened - the two disagreeing read as data loss. */
+    fun archivedRunCount(): Int = store.list().size
+
+    private val account = Account(settings)
+
+    /** Name of whoever is signed in on this device, for the settings screen. */
+    var accountName by mutableStateOf(settings.accountName)
+        private set
+
+    /** Starts sign-in; the caller opens the returned URL in a browser. */
+    fun beginSignIn(open: (String) -> Unit) {
+        viewModelScope.launch {
+            state = state.copy(archiveStatus = "contacting the server …")
+            when (val r = withContext(Dispatchers.IO) { account.begin() }) {
+                is Account.SignInStart.Browser -> {
+                    state = state.copy(archiveStatus = "continue in your browser …")
+                    open(r.url)
+                }
+                is Account.SignInStart.Unavailable ->
+                    state = state.copy(archiveStatus = r.reason)
+            }
+        }
+    }
+
+    /** Completes sign-in from the echolot://auth redirect. */
+    fun completeSignIn(callbackUri: String) {
+        viewModelScope.launch {
+            val msg = withContext(Dispatchers.IO) { account.complete(callbackUri) }
+            accountName = settings.accountName
+            state = state.copy(archiveStatus = msg)
+        }
+    }
+
+    fun signOut() {
+        viewModelScope.launch {
+            val msg = withContext(Dispatchers.IO) { account.signOut() }
+            accountName = settings.accountName
+            state = state.copy(archiveStatus = msg)
+        }
+    }
+
     /** Redeems an enrollment link, from a paste or from an echolot:// deep link. */
     fun enroll(link: String, deviceName: String? = android.os.Build.MODEL) {
+        // Already enrolled? Ask first. The old credential is gone the moment this succeeds, and a
+        // link followed from a web page is one tap — far too little deliberation for that.
+        if (settings.serverConfigured) {
+            val target = app.echo_lot.protocol.EnrollmentLink.parse(link)?.controlUrl ?: link
+            state = state.copy(
+                pendingEnroll = PendingEnroll(
+                    link = link,
+                    // Compared against the link's URL, which names the server publicly — so this
+                    // has to be the public name too. Using the endpoint made re-enrolling with the
+                    // same server look like a move to a different one, because the endpoint and
+                    // the public name are deliberately different strings.
+                    currentServer = settings.serverPublicUrl,
+                    newServer = target,
+                )
+            )
+            return
+        }
+        doEnroll(link, deviceName)
+    }
+
+    /** The user confirmed replacing an existing enrollment. */
+    fun confirmEnroll(deviceName: String? = android.os.Build.MODEL) {
+        val pending = state.pendingEnroll ?: return
+        state = state.copy(pendingEnroll = null)
+        doEnroll(pending.link, deviceName)
+    }
+
+    fun cancelEnroll() {
+        state = state.copy(
+            pendingEnroll = null,
+            enrollStatus = "Kept the existing enrollment; nothing changed.",
+        )
+    }
+
+    private fun doEnroll(link: String, deviceName: String?) {
         viewModelScope.launch {
-            state = state.copy(archiveStatus = "enrolling …")
-            state = state.copy(archiveStatus = withContext(Dispatchers.IO) { store.enroll(link, deviceName) })
+            state = state.copy(enrollStatus = "Enrolling …")
+            val result = withContext(Dispatchers.IO) { store.enroll(link, deviceName) }
+            // A new server means a new canary zone; the old one would describe somebody else's
+            // deployment. Cleared rather than kept, and relearned from the next profile fetch.
+            settings.canaryZone = ""
+            state = state.copy(enrollStatus = result)
         }
     }
 
@@ -249,23 +393,45 @@ class RunViewModel(app: Application) : AndroidViewModel(app) {
         val entries = NetworkInventory.snapshot(ctx)
         val networks = entries.map { it.model }.also { runNetworks = it }
 
+        // What will this run be prevented from measuring? Decided up front, from one throwaway
+        // bind per network, so the document can say so instead of leaving it to be inferred from
+        // per-test `attempted: false` breadcrumbs (measurement-schema.md §3 `constraints`).
+        runConstraints = app.echo_lot.probe.ConstraintDetector.detect(ctx, entries)
+
         val probes: List<Probe> = listOf(
             LinkSnapshotProbe(entries),
             RouterIdentityProbe(entries),
             IcmpProbe(entries, v6 = false),
             IcmpProbe(entries, v6 = true),
+            // Folded from the prober after hardware validation: errqueue traceroute (no root,
+            // no JNI) and the mDNS service inventory / VLAN-leakage detector.
+            app.echo_lot.probe.TracerouteProbe(),
+            app.echo_lot.probe.MdnsInventoryProbe(),
             CaptivePortalProbe(entries),
             // Canary zone served by the Echolot probe server (probe-protocol §6.1). Hardcoded to
             // the reference deployment until profiles/enrollment land in the UI.
-            DnsCanaryProbe(canaryZone = "c.echo-lot.app", sessionPrefix = "adhoc"),
-            StunProbe(serverHost = "fmr-1.echo-lot.app"),
+            // Both target whatever server this device is enrolled with, not the deployment the
+            // app happened to be developed against. With no server configured they get blank
+            // strings and report themselves skipped, which is the honest outcome — the
+            // alternative measures someone else's infrastructure and calls it your network.
+            // Before the canary: "can this device resolve at all" has to be answered before
+            // "are the answers being tampered with" means anything.
+            app.echo_lot.probe.DnsResolverProbe(entries),
+            DnsCanaryProbe(canaryZone = settings.canaryZone, sessionPrefix = "adhoc"),
+            StunProbe(serverHost = settings.serverHost()),
+            // Corroboration for icmp.ping6's silence: a real TCP connection over IPv6. Only its
+            // failure, on a network that advertises IPv6, justifies calling IPv6 broken.
+            app.echo_lot.probe.V6ConnectProbe(entries, serverHost = settings.serverHost()),
         )
 
         // Plan the run first: the Shizuku battery is counted alongside the app-tier probes so
-        // the bar reflects the whole run. Estimates are per-probe (see Probe.estimatedMs).
-        val shizukuEstimateMs = 8_000L
+        // the bar reflects the whole run. Estimates prefer what THIS device measured on recent
+        // runs (Settings EMA); Probe.estimatedMs is only the cold-start seed — a fixed table
+        // cannot know whether ICMPv6 answers in milliseconds here or waits out its timeout.
+        fun estimateOf(p: Probe) = settings.learnedDurationMs(p.type) ?: p.estimatedMs
+        val shizukuEstimateMs = settings.learnedDurationMs(SHIZUKU_DURATION_KEY) ?: 8_000L
         val totalSteps = probes.size + 1
-        var remainingMs = probes.sumOf { it.estimatedMs } + shizukuEstimateMs
+        var remainingMs = probes.sumOf { estimateOf(it) } + shizukuEstimateMs
         state = state.copy(stepsDone = 0, stepsTotal = totalSteps,
             etaSeconds = ((remainingMs + 999) / 1000).toInt())
 
@@ -285,22 +451,33 @@ class RunViewModel(app: Application) : AndroidViewModel(app) {
                 }
             )
             tests.add(result); collected.add(result)
-            remainingMs -= p.estimatedMs
+            settings.recordDurationMs(p.type, (result.endedMonoNs - result.startedMonoNs) / 1_000_000)
+            remainingMs -= estimateOf(p)
         }
 
-        // Shizuku shell tier — self-degrades to UNSUPPORTED when Shizuku isn't running.
+        // Shizuku shell tier — self-degrades to UNSUPPORTED when Shizuku isn't running. One
+        // battery, three tests: the raw captures plus the parsed ra_source/arp_watch views.
         step("link.ip_monitor (shizuku)", done = probes.size, total = totalSteps, etaMs = shizukuEstimateMs)
-        val shizukuTest = try {
+        val shizukuT0 = System.nanoTime()
+        val shizukuTests = try {
             ShizukuProbe().run(ctx, ids::uuid, ids::monoNs)
         } catch (t: Throwable) {
-            Test(
+            listOf(Test(
                 id = ids.uuid(), type = TestType.LINK_IP_MONITOR, tier = Tier.SHIZUKU,
                 startedMonoNs = ids.monoNs(), endedMonoNs = ids.monoNs(),
                 status = TestStatus.FAILED, error = TestError("uncaught", t.message ?: t.javaClass.simpleName),
-            )
+            ))
+        }
+        // One key for the whole step: the three tests come out of one shell battery, and the
+        // bar plans them as one step.
+        settings.recordDurationMs(SHIZUKU_DURATION_KEY, (System.nanoTime() - shizukuT0) / 1_000_000)
+        tests.addAll(shizukuTests); collected.addAll(shizukuTests)
+        // "Shizuku tier ran" is the battery's verdict — the derived tests can be PARTIAL on a
+        // perfectly healthy shell tier (e.g. an RA-less v4-only link).
+        runShizukuOk = shizukuTests.any {
+            it.type == TestType.LINK_IP_MONITOR &&
+                (it.status == TestStatus.OK || it.status == TestStatus.PARTIAL)
         }
-        tests.add(shizukuTest); collected.add(shizukuTest)
-        runShizukuOk = shizukuTest.status == TestStatus.OK || shizukuTest.status == TestStatus.PARTIAL
 
         return buildDocument(tests)
     }
@@ -321,39 +498,189 @@ class RunViewModel(app: Application) : AndroidViewModel(app) {
                     androidSdk = Build.VERSION.SDK_INT, androidRelease = Build.VERSION.RELEASE,
                 ),
                 tiers = Tiers(app = true, shizuku = runShizukuOk),
+                constraints = runConstraints,
             ),
             networks = runNetworks,
             tests = tests,
             findings = findings,
-            summary = Verdicts.derive(tests, findings),
+            summary = Verdicts.derive(tests, findings, runConstraints),
         )
     }
 
     /**
-     * Was IPv6 actually provisioned on any network? A global (non-link-local) v6 address or a
-     * v6 default route means the network claims to offer IPv6 — link-local only does not count.
+     * Was IPv6 provisioned on the network a test actually ran over?
+     *
+     * This deliberately asks about one network rather than about the device. Answering "does any
+     * network here have IPv6" produces a real false positive on a phone, and it is not hypothetical:
+     * an IPv4-only wifi with working cellular alongside it reports "IPv6 is configured, but ICMPv6
+     * gets no reply" — configured on cellular, pinged over wifi, and the two never met.
+     *
+     * A global (non-link-local) address or a v6 default route means the network claims to offer
+     * IPv6; link-local only does not count, since every interface has one.
      */
-    private fun ipv6Provisioned(networks: List<app.echo_lot.measurement.Network>): Boolean =
-        networks.any { n ->
+    private fun ipv6Provisioned(
+        networks: List<app.echo_lot.measurement.Network>,
+        networkRef: String?,
+    ): Boolean {
+        // No reference means the test was not per-network; fall back to the device-wide reading
+        // rather than silently reporting nothing.
+        val scope = networks.filter { networkRef == null || it.id == networkRef }
+        return scope.any { n ->
             n.link.addresses.any { a ->
                 a.addr.contains(':') &&
                     !a.addr.startsWith("fe80", ignoreCase = true) &&
                     !a.addr.startsWith("::1")
             } || n.link.routes.any { it.dst == "::/0" }
         }
+    }
+
+    /**
+     * Per-network ICMP outcomes, keyed by network id.
+     *
+     * Reads the structured evidence the probe records rather than its prose detail — a finding
+     * that depended on the wording of a human-readable string would break silently the first time
+     * that wording improved.
+     */
+    /** What one network's ICMP attempt did: whether it ran at all, and whether it was answered. */
+    private data class IcmpOutcome(val attempted: Boolean, val ok: Boolean)
+
+    /**
+     * Per-network ICMP outcomes, keyed by network id.
+     *
+     * Reads the structured evidence the probe records rather than its prose detail — a finding
+     * that depended on the wording of a human-readable string would break silently the first time
+     * that wording improved.
+     */
+    private fun icmpResults(t: Test): Map<String, IcmpOutcome> {
+        val out = HashMap<String, IcmpOutcome>()
+        val ev = t.evidence ?: return out
+        for ((_, v) in ev) {
+            val o = v as? kotlinx.serialization.json.JsonObject ?: continue
+            val ref = (o["network_ref"] as? kotlinx.serialization.json.JsonPrimitive)?.content ?: continue
+            fun flag(k: String) = (o[k] as? kotlinx.serialization.json.JsonPrimitive)?.content == "true"
+            out[ref] = IcmpOutcome(attempted = flag("attempted"), ok = flag("ok"))
+        }
+        return out
+    }
+
+    /** Human-facing name for the network a test ran over; falls back to something readable. */
+    private fun ifaceOf(networks: List<app.echo_lot.measurement.Network>, ref: String?): String =
+        networks.firstOrNull { it.id == ref }?.iface?.takeIf { it.isNotBlank() } ?: "this network"
 
     /** Minimal first-pass findings from device-tier evidence; the registry grows with the suite. */
     private fun deriveFindings(tests: List<Test>, networks: List<app.echo_lot.measurement.Network>): List<Finding> {
         val out = ArrayList<Finding>()
         val ids = RunIds()
+        val linkEvidence = tests.filter { it.type == TestType.LINK_SNAPSHOT }.map { EvidenceRef(it.id) }
+
+        // Said as a finding, not only as run.constraints: the constraints block is for machines
+        // aggregating thousands of runs, this is for the person reading this one. Both must exist —
+        // a constrained run with a quiet findings list still reads as "nothing wrong here".
+        if (runConstraints.constrained) {
+            val blocked = runConstraints.unmeasuredNetworks
+                .joinToString(", ") { id -> ifaceOf(networks, id) }
+                .ifBlank { "the underlying networks" }
+            // Three distinct situations share this finding code, and naming the wrong one costs
+            // trust: claiming "a VPN is active" right after the user disconnected theirs is how
+            // this text was first proven wrong on hardware.
+            val (title, description) = when {
+                runConstraints.vpnActive && runConstraints.perNetworkBlocked ->
+                    "A VPN is active — $blocked could not be measured" to
+                        ("Android refuses to let apps send on the networks beneath an active " +
+                            "VPN (that is how it prevents traffic leaking around the tunnel), " +
+                            "so every per-network test here measured the tunnel or nothing. " +
+                            "Nothing in this run says anything about $blocked. To measure them, " +
+                            "disconnect the VPN and run again.")
+                runConstraints.perNetworkBlocked ->
+                    "The OS refused sends on $blocked" to
+                        ("Android denied this app permission to send on $blocked (EPERM on " +
+                            "bind). That is the restriction a VPN imposes on the networks " +
+                            "beneath it — a tunnel disconnected moments ago can still leave it " +
+                            "in place while it tears down. Nothing in this run says anything " +
+                            "about $blocked; wait a few seconds and run again.")
+                else ->
+                    "A VPN holds the default route" to
+                        ("Everything using the default route in this run describes the tunnel, " +
+                            "not the network it rides on. Per-network measurements were " +
+                            "permitted and did measure the underlying networks.")
+            }
+            out.add(
+                Finding(
+                    id = ids.uuid(),
+                    code = FindingRegistry.MEASUREMENT_VPN_CONSTRAINED.code,
+                    category = FindingRegistry.MEASUREMENT_VPN_CONSTRAINED.category,
+                    severity = FindingRegistry.MEASUREMENT_VPN_CONSTRAINED.severity,
+                    confidence = Confidence.HIGH,
+                    title = title,
+                    description = description,
+                    evidenceRefs = linkEvidence,
+                )
+            )
+        }
+        val shapes = V6Analysis.classify(networks)
+        // Named per interface: on a phone several networks are up at once, and "IPv6 is broken" is
+        // useless when wifi is the broken one and cellular is fine.
+        for (sh in shapes.filter { it.addressWithoutRoute }) {
+            val where = if (sh.iface.isBlank()) "This device" else sh.iface
+            out.add(
+                Finding(
+                    id = ids.uuid(),
+                    code = FindingRegistry.V6_NO_DEFAULT_ROUTE.code,
+                    category = FindingRegistry.V6_NO_DEFAULT_ROUTE.category,
+                    severity = if (sh.tunnel) Severity.INFO else Severity.MEDIUM,
+                    confidence = Confidence.HIGH,
+                    title = if (sh.tunnel) {
+                        "IPv6 reaches only the destinations a tunnel routes ($where)"
+                    } else {
+                        "IPv6 address with no default route ($where)"
+                    },
+                    description = "$where has a global IPv6 address but no IPv6 default route, so " +
+                        "IPv6 reaches only destinations covered by a specific route. " +
+                        if (sh.tunnel) {
+                            "A tunnel interface holds those routes, so this looks deliberate. " +
+                                "Worth knowing rather than fixing: applications holding a global " +
+                                "address will still try IPv6 first and stall for anything outside " +
+                                "the tunnel's routes."
+                        } else {
+                            "Nothing is routing the rest, so the network handed out an address it " +
+                                "does not carry traffic for — applications will try IPv6 first " +
+                                "and wait for it to fail."
+                        },
+                    evidenceRefs = linkEvidence,
+                )
+            )
+        }
+        for (sh in shapes.filter { it.routeWithoutAddress }) {
+            val where = if (sh.iface.isBlank()) "This network" else sh.iface
+            out.add(
+                Finding(
+                    id = ids.uuid(),
+                    code = FindingRegistry.V6_ROUTE_WITHOUT_ADDRESS.code,
+                    category = FindingRegistry.V6_ROUTE_WITHOUT_ADDRESS.category,
+                    severity = Severity.MEDIUM,
+                    confidence = Confidence.HIGH,
+                    title = "IPv6 router advertised, but no address was configured ($where)",
+                    description = "$where has an IPv6 default route but no global IPv6 address. " +
+                        "The router is advertising itself as an IPv6 gateway while SLAAC produced " +
+                        "no usable address — a missing prefix option, a prefix without the " +
+                        "autonomous flag, or DHCPv6-only addressing that did not complete. Hosts " +
+                        "believe IPv6 is available and pay a connection timeout on every " +
+                        "dual-stack destination before falling back to IPv4, which is felt as " +
+                        "general slowness with no packet loss to explain it.",
+                    evidenceRefs = linkEvidence,
+                )
+            )
+        }
+
         for (t in tests) {
             if (t.type == TestType.NET_CAPTIVE_PORTAL) {
                 val ev = t.evidence?.toString() ?: ""
                 when {
                     ev.contains("\"captive_portal\"") -> out.add(
                         Finding(
-                            id = ids.uuid(), code = "connectivity.captive_portal", category = Category.CONNECTIVITY,
-                            severity = Severity.MEDIUM, confidence = Confidence.HIGH,
+                            id = ids.uuid(), code = FindingRegistry.CAPTIVE_PORTAL.code,
+                            category = FindingRegistry.CAPTIVE_PORTAL.category,
+                            severity = FindingRegistry.CAPTIVE_PORTAL.severity, confidence = Confidence.HIGH,
                             title = "Captive portal intercepting connections",
                             description = "The generate_204 check returned a redirect or a page instead of HTTP 204 — a captive portal (login/splash page) is intercepting traffic on this network.",
                             evidenceRefs = listOf(EvidenceRef(t.id)),
@@ -361,8 +688,9 @@ class RunViewModel(app: Application) : AndroidViewModel(app) {
                     )
                     t.status == TestStatus.FAILED -> out.add(
                         Finding(
-                            id = ids.uuid(), code = "connectivity.no_internet", category = Category.CONNECTIVITY,
-                            severity = Severity.HIGH, confidence = Confidence.HIGH,
+                            id = ids.uuid(), code = FindingRegistry.NO_INTERNET.code,
+                            category = FindingRegistry.NO_INTERNET.category,
+                            severity = FindingRegistry.NO_INTERNET.severity, confidence = Confidence.HIGH,
                             title = "No working internet on any network",
                             description = "Android's own generate_204 connectivity checks failed on every active network (no HTTP 204) — this device has no validated internet path.",
                             evidenceRefs = listOf(EvidenceRef(t.id)),
@@ -375,8 +703,9 @@ class RunViewModel(app: Application) : AndroidViewModel(app) {
                 if (ev.contains("MISMATCH")) {
                     out.add(
                         Finding(
-                            id = ids.uuid(), code = "dns.answer_rewritten", category = Category.DNS,
-                            severity = Severity.HIGH, confidence = Confidence.HIGH,
+                            id = ids.uuid(), code = FindingRegistry.DNS_ANSWER_REWRITTEN.code,
+                            category = FindingRegistry.DNS_ANSWER_REWRITTEN.category,
+                            severity = FindingRegistry.DNS_ANSWER_REWRITTEN.severity, confidence = Confidence.HIGH,
                             title = "DNS answers are being rewritten",
                             description = "A canary reference record returned different RDATA than the spec-defined ground truth — something on the path is rewriting DNS answers (interception, filtering, or a middlebox).",
                             evidenceRefs = listOf(EvidenceRef(t.id)),
@@ -385,8 +714,9 @@ class RunViewModel(app: Application) : AndroidViewModel(app) {
                 } else if (ev.contains("\"reached_authoritative\":false")) {
                     out.add(
                         Finding(
-                            id = ids.uuid(), code = "dns.authoritative_unreachable", category = Category.DNS,
-                            severity = Severity.MEDIUM, confidence = Confidence.MEDIUM,
+                            id = ids.uuid(), code = FindingRegistry.DNS_AUTHORITATIVE_UNREACHABLE.code,
+                            category = FindingRegistry.DNS_AUTHORITATIVE_UNREACHABLE.category,
+                            severity = FindingRegistry.DNS_AUTHORITATIVE_UNREACHABLE.severity, confidence = Confidence.MEDIUM,
                             title = "Canary queries don't reach the authoritative server",
                             description = "A per-run nonce name (which cannot be cached) was not answered by the canary server — the resolver is intercepting or failing to reach it.",
                             evidenceRefs = listOf(EvidenceRef(t.id)),
@@ -399,8 +729,9 @@ class RunViewModel(app: Application) : AndroidViewModel(app) {
                 if (ev.contains("address/port-dependent (symmetric NAT")) {
                     out.add(
                         Finding(
-                            id = ids.uuid(), code = "nat.symmetric", category = Category.NAT,
-                            severity = Severity.MEDIUM, confidence = Confidence.HIGH,
+                            id = ids.uuid(), code = FindingRegistry.NAT_SYMMETRIC.code,
+                            category = FindingRegistry.NAT_SYMMETRIC.category,
+                            severity = FindingRegistry.NAT_SYMMETRIC.severity, confidence = Confidence.HIGH,
                             title = "Symmetric NAT — peer-to-peer connections need a relay",
                             description = "The NAT assigns a different external port per destination (address/port-dependent mapping). Direct peer-to-peer connections (calls, games, file transfer) will usually fail and fall back to relays.",
                             evidenceRefs = listOf(EvidenceRef(t.id)),
@@ -408,29 +739,185 @@ class RunViewModel(app: Application) : AndroidViewModel(app) {
                     )
                 }
             }
-            if (t.type == TestType.ICMP_PING6 && t.status == TestStatus.FAILED) {
-                // A network with no IPv6 at all is NORMAL — most networks are still IPv4-only,
-                // and that is not a defect. What IS a defect is IPv6 that the network claims to
-                // provide (a global address or a default route from RA/DHCPv6) but that does not
-                // work: that causes Happy-Eyeballs delays, timeouts and hangs. So the severity
-                // depends on whether v6 was provisioned at all.
-                if (ipv6Provisioned(networks)) {
+            if (t.type == TestType.DNS_RESOLVER && t.status == TestStatus.OK) {
+                // One finding per network: on a phone the wifi resolver can be wedged while
+                // cellular is fine, and "DNS is broken" would be wrong about half the device.
+                val ev = t.evidence
+                if (ev != null) {
+                    for ((_, v) in ev) {
+                        val o = v as? kotlinx.serialization.json.JsonObject ?: continue
+                        fun str(k: String) =
+                            (o[k] as? kotlinx.serialization.json.JsonPrimitive)?.content
+                        val verdict = str("verdict")
+                        val ref0 = str("network_ref")
+                        val iface0 = networks.firstOrNull { it.id == ref0 }?.iface
+                            ?.takeIf { it.isNotBlank() } ?: "this network"
+                        if (verdict == "search domain swallows queries") {
+                            // Severity follows the harm, not the shape: the same misconfiguration
+                            // is fatal on a resolver that tries the search form and invisible on
+                            // one that does not, and saying "high" for a network that currently
+                            // resolves fine would be crying wolf.
+                            val breaking = str("system_resolves") != "true"
                             out.add(
                                 Finding(
-                            id = ids.uuid(), code = "ipv6.broken", category = Category.IPV6,
-                            severity = Severity.MEDIUM, confidence = Confidence.HIGH,
-                            title = "IPv6 is configured but not working",
-                            description = "This network advertises IPv6 (a global address and/or a default route), but ICMPv6 got no reply on any network. Half-configured IPv6 is worse than none: connections try IPv6 first and stall before falling back.",
+                                    id = ids.uuid(),
+                                    code = FindingRegistry.DNS_SEARCH_DOMAIN_UNANSWERED.code,
+                                    category = FindingRegistry.DNS_SEARCH_DOMAIN_UNANSWERED.category,
+                                    severity = if (breaking) Severity.HIGH else Severity.MEDIUM,
+                                    confidence = Confidence.HIGH,
+                                    title = "The network's search domain swallows DNS queries ($iface0)",
+                                    description = "This network hands out " +
+                                        "${str("search_domains") ?: "a search domain"} as a DNS " +
+                                        "search domain, but its server never answers queries under " +
+                                        "it — not even to say the name does not exist. Resolvers " +
+                                        "append that domain to lookups, so they wait for a reply " +
+                                        "that never comes. " +
+                                        (if (breaking) {
+                                            "That is why names are not resolving on this device."
+                                        } else {
+                                            "Name resolution still works here, because this " +
+                                                "resolver tries the plain name first — another " +
+                                                "device on the same network may fail outright."
+                                        }) +
+                                        " Fix it on the router: either stop advertising the search " +
+                                        "domain, or make the server answer for it, including " +
+                                        "NXDOMAIN for names it does not have. Note that .local is " +
+                                        "reserved for mDNS (RFC 6762) and is widely dropped by " +
+                                        "design; home.arpa (RFC 8375) is the name reserved for this.",
                                     evidenceRefs = listOf(EvidenceRef(t.id)),
                                 )
                             )
-                } else {
+                            continue
+                        }
+                        if (verdict != "server answers, device resolver does not") continue
+                        val ref = str("network_ref")
+                        val where = networks.firstOrNull { it.id == ref }?.iface
+                            ?.takeIf { it.isNotBlank() } ?: "this network"
                         out.add(
                             Finding(
-                            id = ids.uuid(), code = "ipv6.not_offered", category = Category.IPV6,
-                            severity = Severity.INFO, confidence = Confidence.HIGH,
+                                id = ids.uuid(),
+                                code = FindingRegistry.DNS_SYSTEM_RESOLVER_BROKEN.code,
+                                category = FindingRegistry.DNS_SYSTEM_RESOLVER_BROKEN.category,
+                                severity = FindingRegistry.DNS_SYSTEM_RESOLVER_BROKEN.severity,
+                                confidence = Confidence.HIGH,
+                                title = "This device cannot resolve names, but the DNS server is fine ($where)",
+                                description = "A DNS query sent straight from this device was " +
+                                    "answered by ${str("servers") ?: "the configured server"} with " +
+                                    "a valid result, yet asking Android to resolve the same name " +
+                                    "fails. Whatever is wrong sits between this device's resolver " +
+                                    "and a server that demonstrably works. " +
+                                    "Turning wifi off and on, or rejoining the network, clears the " +
+                                    "common case. If it survives a restart it is not a stuck " +
+                                    "resolver: look for something on this device that filters DNS " +
+                                    "— an ad blocker, a private-DNS or VPN app — or a per-device " +
+                                    "rule on the router aimed at this client.",
+                                evidenceRefs = listOf(EvidenceRef(t.id)),
+                            )
+                        )
+                    }
+                }
+            }
+            if (t.type == TestType.ICMP_PING6) {
+                // A network with no IPv6 at all is NORMAL — most networks are still IPv4-only, and
+                // that is not a defect. What IS a defect is IPv6 the network claims to provide (a
+                // global address or a default route from RA/DHCPv6) that does not work: that causes
+                // Happy-Eyeballs delays, timeouts and hangs.
+                //
+                // Judged per network, from the per-network evidence rather than the aggregate
+                // status. The aggregate can only say "some network answered", and on a phone with
+                // wifi and cellular up at once that is how "IPv6 is configured but gets no reply"
+                // ends up describing a network where IPv6 was never configured in the first place.
+                val results = icmpResults(t)
+                // The corroborating witness: did a real TCP connection over IPv6 work on this
+                // network? Same evidence shape as the ICMP probe, so the same parser reads it.
+                val v6ConnTest = tests.firstOrNull { it.type == TestType.V6_BROKENNESS }
+                val v6Conn = v6ConnTest?.let { icmpResults(it) } ?: emptyMap()
+                var anyV6Network = false
+                for (n in networks) {
+                    val provisioned = ipv6Provisioned(networks, n.id)
+                    if (provisioned) anyV6Network = true
+                    val r = results[n.id] ?: continue
+                    // Silence is only evidence if something was actually sent. A bind that failed
+                    // with EPERM says the app could not use the interface, which is a fact about
+                    // this app's permissions and says nothing whatsoever about the network.
+                    if (!provisioned || !r.attempted || r.ok) continue
+                    val where = n.iface?.takeIf { it.isNotBlank() } ?: "this network"
+                    val conn = v6Conn[n.id]
+                    val evidence = listOfNotNull(
+                        EvidenceRef(t.id), v6ConnTest?.let { EvidenceRef(it.id) },
+                    )
+                    when {
+                        // TCP over IPv6 worked: the silence is filtering, and can be said so.
+                        conn?.ok == true -> out.add(
+                            Finding(
+                                id = ids.uuid(), code = FindingRegistry.V6_NO_ICMP_REPLY.code,
+                                category = FindingRegistry.V6_NO_ICMP_REPLY.category,
+                                severity = FindingRegistry.V6_NO_ICMP_REPLY.severity,
+                                confidence = Confidence.HIGH,
+                                title = "ICMPv6 is filtered here — IPv6 itself works ($where)",
+                                description = "$where answered a real TCP connection over IPv6, " +
+                                    "so IPv6 works — but ICMPv6 echo got no reply, so something " +
+                                    "on this network filters ICMPv6. That is a fault in its own " +
+                                    "right even though connections succeed: Path MTU Discovery " +
+                                    "depends on ICMPv6, so large packets can vanish rather than " +
+                                    "being reported as too big.",
+                                evidenceRefs = evidence,
+                            )
+                        )
+                        // Both transports failed on a network that advertises IPv6: broken, and
+                        // now with the evidence the original v6.broken never had.
+                        conn != null && conn.attempted -> out.add(
+                            Finding(
+                                id = ids.uuid(), code = FindingRegistry.V6_BROKEN.code,
+                                category = FindingRegistry.V6_BROKEN.category,
+                                severity = FindingRegistry.V6_BROKEN.severity,
+                                confidence = Confidence.HIGH,
+                                title = "IPv6 is advertised but does not work ($where)",
+                                description = "$where advertises IPv6 (a global address and/or a " +
+                                    "default route), but neither ICMPv6 echo nor a TCP connection " +
+                                    "over IPv6 got through — two independent transports, both " +
+                                    "silent. Applications will try IPv6 first and wait out a " +
+                                    "timeout on every dual-stack destination before falling back " +
+                                    "to IPv4, felt as everything being slow with no loss to " +
+                                    "explain it. The network is announcing a service it does not " +
+                                    "deliver; the fix belongs on the router or upstream.",
+                                evidenceRefs = evidence,
+                            )
+                        )
+                        // No corroboration available (no server configured, or the connect never
+                        // got as far as sending): the honest two-explanation reading stands.
+                        else -> out.add(
+                            Finding(
+                                id = ids.uuid(), code = FindingRegistry.V6_NO_ICMP_REPLY.code,
+                                category = FindingRegistry.V6_NO_ICMP_REPLY.category,
+                                severity = FindingRegistry.V6_NO_ICMP_REPLY.severity,
+                                confidence = Confidence.MEDIUM,
+                                title = "IPv6 is configured, but ICMPv6 gets no reply ($where)",
+                                description = "$where advertises IPv6 (a global address and/or a " +
+                                    "default route), but ICMPv6 echo got no reply over it. That has " +
+                                    "two explanations which look identical from here: IPv6 is broken, " +
+                                    "or ICMPv6 is filtered while IPv6 itself works. Filtering is " +
+                                    "common and is a fault in its own right — it breaks Path MTU " +
+                                    "Discovery, so large packets vanish rather than being reported as " +
+                                    "too big.",
+                                evidenceRefs = listOf(EvidenceRef(t.id)),
+                            )
+                        )
+                    }
+                }
+                if (!anyV6Network) {
+                    // Said once for the device, not once per interface: "this network is IPv4-only"
+                    // repeated per interface reads as several problems instead of one observation.
+                    out.add(
+                        Finding(
+                            id = ids.uuid(), code = FindingRegistry.V6_NOT_OFFERED.code,
+                            category = FindingRegistry.V6_NOT_OFFERED.category,
+                            severity = FindingRegistry.V6_NOT_OFFERED.severity,
+                            confidence = Confidence.HIGH,
                             title = "IPv4-only network (no IPv6 offered)",
-                            description = "No IPv6 address or default route was provisioned, so IPv6 tests could not run. This is normal — many networks are still IPv4-only and it is not a fault.",
+                            description = "No IPv6 address or default route was provisioned on " +
+                                "any active network, so IPv6 tests could not run. This is normal " +
+                                "— many networks are still IPv4-only and it is not a fault.",
                             evidenceRefs = listOf(EvidenceRef(t.id)),
                         )
                     )
@@ -446,4 +933,9 @@ class RunViewModel(app: Application) : AndroidViewModel(app) {
             etaSeconds = if (etaMs >= 0) ((etaMs + 999) / 1000).toInt() else state.etaSeconds,
         )
     }
+
+    private companion object {
+        /** Duration-learning key for the Shizuku step, which is three tests but one battery. */
+        const val SHIZUKU_DURATION_KEY = "shizuku.battery"
+    }
 }

echolot-app/app/src/main/kotlin/app/echo_lot/app/Settings.kt

@@ -50,6 +50,35 @@ class Settings(context: Context) {
         maxTotalBytes = maxTotalMb.toLong() * 1024 * 1024,
     )
 
+    // ---- run-duration learning ---------------------------------------------------------
+
+    /**
+     * Learned duration of one test type on THIS device, or null before the first run.
+     *
+     * The static Probe.estimatedMs values are only cold-start seeds: real durations depend on
+     * the phone and the network it stands in (ICMPv6 answers in milliseconds where IPv6 works
+     * and waits out full timeouts where it does not), so a fixed table is wrong for almost
+     * everyone almost always. What was measured last time is the only estimate that tracks
+     * reality.
+     */
+    fun learnedDurationMs(type: String): Long? =
+        prefs.getLong("$DURATION_PREFIX$type", -1L).takeIf { it > 0 }
+
+    /**
+     * Feeds one measured duration into the estimate — EMA, 70 % old / 30 % new. Heavy enough
+     * on history that a single odd run (a captive portal stalling DNS) does not whipsaw the
+     * bar, light enough that a real change (enrolling with a server un-skips three probes)
+     * converges within a few runs. Recorded whatever the test's status: a probe that skips in
+     * 2 ms will keep skipping in 2 ms until circumstances change, and then the EMA follows.
+     */
+    fun recordDurationMs(type: String, ms: Long) {
+        if (ms < 0) return
+        val key = "$DURATION_PREFIX$type"
+        val old = prefs.getLong(key, -1L)
+        val next = if (old <= 0) ms else (old * 7 + ms * 3) / 10
+        prefs.edit().putLong(key, next).apply()
+    }
+
     // ---- upload ----------------------------------------------------------------------
 
     /**
@@ -97,6 +126,41 @@ class Settings(context: Context) {
         get() = prefs.getString(SERVER_PIN, "") ?: ""
         set(v) = prefs.edit().putString(SERVER_PIN, v.trim()).apply()
 
+    /**
+     * The address the operator handed out, for showing to a person.
+     *
+     * Separate from [serverUrl], which is the endpoint actually dialled. They differ when the
+     * server publishes one public name and points devices at another to select its pinned
+     * certificate — a detail worth keeping out of the user's face but not out of the settings.
+     */
+    var serverPublicUrl: String
+        get() = (prefs.getString(SERVER_PUBLIC_URL, "") ?: "").ifBlank { serverUrl }
+        set(v) = prefs.edit().putString(SERVER_PUBLIC_URL, v.trim()).apply()
+
+    /**
+     * What the server said about itself, last time it was asked: addresses, ports, capabilities.
+     *
+     * Cached as a rendered block rather than as fields, because it is shown and never acted on —
+     * these are facts to read, not settings to apply, and storing them as settings would invite
+     * exactly the confusion of an editable box that changes nothing.
+     */
+    var serverFacts: String
+        get() = prefs.getString(SERVER_FACTS, "") ?: ""
+        set(v) = prefs.edit().putString(SERVER_FACTS, v).apply()
+
+    /**
+     * The server's own addresses, learned from its profile, for reaching it when DNS will not.
+     *
+     * Only the primaries: the alternate pair exists for NAT behaviour discovery and does not carry
+     * the control plane, so falling back to one would fail for a second, unrelated reason.
+     */
+    var serverAddrs: String
+        get() = prefs.getString(SERVER_ADDRS, "") ?: ""
+        set(v) = prefs.edit().putString(SERVER_ADDRS, v).apply()
+
+    fun serverAddrList(): List<String> =
+        serverAddrs.split(',').map { it.trim() }.filter { it.isNotEmpty() }
+
     var serverCredential: String
         get() = prefs.getString(SERVER_CRED, "") ?: ""
         set(v) = prefs.edit().putString(SERVER_CRED, v.trim()).apply()
@@ -104,6 +168,58 @@ class Settings(context: Context) {
     val serverConfigured: Boolean
         get() = serverUrl.isNotBlank() && serverPin.isNotBlank() && serverCredential.isNotBlank()
 
+    /**
+     * The DNS zone this server is authoritative for, learned from its profile.
+     *
+     * Cached because the canary probe runs at device tier, before anything has talked to the
+     * server, and a probe that had to make a control-plane call first would fail on exactly the
+     * networks worth measuring. Empty means "not known yet", and the probe reports itself as
+     * skipped rather than inventing a zone.
+     */
+    var canaryZone: String
+        get() = prefs.getString(CANARY_ZONE, "") ?: ""
+        set(v) = prefs.edit().putString(CANARY_ZONE, v.trim()).apply()
+
+    /**
+     * Host part of the configured server URL, for probes that address it directly (STUN).
+     *
+     * Derived rather than stored: a second copy of the server's name is a second thing to keep in
+     * step, and it would go stale the moment someone re-enrolled against a different server.
+     */
+    fun serverHost(): String = runCatching {
+        java.net.URI(serverUrl).host?.takeIf { it.isNotBlank() }
+    }.getOrNull() ?: ""
+
+    // ---- account ---------------------------------------------------------------------
+
+    /**
+     * The PKCE verifier and state for a sign-in that is out at the browser.
+     *
+     * Persisted rather than held in memory because handing control to a browser backgrounds this
+     * process, and Android may kill it before the callback returns. An in-memory value works on a
+     * developer's device and fails on a phone under memory pressure.
+     */
+    var pendingVerifier: String
+        get() = prefs.getString(PENDING_VERIFIER, "") ?: ""
+        set(v) = prefs.edit().putString(PENDING_VERIFIER, v).apply()
+
+    var pendingState: String
+        get() = prefs.getString(PENDING_STATE, "") ?: ""
+        set(v) = prefs.edit().putString(PENDING_STATE, v).apply()
+
+    fun clearPendingAuth() = prefs.edit().remove(PENDING_VERIFIER).remove(PENDING_STATE).apply()
+
+    /** Display name of whoever is signed in on this device; empty when nobody is. */
+    var accountName: String
+        get() = prefs.getString(ACCOUNT_NAME, "") ?: ""
+        set(v) = prefs.edit().putString(ACCOUNT_NAME, v).apply()
+
+    var accountId: String
+        get() = prefs.getString(ACCOUNT_ID, "") ?: ""
+        set(v) = prefs.edit().putString(ACCOUNT_ID, v).apply()
+
+    val signedIn: Boolean get() = accountName.isNotBlank()
+
     private fun hex(s: String) = ByteArray(s.length / 2) {
         ((Character.digit(s[it * 2], 16) shl 4) or Character.digit(s[it * 2 + 1], 16)).toByte()
     }
@@ -120,5 +236,14 @@ class Settings(context: Context) {
         const val SERVER_URL = "server_url"
         const val SERVER_PIN = "server_pin"
         const val SERVER_CRED = "server_credential"
+        const val SERVER_PUBLIC_URL = "server_public_url"
+        const val SERVER_FACTS = "server_facts"
+        const val SERVER_ADDRS = "server_addrs"
+        const val CANARY_ZONE = "server_canary_zone"
+        const val PENDING_VERIFIER = "pending_auth_verifier"
+        const val PENDING_STATE = "pending_auth_state"
+        const val ACCOUNT_NAME = "account_name"
+        const val ACCOUNT_ID = "account_id"
+        const val DURATION_PREFIX = "duration_ms."
     }
 }

echolot-app/app/src/main/kotlin/app/echo_lot/app/SettingsScreen.kt

@@ -4,19 +4,24 @@
 package app.echo_lot.app
 
 import androidx.compose.foundation.layout.Arrangement
+import androidx.compose.foundation.layout.safeDrawingPadding
 import androidx.compose.foundation.layout.Column
 import androidx.compose.foundation.layout.Row
 import androidx.compose.foundation.layout.Spacer
 import androidx.compose.foundation.layout.fillMaxWidth
 import androidx.compose.foundation.layout.height
+import androidx.compose.foundation.layout.width
 import androidx.compose.foundation.layout.padding
+import androidx.compose.foundation.shape.RoundedCornerShape
 import androidx.compose.foundation.rememberScrollState
 import androidx.compose.foundation.verticalScroll
 import androidx.compose.material3.Button
 import androidx.compose.material3.Card
 import androidx.compose.material3.FilterChip
+import androidx.compose.material3.LocalContentColor
 import androidx.compose.material3.MaterialTheme
 import androidx.compose.material3.OutlinedTextField
+import androidx.compose.material3.Surface
 import androidx.compose.material3.Switch
 import androidx.compose.material3.Text
 import androidx.compose.material3.TextButton
@@ -29,6 +34,7 @@ import androidx.compose.ui.Alignment
 import androidx.compose.ui.Modifier
 import androidx.compose.ui.text.font.FontFamily
 import androidx.compose.ui.unit.dp
+import androidx.compose.ui.unit.sp
 import app.echo_lot.privacy.PrivacyLevel
 
 /**
@@ -47,8 +53,12 @@ fun SettingsScreen(
     onDeleteAll: () -> Unit,
     onPreviewUpload: () -> Unit,
     onCheckServer: () -> Unit,
+    accountName: String,
+    onSignIn: () -> Unit,
+    onSignOut: () -> Unit,
     onEnroll: (String) -> Unit,
     serverStatus: String?,
+    enrollStatus: String?,
     onBack: () -> Unit,
 ) {
     // SharedPreferences is not observable, so mirror each value into Compose state and write
@@ -61,12 +71,25 @@ fun SettingsScreen(
     var privacy by remember { mutableStateOf(settings.privacyLevel) }
     var stableSalt by remember { mutableStateOf(settings.stableSalt) }
     var enrollLink by remember { mutableStateOf("") }
-    var serverUrl by remember { mutableStateOf(settings.serverUrl) }
+    // The public name, which is what the operator handed out and what a person recognises. The
+    // endpoint actually dialled is shown beneath it when the two differ, rather than hidden — a
+    // network engineer debugging a connection wants to see where it really goes.
+    var serverUrl by remember { mutableStateOf(settings.serverPublicUrl) }
     var serverPin by remember { mutableStateOf(settings.serverPin) }
     var serverCred by remember { mutableStateOf(settings.serverCredential) }
+    // Enrolling is asynchronous, so these are re-read when its result lands rather than when the
+    // button is pressed — reading them immediately showed the previous server's values and looked
+    // exactly like an enrollment that had silently done nothing.
+    var serverFacts by remember { mutableStateOf(settings.serverFacts) }
+    androidx.compose.runtime.LaunchedEffect(enrollStatus, serverStatus) {
+        serverFacts = settings.serverFacts
+        serverUrl = settings.serverPublicUrl
+        serverPin = settings.serverPin
+        serverCred = settings.serverCredential
+    }
 
     Column(
-        Modifier.fillMaxWidth().verticalScroll(rememberScrollState()).padding(16.dp),
+        Modifier.fillMaxWidth().safeDrawingPadding().verticalScroll(rememberScrollState()).padding(16.dp),
         verticalArrangement = Arrangement.spacedBy(12.dp),
     ) {
         Row(verticalAlignment = Alignment.CenterVertically) {
@@ -129,18 +152,59 @@ fun SettingsScreen(
                 }
                 Text(privacyExplanation(privacy), style = MaterialTheme.typography.bodySmall)
 
+                // At FULL nothing is pseudonymized, so a salt has nothing to act on. Shown
+                // disabled rather than hidden: the setting is still stored and still applies the
+                // moment the level changes, and a control that vanishes hides that fact.
                 Toggle(
                     label = "Stable pseudonyms across runs",
-                    detail = "Lets you compare uploaded runs over time (same SSID reads the same " +
-                        "each time). It also links your uploads together, so leave it off on a " +
-                        "server you don't run yourself.",
-                    checked = stableSalt,
+                    detail = if (privacy == PrivacyLevel.FULL) {
+                        "Not used at this level — nothing is pseudonymized, so there is nothing " +
+                            "to keep stable. Choose balanced or strict to use this."
+                    } else {
+                        "Lets you compare uploaded runs over time (same SSID reads the same " +
+                            "each time). It also links your uploads together, so leave it off on " +
+                            "a server you don't run yourself."
+                    },
+                    checked = stableSalt && privacy != PrivacyLevel.FULL,
+                    enabled = privacy != PrivacyLevel.FULL,
                 ) { stableSalt = it; settings.stableSalt = it }
 
                 TextButton(onClick = onPreviewUpload) { Text("Preview what an upload would send") }
             }
         }
 
+        // ---- account ----
+        Card(Modifier.fillMaxWidth()) {
+            Column(Modifier.padding(14.dp), verticalArrangement = Arrangement.spacedBy(8.dp)) {
+                Text("Account", style = MaterialTheme.typography.titleMedium)
+                if (accountName.isNotBlank()) {
+                    Text("Signed in as $accountName", style = MaterialTheme.typography.bodyMedium)
+                    Text(
+                        "Runs from every device signed in to this account share one history.",
+                        style = MaterialTheme.typography.bodySmall,
+                    )
+                    TextButton(onClick = onSignOut) { Text("Sign out") }
+                } else {
+                    Text(
+                        "Signing in is optional. It links this device to an account on your " +
+                            "server, so several devices share one history — and some servers only " +
+                            "accept uploads from a signed-in device.",
+                        style = MaterialTheme.typography.bodySmall,
+                    )
+                    Button(onClick = onSignIn, enabled = settings.serverConfigured) {
+                        Text("Sign in")
+                    }
+                    if (!settings.serverConfigured) {
+                        Text(
+                            "Enrol with a server first — the account belongs to the server, not " +
+                                "to the app.",
+                            style = MaterialTheme.typography.bodySmall,
+                        )
+                    }
+                }
+            }
+        }
+
         // ---- upload ----
         Card(Modifier.fillMaxWidth()) {
             Column(Modifier.padding(14.dp), verticalArrangement = Arrangement.spacedBy(8.dp)) {
@@ -172,17 +236,38 @@ fun SettingsScreen(
                     onClick = {
                         onEnroll(enrollLink)
                         enrollLink = "" // spent either way; leaving it around invites a retry
-                        serverUrl = settings.serverUrl
-                        serverPin = settings.serverPin
-                        serverCred = settings.serverCredential
                     },
                     enabled = enrollLink.isNotBlank(),
                 ) { Text("Enroll") }
+                // Beside the button that caused it. Enrolling is asynchronous, so without this the
+                // only sign of success is three fields quietly changing further down the card.
+                enrollStatus?.let {
+                    Text(it, style = MaterialTheme.typography.bodySmall)
+                }
 
                 OutlinedTextField(
-                    value = serverUrl, onValueChange = { serverUrl = it; settings.serverUrl = it },
+                    value = serverUrl,
+                    onValueChange = {
+                        serverUrl = it
+                        // Typed by hand there is no discovery to consult, so what was entered is
+                        // both the public name and the endpoint. Setting only one of them would
+                        // leave the app dialling the previous server.
+                        settings.serverUrl = it
+                        settings.serverPublicUrl = it
+                    },
                     label = { Text("Server URL") }, singleLine = true, modifier = Modifier.fillMaxWidth(),
                 )
+                // Directly under the field it explains. Anywhere else it reads as a stray sentence
+                // about some other part of the screen.
+                if (settings.serverUrl.isNotBlank() && settings.serverUrl != settings.serverPublicUrl) {
+                    Text(
+                        "Connects to ${settings.serverUrl} — this server publishes one name and " +
+                            "points devices at another, so its pinned certificate can share a port " +
+                            "with its web interface.",
+                        style = MaterialTheme.typography.bodySmall,
+                        color = LocalContentColor.current.copy(alpha = 0.7f),
+                    )
+                }
                 OutlinedTextField(
                     value = serverPin, onValueChange = { serverPin = it; settings.serverPin = it },
                     label = { Text("Certificate pin (SPKI, base64)") }, singleLine = true,
@@ -209,6 +294,52 @@ fun SettingsScreen(
                 serverStatus?.let {
                     Text(it, style = MaterialTheme.typography.bodySmall)
                 }
+                // What the server reported, placed under the button that asks it rather than among
+                // the fields above: these are facts to read, not settings to apply, and an
+                // editable-looking box that changes nothing is worse than no box at all.
+                //
+                // Monospaced so the addresses line up under each other — column alignment is most
+                // of what makes a list of IPs quicker to read than prose.
+                if (serverFacts.isNotBlank()) {
+                    Surface(
+                        color = MaterialTheme.colorScheme.surfaceVariant,
+                        shape = RoundedCornerShape(8.dp),
+                        modifier = Modifier.fillMaxWidth(),
+                    ) {
+                        Column(
+                            Modifier.padding(horizontal = 12.dp, vertical = 10.dp),
+                            verticalArrangement = Arrangement.spacedBy(2.dp),
+                        ) {
+                            Text(
+                                "WHAT THIS SERVER REPORTS",
+                                style = MaterialTheme.typography.labelSmall,
+                                color = LocalContentColor.current.copy(alpha = 0.7f),
+                            )
+                            // Real columns rather than padded text: the label column has a fixed
+                            // width, so values line up whatever the font does, and a long value
+                            // wraps inside its own column instead of under the labels.
+                            for (line in serverFacts.lines()) {
+                                val label = line.substringBefore('|')
+                                val value = line.substringAfter('|', "")
+                                Row(Modifier.fillMaxWidth()) {
+                                    Text(
+                                        label,
+                                        style = MaterialTheme.typography.bodySmall,
+                                        color = LocalContentColor.current.copy(alpha = 0.7f),
+                                        modifier = Modifier.width(72.dp),
+                                    )
+                                    Text(
+                                        value,
+                                        style = MaterialTheme.typography.bodySmall.copy(
+                                            fontFamily = FontFamily.Monospace,
+                                        ),
+                                        modifier = Modifier.weight(1f),
+                                    )
+                                }
+                            }
+                        }
+                    }
+                }
                 Text(
                     "This app is ${BuildConfig.APP_SEMVER} and speaks probe protocol " +
                         "${app.echo_lot.protocol.Compat.PROTOCOL_VERSION}. It works with servers " +
@@ -236,13 +367,22 @@ private fun privacyExplanation(level: PrivacyLevel): String = when (level) {
 }
 
 @Composable
-private fun Toggle(label: String, detail: String, checked: Boolean, onChange: (Boolean) -> Unit) {
+private fun Toggle(
+    label: String,
+    detail: String,
+    checked: Boolean,
+    enabled: Boolean = true,
+    onChange: (Boolean) -> Unit,
+) {
     Row(Modifier.fillMaxWidth(), verticalAlignment = Alignment.Top) {
         Column(Modifier.weight(1f)) {
-            Text(label, style = MaterialTheme.typography.bodyMedium)
-            Text(detail, style = MaterialTheme.typography.bodySmall)
+            // Dimmed together with the switch, so "this does nothing right now" reads at a glance
+            // instead of only on close inspection.
+            val alpha = if (enabled) 1f else 0.5f
+            Text(label, style = MaterialTheme.typography.bodyMedium, color = LocalContentColor.current.copy(alpha = alpha))
+            Text(detail, style = MaterialTheme.typography.bodySmall, color = LocalContentColor.current.copy(alpha = alpha))
         }
-        Switch(checked = checked, onCheckedChange = onChange)
+        Switch(checked = checked, onCheckedChange = onChange, enabled = enabled)
     }
 }
 

echolot-app/core-archive/src/main/kotlin/app/echo_lot/archive/RunArchive.kt

@@ -31,10 +31,23 @@ data class ArchivedRun(
     val verdict: String? = null,
     @SerialName("finding_count") val findingCount: Int = 0,
     @SerialName("size_bytes") val sizeBytes: Long = 0,
+    /**
+     * How the *archived* document is redacted. Always "full" in practice, because the archive
+     * deliberately keeps the unredacted run - see the package doc. This is not what was uploaded.
+     */
     val anonymization: String = "full",
     /** Whether this run has been accepted by a server, so history can show what is backed up. */
     val uploaded: Boolean = false,
     @SerialName("uploaded_to") val uploadedTo: String? = null,
+    /**
+     * The level the run was *uploaded* at, which is a different document from the archived one.
+     *
+     * Kept separately because conflating the two is actively misleading: the history row showed
+     * the archive's own level ("full") directly beneath "uploaded to fmr", which reads as "the
+     * complete data was uploaded" when a redacted copy had been sent. A privacy display that
+     * overstates what left the device is worse than none.
+     */
+    @SerialName("uploaded_as") val uploadedAs: String? = null,
 )
 
 /**
@@ -119,7 +132,7 @@ class RunArchive(private val dir: File, private val now: () -> Long = System::cu
     fun deleteAll(): Int = list().count { delete(it.id) }
 
     /** Records that a server accepted this run, so history can distinguish backed-up from local. */
-    fun markUploaded(id: String, serverName: String) {
+    fun markUploaded(id: String, serverName: String, uploadedAs: String? = null) {
         val f = File(dir, safe(id) + META_EXT)
         val meta = runCatching { json.decodeFromString(ArchivedRun.serializer(), f.readText()) }.getOrNull()
             ?: return
@@ -127,7 +140,7 @@ class RunArchive(private val dir: File, private val now: () -> Long = System::cu
             f,
             json.encodeToString(
                 ArchivedRun.serializer(),
-                meta.copy(uploaded = true, uploadedTo = serverName),
+                meta.copy(uploaded = true, uploadedTo = serverName, uploadedAs = uploadedAs),
             ),
         )
     }
@@ -181,7 +194,10 @@ class RunArchive(private val dir: File, private val now: () -> Long = System::cu
             id = id,
             savedAtEpochMs = now(),
             startedAt = run["started_at"]?.jsonPrimitive?.content,
-            verdict = doc["summary"]?.jsonObject?.get("verdict")?.jsonPrimitive?.content,
+            // The schema calls it `overall` (Summary.overall); reading `verdict` here silently
+            // yielded null for every run, so the history list's most prominent element - the
+            // coloured verdict - was blank on every row.
+            verdict = doc["summary"]?.jsonObject?.get("overall")?.jsonPrimitive?.content,
             findingCount = (doc["findings"] as? kotlinx.serialization.json.JsonArray)?.size ?: 0,
             sizeBytes = size,
             anonymization = run["privacy"]?.jsonObject?.get("anonymization")?.jsonPrimitive?.content ?: "full",

echolot-app/core-archive/src/test/kotlin/app/echo_lot/archive/RunArchiveTest.kt

@@ -25,7 +25,7 @@ class RunArchiveTest {
     private fun doc(id: String, findings: Int = 1, pad: Int = 0): String {
         val f = (1..findings).joinToString(",") { """{"id":"f$it"}""" }
         return """{"run":{"id":"$id","started_at":"2026-08-01T10:00:00Z","privacy":{"anonymization":"balanced"}},""" +
-            """"findings":[$f],"summary":{"verdict":"warn"},"pad":"${"x".repeat(pad)}"}"""
+            """"findings":[$f],"summary":{"overall":"warn"},"pad":"${"x".repeat(pad)}"}"""
     }
 
     @Test
@@ -120,13 +120,40 @@ class RunArchiveTest {
     fun uploadStateIsRecorded() {
         val a = archive()
         a.save(doc("run-1"))
-        a.markUploaded("run-1", "fmr")
+        a.markUploaded("run-1", "fmr", "balanced")
         val meta = a.list().single()
         assertTrue(meta.uploaded)
         assertEquals("fmr", meta.uploadedTo)
         assertEquals("run-1", meta.id, "marking upload must not disturb the rest of the entry")
     }
 
+    // The archive's own level and the level a run was uploaded at describe *different documents*.
+    // Showing the archive's ("full", because the archive is deliberately unredacted) next to
+    // "uploaded to fmr" reads as "the complete data was uploaded" when a redacted copy was sent —
+    // a privacy display that overstates what left the device is worse than none.
+    @Test
+    fun theUploadedLevelIsRecordedSeparatelyFromTheArchivedOne() {
+        val a = archive()
+        // A real archived document carries no privacy stamp: the anonymizer never runs on the
+        // archive. The shared doc() fixture has one, which is exactly the unrealism that let this
+        // confusion through in the first place.
+        a.save("""{"run":{"id":"run-1"},"findings":[],"summary":{"overall":"green"}}""")
+        a.markUploaded("run-1", "fmr", "balanced")
+        val meta = a.list().single()
+        assertEquals("full", meta.anonymization, "the archived copy is unredacted, by design")
+        assertEquals("balanced", meta.uploadedAs, "the uploaded copy was redacted, and must say so")
+    }
+
+    // The verdict is read from `summary.overall` — the schema's actual field name. Reading
+    // `summary.verdict` silently yielded null for every run, so the history list's most prominent
+    // element was blank on every row while everything else looked fine.
+    @Test
+    fun theVerdictComesFromTheSchemasOverallField() {
+        val a = archive()
+        a.save("""{"run":{"id":"r1"},"findings":[],"summary":{"overall":"yellow"}}""")
+        assertEquals("yellow", a.list().single().verdict)
+    }
+
     @Test
     fun deleteRemovesBothFiles() {
         val a = archive()

echolot-app/core-engine/src/main/kotlin/app/echo_lot/engine/DownstreamMeasurement.kt

@@ -118,7 +118,7 @@ class DownstreamMeasurement(private val ids: IdSource) {
         if (!inOrder) {
             findings.add(
                 finding(
-                    "mtu.fragments_blocked", Category.MTU, Severity.MEDIUM, testId,
+                    FindingRegistry.FRAGMENTS_BLOCKED, testId,
                     "IP fragments do not reach this device",
                     "A fragmented datagram sent in the normal order never arrived. Anything that " +
                         "relies on fragmentation — large DNS answers over UDP, some VPN traffic — " +
@@ -133,7 +133,7 @@ class DownstreamMeasurement(private val ids: IdSource) {
             }.joinToString(" or ")
             findings.add(
                 finding(
-                    "mtu.fragment_reorder_sensitive", Category.MTU, Severity.LOW, testId,
+                    FindingRegistry.FRAGMENT_REORDER_SENSITIVE, testId,
                     "Fragments are dropped when they arrive $which",
                     "In-order fragments are delivered, but the same datagram sent $which is not. " +
                         "Something on the path only reassembles when the first fragment (the one " +
@@ -168,6 +168,10 @@ class DownstreamMeasurement(private val ids: IdSource) {
         trainCount: Int = 100,
         trainSizeBytes: Int = 300,
         trainIntervalUs: Int = 3_000,
+        // DSCP to mark the train with (0-63), or -1 to leave packets unmarked. Pairing a marked
+        // downtrain with the server-observed DSCP of an upstream train is the two-direction
+        // sec.dscp_ecn_survival measurement.
+        trainDscp: Int = -1,
     ): Pair<List<Test>, List<Finding>> {
         val tests = ArrayList<Test>()
         val findings = ArrayList<Finding>()
@@ -175,7 +179,8 @@ class DownstreamMeasurement(private val ids: IdSource) {
         val df = bigSend(credential, sessionId, control, probe, sessionRef, sizes, df = true)
         val frag = bigSend(credential, sessionId, control, probe, sessionRef, sizes, df = false)
         val train = downTrain(
-            credential, sessionId, control, probe, sessionRef, trainCount, trainSizeBytes, trainIntervalUs,
+            credential, sessionId, control, probe, sessionRef, trainCount, trainSizeBytes,
+            trainIntervalUs, trainDscp,
         )
 
         tests.add(df.test); tests.add(frag.test); tests.add(train.test)
@@ -198,7 +203,7 @@ class DownstreamMeasurement(private val ids: IdSource) {
             if (ipMtu < 1500) {
                 findings.add(
                     finding(
-                        "mtu.reduced_downstream", Category.MTU, Severity.LOW, df.test.id,
+                        FindingRegistry.MTU_REDUCED_DOWNSTREAM, df.test.id,
                         "Downstream path MTU is $ipMtu bytes, below 1500",
                         "The largest datagram that reached this device without fragmenting was " +
                             "$pathMtu bytes of payload ($ipMtu on the wire). Tunnels (PPPoE, VPN, " +
@@ -213,7 +218,7 @@ class DownstreamMeasurement(private val ids: IdSource) {
             if (fragLargest <= pathMtu && sizes.any { it > pathMtu }) {
                 findings.add(
                     finding(
-                        "mtu.downstream_blackhole", Category.MTU, Severity.MEDIUM, frag.test.id,
+                        FindingRegistry.MTU_DOWNSTREAM_BLACKHOLE, frag.test.id,
                         "Datagrams above $pathMtu bytes are dropped downstream, fragmented or not",
                         "Nothing larger than $pathMtu bytes arrived, even when the network was " +
                             "free to fragment it. Traffic that relies on large responses will " +
@@ -226,7 +231,7 @@ class DownstreamMeasurement(private val ids: IdSource) {
         if (train.received == 0) {
             findings.add(
                 finding(
-                    "connectivity.downstream_blocked", Category.CONNECTIVITY, Severity.HIGH, train.test.id,
+                    FindingRegistry.DOWNSTREAM_BLOCKED, train.test.id,
                     "No server-initiated packets arrived",
                     "The server sent ${train.sent} packets toward this device and none arrived, " +
                         "while the round-trip echo worked. Something on the path forwards replies " +
@@ -236,7 +241,7 @@ class DownstreamMeasurement(private val ids: IdSource) {
         } else if (train.lossPct >= 5.0) {
             findings.add(
                 finding(
-                    "connectivity.downstream_loss", Category.CONNECTIVITY, Severity.MEDIUM, train.test.id,
+                    FindingRegistry.LOSS_DOWNSTREAM, train.test.id,
                     "Downstream loss of ${round1(train.lossPct)}%",
                     "${train.sent - train.received} of ${train.sent} packets sent toward this " +
                         "device were lost. Downstream loss is invisible to a round-trip test, " +
@@ -247,7 +252,7 @@ class DownstreamMeasurement(private val ids: IdSource) {
         if (train.reordered > 0) {
             findings.add(
                 finding(
-                    "connectivity.downstream_reorder", Category.CONNECTIVITY, Severity.LOW, train.test.id,
+                    FindingRegistry.DOWNSTREAM_REORDER, train.test.id,
                     "${train.reordered} downstream packet(s) arrived out of order",
                     "Packets arrived in a different order than they were sent. Usually per-packet " +
                         "load balancing across links; harmless for most traffic, not for all of it.",
@@ -338,15 +343,19 @@ class DownstreamMeasurement(private val ids: IdSource) {
 
     private fun downTrain(
         credential: String, sessionId: String, control: ControlClient, probe: ProbeSession,
-        sessionRef: String, count: Int, sizeBytes: Int, intervalUs: Int,
+        sessionRef: String, count: Int, sizeBytes: Int, intervalUs: Int, dscp: Int = -1,
     ): TrainResult {
         val testId = ids.uuid()
         val started = ids.monoNs()
 
+        // dscp is only sent when requested: an older server rejects unknown-value problems
+        // louder than absent keys, and unmarked is the correct default for a plain loss train.
+        val dscpField = if (dscp in 0..63) ""","dscp":$dscp""" else ""
         val reply = runCatching {
             control.action(
                 credential, sessionId,
-                """{"action":"downtrain","count":$count,"size_bytes":$sizeBytes,"interval_us":$intervalUs}""",
+                """{"action":"downtrain","count":$count,"size_bytes":$sizeBytes,""" +
+                    """"interval_us":$intervalUs$dscpField}""",
             )
         }
         if (reply.isFailure) {
@@ -394,6 +403,12 @@ class DownstreamMeasurement(private val ids: IdSource) {
                 interArrivalMsAvg = interArrival.average().takeIf { interArrival.isNotEmpty() }?.let(::round1),
                 interArrivalMsMax = interArrival.maxOrNull()?.let(::round1),
                 sendIntervalUs = intervalUs,
+                dscpRequested = dscp.takeIf { it in 0..63 },
+                // The server says whether it could actually mark (dscp_applied); recorded so a
+                // survival comparison never blames the path for a marking the sender skipped.
+                dscpApplied = reply.getOrNull()?.let {
+                    Regex("\"dscp_applied\"\\s*:\\s*(true|false)").find(it)?.groupValues?.get(1)?.toBoolean()
+                },
             ),
         ) as JsonObject
 
@@ -414,9 +429,17 @@ class DownstreamMeasurement(private val ids: IdSource) {
 
     // ---- helpers ----------------------------------------------------------------------
 
-    private fun finding(code: String, cat: Category, sev: Severity, testId: String, title: String, desc: String) =
+    /**
+     * Builds a finding from a registry entry, which supplies the code, category and severity.
+     *
+     * Taking a [FindingSpec] rather than three loose values is the point: a typo becomes a
+     * compile error, and two call sites cannot disagree about which category a finding belongs
+     * to - a disagreement that would split one fault across two verdict lights.
+     */
+    private fun finding(spec: FindingSpec, testId: String, title: String, desc: String) =
         Finding(
-            id = ids.uuid(), code = code, category = cat, severity = sev, confidence = Confidence.HIGH,
+            id = ids.uuid(), code = spec.code, category = spec.category, severity = spec.severity,
+            confidence = Confidence.HIGH,
             title = title, description = desc, evidenceRefs = listOf(EvidenceRef(testId)),
         )
 
@@ -482,4 +505,6 @@ data class DownTrainMetrics(
     @SerialName("inter_arrival_ms_avg") val interArrivalMsAvg: Double? = null,
     @SerialName("inter_arrival_ms_max") val interArrivalMsMax: Double? = null,
     @SerialName("send_interval_us") val sendIntervalUs: Int,
+    @SerialName("dscp_requested") val dscpRequested: Int? = null,
+    @SerialName("dscp_applied") val dscpApplied: Boolean? = null,
 )

echolot-app/core-engine/src/main/kotlin/app/echo_lot/engine/ServerMeasurement.kt

@@ -47,6 +47,14 @@ class ServerMeasurement(
          * it is a flag rather than an assumption.
          */
         val downstream: Boolean = true,
+        /**
+         * Throughput moves real data — a 5-second run at 50 Mbps is about 30 MB — so it is off
+         * unless asked for. On a metered mobile connection that is the user's money, and a
+         * measurement tool that spends it without being told to is not one people keep installed.
+         */
+        val throughput: Boolean = false,
+        @Suppress("unused") val throughputSeconds: Int = 5,
+        @Suppress("unused") val throughputKbps: Int = 50_000,
     )
 
     fun run(cfg: Config): MeasurementDocument {
@@ -80,6 +88,14 @@ class ServerMeasurement(
             tests.add(test)
             allFindings.addAll(findings)
 
+            // The upstream train needs no grant and no capability beyond udp-probe itself; a
+            // server that predates trains simply never answers the report request, which the
+            // measurement reports as exactly that ambiguity rather than as network loss.
+            val (utTest, utFindings) = UpstreamTrainMeasurement(ids)
+                .run(ps, sessionRef = "sess-1")
+            tests.add(utTest)
+            allFindings.addAll(utFindings)
+
             // Downstream needs a session the server has already seen traffic from — the echo
             // train just provided that — and a server that advertises the grants. Skipped
             // quietly against an older server rather than reported as a failure of the network.
@@ -89,6 +105,15 @@ class ServerMeasurement(
                 tests.addAll(dsTests)
                 allFindings.addAll(dsFindings)
             }
+
+            if (cfg.throughput && profile.supports("throughput")) {
+                val (tpTest, tpFindings) = ThroughputMeasurement(ids).run(
+                    cfg.credential, session.sessionId, control, ps, sessionRef = "sess-1",
+                    durationS = cfg.throughputSeconds, kbps = cfg.throughputKbps,
+                )
+                tests.add(tpTest)
+                allFindings.addAll(tpFindings)
+            }
         }
 
         control.deleteSession(cfg.credential, session.sessionId)
@@ -191,11 +216,11 @@ class ServerMeasurement(
 
         val findings = ArrayList<Finding>()
         if (received == 0) {
-            findings.add(finding("nat.udp_unreachable", Category.CONNECTIVITY, Severity.HIGH, testId,
+            findings.add(finding(FindingRegistry.UDP_UNREACHABLE, testId,
                 "No UDP echo replies from the server",
                 "Every ECHO probe to the server's UDP data plane was lost — the path blocks or drops the session's UDP traffic."))
         } else if (lossPct >= 20.0) {
-            findings.add(finding("connectivity.udp_loss", Category.CONNECTIVITY, Severity.MEDIUM, testId,
+            findings.add(finding(FindingRegistry.UDP_LOSS, testId,
                 "High UDP loss to the server (${round1(lossPct)}%)",
                 "A large fraction of ECHO probes were lost, indicating an unreliable UDP path."))
         }
@@ -203,14 +228,14 @@ class ServerMeasurement(
         directional?.let { d ->
             when {
                 d.noneReachedServer && received == 0 -> findings.add(
-                    finding("nat.udp_unreachable_upstream", Category.CONNECTIVITY, Severity.HIGH, testId,
+                    finding(FindingRegistry.UDP_UNREACHABLE_UPSTREAM, testId,
                         "Nothing reached the server",
                         "The server received none of the ${d.sent} probes, so the traffic is being " +
                             "dropped on the way out, not on the way back. A firewall or NAT on " +
                             "this side of the path is the place to look."),
                 )
                 d.lossUpstreamPct >= 2.0 -> findings.add(
-                    finding("connectivity.loss_upstream", Category.CONNECTIVITY, Severity.MEDIUM, testId,
+                    finding(FindingRegistry.LOSS_UPSTREAM, testId,
                         "${d.lossUpstreamPct} % of probes were lost on the way to the server",
                         "${d.lostUpstream} of ${d.sent} probes never reached the server. The " +
                             "return path is not implicated: replies came back for everything that " +
@@ -219,7 +244,7 @@ class ServerMeasurement(
             }
             if (d.lossDownstreamPct >= 2.0) {
                 findings.add(
-                    finding("connectivity.loss_downstream", Category.CONNECTIVITY, Severity.MEDIUM, testId,
+                    finding(FindingRegistry.LOSS_DOWNSTREAM, testId,
                         "${d.lossDownstreamPct} % of replies were lost on the way back",
                         "The server received ${d.seenByServer} probes and answered them, but " +
                             "${d.lostDownstream} of those replies never arrived. The outbound path " +
@@ -229,7 +254,7 @@ class ServerMeasurement(
         }
 
         if (natRebinding) {
-            findings.add(finding("nat.udp_rebinding", Category.NAT, Severity.MEDIUM, testId,
+            findings.add(finding(FindingRegistry.NAT_UDP_REBINDING, testId,
                 "NAT remapped the UDP source port mid-flow",
                 "The server observed more than one source port for this session (${observedPorts.joinToString()}), i.e. a NAT with a short UDP mapping or per-packet remapping."))
         }
@@ -259,9 +284,17 @@ class ServerMeasurement(
         }
     }
 
-    private fun finding(code: String, cat: Category, sev: Severity, testId: String, title: String, desc: String) =
+    /**
+     * Builds a finding from a registry entry, which supplies the code, category and severity.
+     *
+     * Taking a [FindingSpec] rather than three loose values is the point: a typo becomes a
+     * compile error, and two call sites cannot disagree about which category a finding belongs
+     * to - a disagreement that would split one fault across two verdict lights.
+     */
+    private fun finding(spec: FindingSpec, testId: String, title: String, desc: String) =
         Finding(
-            id = ids.uuid(), code = code, category = cat, severity = sev, confidence = Confidence.HIGH,
+            id = ids.uuid(), code = spec.code, category = spec.category, severity = spec.severity,
+            confidence = Confidence.HIGH,
             title = title, description = desc, evidenceRefs = listOf(EvidenceRef(testId)),
         )
 

echolot-app/core-engine/src/main/kotlin/app/echo_lot/engine/ThroughputMeasurement.kt

@@ -0,0 +1,338 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package app.echo_lot.engine
+
+import app.echo_lot.measurement.*
+import app.echo_lot.protocol.ControlClient
+import app.echo_lot.protocol.ProbeSession
+import app.echo_lot.protocol.Wire
+import kotlinx.serialization.SerialName
+import kotlinx.serialization.Serializable
+import kotlinx.serialization.json.Json
+import kotlinx.serialization.json.JsonObject
+import kotlinx.serialization.json.encodeToJsonElement
+import kotlinx.serialization.json.jsonArray
+import kotlinx.serialization.json.jsonObject
+import kotlinx.serialization.json.jsonPrimitive
+
+/**
+ * Downstream throughput: the server sends at a paced rate for a bounded time and the client
+ * measures what arrives (`perf.throughput_udp`).
+ *
+ * The number this produces is only meaningful with a qualifier attached, and getting that
+ * qualifier right is most of the work here. A throughput test reports the *smallest* limit on the
+ * path, and the sender's own ceiling is one of the candidates: if the server was asked for 50 Mbps
+ * and 50 Mbps arrived, the network was never the constraint and "50 Mbps" says nothing about it.
+ * Reporting that as a capacity measurement would be a confident lie, so the result always carries
+ * [ThroughputMetrics.limitedBy] and a finding is only raised when the network is actually
+ * implicated.
+ *
+ * Comparing against the *sender's* count rather than the requested rate is the other half: the
+ * server reports how much it actually put on the wire, and the gap between that and what arrived
+ * is the loss. A receiver alone cannot tell "the network dropped it" from "the sender never sent
+ * it", and guessing turns a healthy server-side limit into a phantom network fault.
+ */
+class ThroughputMeasurement(private val ids: IdSource) {
+
+    private val json = Json { encodeDefaults = true; explicitNulls = true }
+
+    fun run(
+        credential: String,
+        sessionId: String,
+        control: ControlClient,
+        probe: ProbeSession,
+        sessionRef: String,
+        durationS: Int = 5,
+        kbps: Int = 50_000,
+        sizeBytes: Int = 1200,
+    ): Pair<Test, List<Finding>> {
+        val testId = ids.uuid()
+        val started = ids.monoNs()
+
+        val reply = runCatching {
+            control.action(
+                credential, sessionId,
+                """{"action":"throughput","direction":"down","duration_s":$durationS,""" +
+                    """"kbps":$kbps,"size_bytes":$sizeBytes}""",
+            )
+        }
+        if (reply.isFailure) {
+            return Test(
+                id = testId, type = TestType.PERF_THROUGHPUT_UDP, sessionRef = sessionRef, tier = Tier.APP,
+                startedMonoNs = started, endedMonoNs = ids.monoNs(),
+                status = TestStatus.UNSUPPORTED,
+                error = TestError("action_refused", reply.exceptionOrNull()?.message ?: "throughput refused"),
+            ) to emptyList()
+        }
+
+        // The server may have shortened the run to fit its own byte budget; listen for what it
+        // actually promised, not for what we asked.
+        val plannedMs = parseInt(reply.getOrNull(), "duration_ms") ?: (durationS * 1000)
+
+        // A margin past the planned end so the tail of the run is not counted as loss: packets
+        // still in flight when we stop listening were not dropped, they were merely late.
+        val received = probe.collectGranted(plannedMs + 1_500L)
+            .filter { it.type == Wire.TYPE_THROUGHPUT_DATA }
+
+        val bytes = received.sumOf { it.sizeBytes.toLong() }
+        val spanNs = if (received.size >= 2) {
+            received.maxOf { it.tRxNs } - received.minOf { it.tRxNs }
+        } else {
+            0L
+        }
+        // Measured over the arrival span rather than our listening window, which includes the
+        // request round trip and the trailing margin and would understate the rate.
+        val receivedKbps = if (spanNs > 0) (bytes * 8 * 1_000_000 / spanNs).toInt() else 0
+
+        val sender = senderReport(control, credential, sessionId)
+        val sentPackets = sender?.packets ?: 0
+        val lossPct = if (sentPackets > 0) {
+            round2((sentPackets - received.size).coerceAtLeast(0) * 100.0 / sentPackets)
+        } else {
+            null
+        }
+
+        // Only a run the *clock* ended measured the network. One stopped by our own byte budget
+        // or rate ceiling measured this server.
+        val limitedBy = sender?.limitedBy ?: "unknown"
+        val networkLimited = limitedBy == "duration" &&
+            sender != null && receivedKbps > 0 && receivedKbps < sender.kbps * 9 / 10
+
+        val metrics = json.encodeToJsonElement(
+            ThroughputMetrics(
+                requestedKbps = kbps,
+                plannedDurationMs = plannedMs,
+                packetsReceived = received.size,
+                bytesReceived = bytes,
+                receivedKbps = receivedKbps,
+                senderPackets = sender?.packets,
+                senderBytes = sender?.bytes,
+                senderKbps = sender?.kbps,
+                lossPct = lossPct,
+                limitedBy = limitedBy,
+                measuresNetwork = networkLimited,
+            ),
+        ) as JsonObject
+
+        val findings = ArrayList<Finding>()
+        when {
+            sender == null -> Unit // no sender report: nothing can be concluded, so nothing is
+            received.isEmpty() -> findings.add(
+                finding(
+                    FindingRegistry.THROUGHPUT_NO_DELIVERY, testId,
+                    "No throughput traffic arrived",
+                    "The server sent ${sender.packets} packets and none arrived. This is a " +
+                        "connectivity fault rather than a slow link.",
+                ),
+            )
+            networkLimited -> findings.add(
+                finding(
+                    FindingRegistry.THROUGHPUT_BELOW_OFFERED, testId,
+                    "Downstream throughput ${receivedKbps / 1000} Mbit/s, below the " +
+                        "${sender.kbps / 1000} Mbit/s offered",
+                    "The server sent at ${sender.kbps / 1000} Mbit/s for the full run and " +
+                        "${receivedKbps / 1000} Mbit/s arrived" +
+                        (lossPct?.let { ", losing $it % of packets" } ?: "") +
+                        ". The path could not carry what was offered.",
+                ),
+            )
+        }
+
+        return Test(
+            id = testId, type = TestType.PERF_THROUGHPUT_UDP, sessionRef = sessionRef, tier = Tier.APP,
+            startedMonoNs = started, endedMonoNs = ids.monoNs(),
+            status = if (received.isEmpty()) TestStatus.FAILED else TestStatus.OK,
+            metrics = metrics,
+        ) to findings
+    }
+
+    /**
+     * Upstream throughput: the client sends, the server counts.
+     *
+     * The mirror image of the downstream case, and it needs no grant — the client is generating
+     * its own traffic, so there is no amplification to gate. What it does need is the server's
+     * count: only the far end knows how much arrived, and without that number a sender can
+     * measure how fast it can *transmit*, which is not the same question and is usually just the
+     * speed of the local NIC.
+     */
+    fun runUpstream(
+        credential: String,
+        sessionId: String,
+        control: ControlClient,
+        probe: ProbeSession,
+        sessionRef: String,
+        durationS: Int = 5,
+        kbps: Int = 20_000,
+        sizeBytes: Int = 1200,
+    ): Pair<Test, List<Finding>> {
+        val testId = ids.uuid()
+        val started = ids.monoNs()
+
+        // Zeroes the server's counter so this run measures itself rather than inheriting the
+        // packets of an earlier one on the same session.
+        val reply = runCatching {
+            control.action(credential, sessionId, """{"action":"throughput","direction":"up"}""")
+        }
+        if (reply.isFailure) {
+            return Test(
+                id = testId, type = TestType.PERF_THROUGHPUT_UDP, sessionRef = sessionRef, tier = Tier.APP,
+                startedMonoNs = started, endedMonoNs = ids.monoNs(),
+                status = TestStatus.UNSUPPORTED,
+                error = TestError("action_refused", reply.exceptionOrNull()?.message ?: "refused"),
+            ) to emptyList()
+        }
+
+        val sent = probe.sendThroughput(durationS * 1000L, kbps, sizeBytes)
+        // A moment for the tail of the run to arrive; counting still-in-flight packets as lost
+        // would inflate the loss figure by whatever the path's delay happens to be.
+        Thread.sleep(500)
+        val seen = upstreamCount(control, credential, sessionId)
+
+        val lossPct = if (sent.packets > 0 && seen != null) {
+            round2((sent.packets - seen.packets).coerceAtLeast(0) * 100.0 / sent.packets)
+        } else {
+            null
+        }
+        // The receiver's rate is the measurement. The sender's is what we managed to emit, which
+        // is a property of this phone and its radio, not of the network.
+        val achievedKbps = seen?.kbps ?: 0
+
+        val metrics = json.encodeToJsonElement(
+            UpstreamThroughputMetrics(
+                requestedKbps = kbps,
+                sentPackets = sent.packets,
+                sentBytes = sent.bytes,
+                sentKbps = sent.kbps,
+                receivedPackets = seen?.packets,
+                receivedBytes = seen?.bytes,
+                receivedKbps = achievedKbps,
+                lossPct = lossPct,
+                // Same honesty rule as downstream: if what arrived matches what we offered, the
+                // path was never the constraint and this number says nothing about it.
+                measuresNetwork = seen != null && achievedKbps > 0 && achievedKbps < sent.kbps * 9 / 10,
+            ),
+        ) as JsonObject
+
+        val findings = ArrayList<Finding>()
+        if (seen != null && seen.packets == 0 && sent.packets > 0) {
+            findings.add(
+                finding(
+                    FindingRegistry.THROUGHPUT_NO_DELIVERY, testId,
+                    "No upstream traffic reached the server",
+                    "This device sent ${sent.packets} packets and the server received none. " +
+                        "That is a connectivity fault on the outbound path rather than a slow link.",
+                ),
+            )
+        } else if (lossPct != null && lossPct >= 2.0) {
+            findings.add(
+                finding(
+                    FindingRegistry.THROUGHPUT_BELOW_OFFERED, testId,
+                    "Upstream loss of $lossPct % at ${sent.kbps / 1000} Mbit/s",
+                    "The server received ${seen?.packets} of the ${sent.packets} packets this " +
+                        "device sent. The outbound path could not carry what was offered.",
+                ),
+            )
+        }
+
+        return Test(
+            id = testId, type = TestType.PERF_THROUGHPUT_UDP, sessionRef = sessionRef, tier = Tier.APP,
+            startedMonoNs = started, endedMonoNs = ids.monoNs(),
+            status = if (seen == null || seen.packets == 0) TestStatus.FAILED else TestStatus.OK,
+            metrics = metrics,
+        ) to findings
+    }
+
+    private data class UpstreamCount(val packets: Int, val bytes: Long, val kbps: Int)
+
+    /** The server's tally for this session's upstream run. */
+    private fun upstreamCount(
+        control: ControlClient, credential: String, sessionId: String,
+    ): UpstreamCount? = runCatching {
+        val o = Json.parseToJsonElement(control.observations(credential, sessionId))
+            .jsonObject["throughput_up"]?.jsonObject ?: return null
+        UpstreamCount(
+            packets = o["packets"]?.jsonPrimitive?.content?.toIntOrNull() ?: 0,
+            bytes = o["bytes"]?.jsonPrimitive?.content?.toLongOrNull() ?: 0,
+            kbps = o["kbps"]?.jsonPrimitive?.content?.toIntOrNull() ?: 0,
+        )
+    }.getOrNull()
+
+    private data class SenderReport(
+        val packets: Int, val bytes: Long, val kbps: Int, val limitedBy: String,
+    )
+
+    /** The server's own account of the run, from the observations API. */
+    private fun senderReport(
+        control: ControlClient, credential: String, sessionId: String,
+    ): SenderReport? = runCatching {
+        val arr = Json.parseToJsonElement(control.observations(credential, sessionId))
+            .jsonObject["throughput"]?.jsonArray ?: return null
+        val last = arr.lastOrNull()?.jsonObject ?: return null
+        SenderReport(
+            packets = last["packets"]?.jsonPrimitive?.content?.toIntOrNull() ?: 0,
+            bytes = last["bytes"]?.jsonPrimitive?.content?.toLongOrNull() ?: 0,
+            kbps = last["kbps"]?.jsonPrimitive?.content?.toIntOrNull() ?: 0,
+            limitedBy = last["limited_by"]?.jsonPrimitive?.content ?: "unknown",
+        )
+    }.getOrNull()
+
+    private fun parseInt(body: String?, key: String): Int? =
+        body?.let { Regex("\"$key\"\\s*:\\s*(-?\\d+)").find(it)?.groupValues?.get(1)?.toIntOrNull() }
+
+    /**
+     * Builds a finding from a registry entry, which supplies the code, category and severity.
+     *
+     * Taking a [FindingSpec] rather than three loose values is the point: a typo becomes a
+     * compile error, and two call sites cannot disagree about which category a finding belongs
+     * to - a disagreement that would split one fault across two verdict lights.
+     */
+    private fun finding(spec: FindingSpec, testId: String, title: String, desc: String) =
+        Finding(
+            id = ids.uuid(), code = spec.code, category = spec.category, severity = spec.severity,
+            confidence = Confidence.HIGH,
+            title = title, description = desc, evidenceRefs = listOf(EvidenceRef(testId)),
+        )
+
+    private fun round2(v: Double) = Math.round(v * 100.0) / 100.0
+}
+
+/** Metrics for perf.throughput_udp in the upstream direction. */
+@Serializable
+data class UpstreamThroughputMetrics(
+    val direction: String = "up",
+    @SerialName("requested_kbps") val requestedKbps: Int,
+    @SerialName("sent_packets") val sentPackets: Int,
+    @SerialName("sent_bytes") val sentBytes: Long,
+    /** What this device managed to emit — a property of the phone and its radio, not the path. */
+    @SerialName("sent_kbps") val sentKbps: Int,
+    @SerialName("received_packets") val receivedPackets: Int? = null,
+    @SerialName("received_bytes") val receivedBytes: Long? = null,
+    /** What arrived, measured by the only party that can measure it. This is the result. */
+    @SerialName("received_kbps") val receivedKbps: Int,
+    @SerialName("loss_pct") val lossPct: Double? = null,
+    @SerialName("measures_network") val measuresNetwork: Boolean,
+)
+
+/** Metrics for perf.throughput_udp. */
+@Serializable
+data class ThroughputMetrics(
+    val direction: String = "down",
+    @SerialName("requested_kbps") val requestedKbps: Int,
+    @SerialName("planned_duration_ms") val plannedDurationMs: Int,
+    @SerialName("packets_received") val packetsReceived: Int,
+    @SerialName("bytes_received") val bytesReceived: Long,
+    @SerialName("received_kbps") val receivedKbps: Int,
+    @SerialName("sender_packets") val senderPackets: Int? = null,
+    @SerialName("sender_bytes") val senderBytes: Long? = null,
+    @SerialName("sender_kbps") val senderKbps: Int? = null,
+    /** Against the sender's count, so a server-side limit is never counted as network loss. */
+    @SerialName("loss_pct") val lossPct: Double? = null,
+    /** What ended the run: duration | budget | rate | send_error | unknown. */
+    @SerialName("limited_by") val limitedBy: String,
+    /**
+     * Whether this number says anything about the network. False when the sender's own ceiling
+     * was the binding constraint — in which case the rate is a property of the test, not the path.
+     */
+    @SerialName("measures_network") val measuresNetwork: Boolean,
+)

echolot-app/core-engine/src/main/kotlin/app/echo_lot/engine/UpstreamTrainMeasurement.kt

@@ -0,0 +1,159 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package app.echo_lot.engine
+
+import app.echo_lot.measurement.*
+import app.echo_lot.protocol.ProbeSession
+import kotlinx.serialization.SerialName
+import kotlinx.serialization.Serializable
+import kotlinx.serialization.json.Json
+import kotlinx.serialization.json.JsonObject
+import kotlinx.serialization.json.encodeToJsonElement
+
+/**
+ * train.udp_updown — the client sends a paced train (types 0x03), then asks the server what
+ * arrived (0x04 → 0x05) and lines both views up per sequence number.
+ *
+ * This is the measurement a round trip cannot make: an echo run only says "lost somewhere", the
+ * train's two ledgers say lost on the way OUT, specifically, because the server's report names
+ * exactly which sequence numbers reached it. The downstream direction has its own test
+ * (train.udp_downstream) under a grant; this one needs none, since the client generates all the
+ * traffic itself.
+ *
+ * The evidence is the schema's columnar TrainEvidence: one index per sent packet, with the
+ * server-side columns null where a packet never arrived. Server timestamps are on the server's
+ * own clock — only differences within that clock mean anything unless time.server_offset maps
+ * them (two-clock rule).
+ */
+class UpstreamTrainMeasurement(private val ids: IdSource) {
+
+    private val json = Json { encodeDefaults = true; explicitNulls = true }
+
+    fun run(
+        probe: ProbeSession,
+        sessionRef: String,
+        count: Int = 200,
+        sizeBytes: Int = 200,
+        interPacketMs: Long = 5,
+    ): Pair<Test, List<Finding>> {
+        val testId = ids.uuid()
+        val started = ids.monoNs()
+        // The id only needs to be unique within this session; a clash across sessions is
+        // meaningless because trains are buffered per session on the server.
+        val trainId = (System.nanoTime() and 0x7FFFFFFF).toInt()
+
+        val sent = probe.sendTrain(trainId, count, sizeBytes, interPacketMs)
+        // Let the tail arrive before asking for the ledger; packets still in flight when the
+        // report is cut would read as upstream loss.
+        Thread.sleep(300)
+        val report = probe.trainReport(trainId)
+
+        if (report == null) {
+            return Test(
+                id = testId, type = TestType.TRAIN_UDP_UPDOWN, sessionRef = sessionRef,
+                tier = Tier.APP, startedMonoNs = started, endedMonoNs = ids.monoNs(),
+                status = TestStatus.FAILED,
+                // Honest ambiguity: an old server drops 0x04 silently, and a lost report looks
+                // identical from here. Neither says anything about the train itself.
+                error = TestError(
+                    "no_report",
+                    "no train report arrived — the report was lost, or the server predates trains",
+                ),
+            ) to emptyList()
+        }
+
+        val bySeq = report.rows.associateBy { it.seq }
+        fun col255(v: Int): Int? = v.takeIf { it != 255 } // 255 = "not observed" on the wire
+
+        val evidence = TrainEvidence(
+            epochMonoNs = started,
+            seq = sent.map { it.seq },
+            tTxNs = sent.map { it.tTxNs },
+            tSrvRxNs = sent.map { bySeq[it.seq]?.tRxNs },
+            tRxNs = sent.map { null }, // upstream only: nothing comes back per packet
+            sizeBytes = sent.map { it.sizeBytes },
+            ttlSeenByServer = sent.map { bySeq[it.seq]?.let { r -> col255(r.ttl) } },
+            dscpSeenByServer = sent.map { bySeq[it.seq]?.let { r -> col255(r.dscp) } },
+            ecnSeenByServer = sent.map { bySeq[it.seq]?.let { r -> col255(r.ecn) } },
+            evidenceTruncated = report.truncated,
+        ).toEvidence()
+
+        // Loss against the server's total count, not its row list: rows past the server's buffer
+        // cap are counted but not kept, and treating them as lost would invent loss exactly on
+        // the biggest trains.
+        val lossPct = if (sent.isEmpty()) 0.0 else {
+            (sent.size - report.received).coerceAtLeast(0) * 100.0 / sent.size
+        }
+        val metrics = json.encodeToJsonElement(
+            UpstreamTrainMetrics(
+                sent = sent.size,
+                receivedByServer = report.received,
+                lossPct = round1(lossPct),
+                reportPartsExpected = report.partsExpected,
+                reportPartsReceived = report.partsReceived,
+                truncated = report.truncated,
+            ),
+        ) as JsonObject
+
+        val findings = ArrayList<Finding>()
+        if (sent.isNotEmpty() && report.received == 0) {
+            findings.add(
+                Finding(
+                    id = ids.uuid(),
+                    code = FindingRegistry.UDP_UNREACHABLE_UPSTREAM.code,
+                    category = FindingRegistry.UDP_UNREACHABLE_UPSTREAM.category,
+                    severity = FindingRegistry.UDP_UNREACHABLE_UPSTREAM.severity,
+                    confidence = Confidence.HIGH,
+                    title = "The server received none of ${sent.size} upstream packets",
+                    description = "Every train packet vanished on the way out, while the " +
+                        "report request's reply made it back — the outbound path drops this " +
+                        "traffic, the return path works.",
+                    evidenceRefs = listOf(EvidenceRef(testId)),
+                ),
+            )
+        } else if (lossPct >= 2.0) {
+            findings.add(
+                Finding(
+                    id = ids.uuid(),
+                    code = FindingRegistry.LOSS_UPSTREAM.code,
+                    category = FindingRegistry.LOSS_UPSTREAM.category,
+                    severity = FindingRegistry.LOSS_UPSTREAM.severity,
+                    confidence = Confidence.HIGH,
+                    title = "Upstream loss of ${round1(lossPct)} %",
+                    description = "The server received ${report.received} of the ${sent.size} " +
+                        "packets this device sent, and its per-sequence ledger names the " +
+                        "missing ones. This is outbound loss specifically; the return path " +
+                        "delivered the report.",
+                    evidenceRefs = listOf(EvidenceRef(testId)),
+                ),
+            )
+        }
+
+        val status = when {
+            report.received == 0 && sent.isNotEmpty() -> TestStatus.FAILED
+            report.partsReceived < report.partsExpected -> TestStatus.PARTIAL
+            else -> TestStatus.OK
+        }
+        return Test(
+            id = testId, type = TestType.TRAIN_UDP_UPDOWN, sessionRef = sessionRef, tier = Tier.APP,
+            startedMonoNs = started, endedMonoNs = ids.monoNs(),
+            status = status, evidence = evidence, metrics = metrics,
+        ) to findings
+    }
+
+    private fun round1(v: Double) = Math.round(v * 10.0) / 10.0
+}
+
+/** Metrics for train.udp_updown. */
+@Serializable
+data class UpstreamTrainMetrics(
+    val sent: Int,
+    /** The server's total count — includes packets past its row buffer (counted, not listed). */
+    @SerialName("received_by_server") val receivedByServer: Int,
+    @SerialName("loss_pct") val lossPct: Double,
+    @SerialName("report_parts_expected") val reportPartsExpected: Int,
+    @SerialName("report_parts_received") val reportPartsReceived: Int,
+    /** The server's row buffer overflowed: rows are a sample, the count is still complete. */
+    val truncated: Boolean,
+)

echolot-app/core-engine/src/test/kotlin/app/echo_lot/engine/LiveDownstreamTest.kt

@@ -44,7 +44,9 @@ class LiveDownstreamTest {
         for (t in tests) println("${t.type}  status=${t.status}  metrics=${t.metrics}")
         for (f in findings) println("finding ${f.code} [${f.severity}] ${f.title}")
 
-        assertEquals(3, tests.size, "expected pmtud_down, frag_delivery and a downstream train")
+        // Assert on what is present, not on how many: adding a measurement should not be a
+        // test edit. (It was, once — hence the note.)
+        assertTrue(tests.size >= 3, "expected at least the three downstream tests, got ${tests.size}")
         val byType = tests.associateBy { it.type }
 
         val pmtud = assertNotNull(byType[TestType.MTU_PMTUD_DOWN], "no mtu.pmtud_down test")
@@ -58,6 +60,17 @@ class LiveDownstreamTest {
         val frag = assertNotNull(byType[TestType.MTU_FRAG_DELIVERY], "no mtu.frag_delivery test")
         assertNotNull(frag.metrics?.get("largest_delivered_bytes"))
 
+        // Fragment ordering runs only when fragments arrive at all, and only against a server
+        // that can craft them — so it is checked when present rather than required.
+        byType[TestType.MTU_FRAG_ORDERING]?.let { fo ->
+            val m = fo.metrics?.toString() ?: ""
+            println("fragment ordering: ${fo.status} $m")
+            if (fo.status != TestStatus.UNSUPPORTED) {
+                assertTrue(m.contains("in_order"), "no per-ordering result: $m")
+                assertTrue(m.contains("reversed"), "reversed ordering was never attempted: $m")
+            }
+        }
+
         val train = assertNotNull(byType[TestType.TRAIN_UDP_DOWNSTREAM], "no downstream train")
         assertNotNull(train.evidence, "a train without columnar evidence is not recomputable")
         val received = train.metrics?.get("received")?.toString()?.toIntOrNull() ?: 0

echolot-app/core-engine/src/test/kotlin/app/echo_lot/engine/LiveThroughputTest.kt

@@ -0,0 +1,105 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package app.echo_lot.engine
+
+import app.echo_lot.measurement.TestStatus
+import app.echo_lot.protocol.ControlClient
+import app.echo_lot.protocol.ProbeSession
+import kotlin.test.Test
+import kotlin.test.assertEquals
+import kotlin.test.assertNotNull
+import kotlin.test.assertTrue
+
+/**
+ * Downstream throughput against a LIVE server. Self-skips without ECHOLOT_LIVE_*.
+ *
+ * The assertions are about *honesty* rather than speed: a rate is only a measurement if the run
+ * was ended by the clock and the sender's own count backs it up. A test that just asserted "some
+ * Mbps arrived" would pass equally well against a broken implementation.
+ */
+class LiveThroughputTest {
+
+    private val url = System.getenv("ECHOLOT_LIVE_URL")
+    private val pin = System.getenv("ECHOLOT_LIVE_PIN")
+    private val cred = System.getenv("ECHOLOT_LIVE_CRED")
+    private val udp = System.getenv("ECHOLOT_LIVE_UDP")
+    private val target = System.getenv("ECHOLOT_LIVE_TARGET") ?: "fmr"
+
+    @Test
+    fun measuresDownstreamRateAndSaysWhatLimitedIt() {
+        if (url == null || pin == null || cred == null || udp == null) {
+            println("LiveThroughputTest skipped (no ECHOLOT_LIVE_* env)"); return
+        }
+        val control = ControlClient(url, setOf(pin), "0.2.0")
+        val session = control.createSession(cred, target)
+        val (host, port) = udp.split(":").let { it[0] to it[1].toInt() }
+
+        val (test, findings) = ProbeSession(cred, session, host, port).use { ps ->
+            ps.echo() // prime: the grant binds to the observed source
+            ThroughputMeasurement(SystemIdSource()).run(
+                cred, session.sessionId, control, ps, sessionRef = "sess-1",
+                durationS = 3, kbps = 20_000,
+            )
+        }
+        control.deleteSession(cred, session.sessionId)
+
+        val m = assertNotNull(test.metrics).toString()
+        println("throughput: ${test.status} $m")
+        for (f in findings) println("finding ${f.code} [${f.severity}] ${f.title}")
+
+        assertEquals(TestStatus.OK, test.status, "no throughput traffic arrived: $m")
+
+        // The sender's own count must be present — without it, loss cannot be attributed and the
+        // number is not a measurement.
+        assertTrue(m.contains("sender_packets"), "no sender report to compare against: $m")
+        assertTrue(m.contains("limited_by"), "the result must say what ended the run: $m")
+
+        val received = Regex(""""received_kbps":(\d+)""").find(m)?.groupValues?.get(1)?.toInt()
+        assertNotNull(received)
+        assertTrue(received > 0, "measured 0 kbps: $m")
+        println("received ${received / 1000} Mbit/s")
+
+        // A run this short and this far below the ceiling should end on the clock. Anything else
+        // means the grant was the constraint, and then the rate says nothing about the path.
+        assertTrue(m.contains(""""limited_by":"duration""""),
+            "the run did not end on the clock, so the rate measures the server, not the path: $m")
+    }
+
+    // Upstream is the direction only the far end can measure. The assertion that matters is that
+    // the server's count is present and plausible against what we sent — a test that only checked
+    // "we transmitted some Mbps" would pass against a server that counted nothing at all.
+    @Test
+    fun measuresUpstreamAgainstTheServersCount() {
+        if (url == null || pin == null || cred == null || udp == null) {
+            println("LiveThroughputTest(up) skipped"); return
+        }
+        val control = ControlClient(url, setOf(pin), "0.2.0")
+        val session = control.createSession(cred, target)
+        val (host, port) = udp.split(":").let { it[0] to it[1].toInt() }
+
+        val (test, findings) = ProbeSession(cred, session, host, port).use { ps ->
+            ps.echo()
+            ThroughputMeasurement(SystemIdSource()).runUpstream(
+                cred, session.sessionId, control, ps, sessionRef = "sess-1",
+                durationS = 3, kbps = 10_000,
+            )
+        }
+        control.deleteSession(cred, session.sessionId)
+
+        val m = assertNotNull(test.metrics).toString()
+        println("upstream: ${test.status} $m")
+        for (f in findings) println("finding ${f.code} [${f.severity}] ${f.title}")
+
+        assertEquals(TestStatus.OK, test.status, "the server counted nothing: $m")
+        val recv = Regex(""""received_packets":(\d+)""").find(m)?.groupValues?.get(1)?.toInt()
+        val sent = Regex(""""sent_packets":(\d+)""").find(m)?.groupValues?.get(1)?.toInt()
+        assertNotNull(recv); assertNotNull(sent)
+        assertTrue(sent > 100, "barely anything was sent, so the rate means nothing: $m")
+        assertTrue(recv > 0, "the server received none of $sent packets: $m")
+        // The counts should be close on a healthy path; wildly different means the two sides are
+        // counting different things rather than the network losing packets.
+        assertTrue(recv <= sent, "the server counted MORE than we sent — the counter is not being reset")
+        println("sent $sent, server saw $recv")
+    }
+}

echolot-app/core-engine/src/test/kotlin/app/echo_lot/engine/LiveUpstreamTrainTest.kt

@@ -0,0 +1,65 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package app.echo_lot.engine
+
+import app.echo_lot.measurement.TestStatus
+import app.echo_lot.protocol.ControlClient
+import app.echo_lot.protocol.ProbeSession
+import kotlin.test.Test
+import kotlin.test.assertEquals
+import kotlin.test.assertNotNull
+import kotlin.test.assertTrue
+
+/**
+ * Upstream train (types 0x03-0x05) against a LIVE server. Self-skips without ECHOLOT_LIVE_*.
+ *
+ * What is asserted is the ledger property: the server's report must account for what was sent,
+ * per sequence number, because directional loss attribution is the entire reason trains exist —
+ * a test that only checked "a report came back" would pass against a server that counts nothing.
+ */
+class LiveUpstreamTrainTest {
+
+    private val url = System.getenv("ECHOLOT_LIVE_URL")
+    private val pin = System.getenv("ECHOLOT_LIVE_PIN")
+    private val cred = System.getenv("ECHOLOT_LIVE_CRED")
+    private val udp = System.getenv("ECHOLOT_LIVE_UDP")
+    private val target = System.getenv("ECHOLOT_LIVE_TARGET") ?: "fmr"
+
+    @Test
+    fun serverLedgerAccountsForTheTrain() {
+        if (url == null || pin == null || cred == null || udp == null) {
+            println("LiveUpstreamTrainTest skipped (no ECHOLOT_LIVE_* env)"); return
+        }
+        val control = ControlClient(url, setOf(pin), "0.2.0")
+        val session = control.createSession(cred, target)
+        val (host, port) = udp.split(":").let { it[0] to it[1].toInt() }
+
+        val (test, findings) = ProbeSession(cred, session, host, port).use { ps ->
+            ps.echo() // prime the session so its source is known
+            UpstreamTrainMeasurement(SystemIdSource()).run(
+                ps, sessionRef = "sess-1", count = 120, sizeBytes = 200, interPacketMs = 3,
+            )
+        }
+        control.deleteSession(cred, session.sessionId)
+
+        val m = assertNotNull(test.metrics).toString()
+        println("updown: ${test.status} $m")
+        for (f in findings) println("finding ${f.code} [${f.severity}] ${f.title}")
+
+        assertEquals(TestStatus.OK, test.status, "train report incomplete or absent: $m")
+
+        val sent = Regex(""""sent":(\d+)""").find(m)?.groupValues?.get(1)?.toInt()
+        val received = Regex(""""received_by_server":(\d+)""").find(m)?.groupValues?.get(1)?.toInt()
+        assertNotNull(sent); assertNotNull(received)
+        assertTrue(sent > 0, "nothing was sent: $m")
+        // Over a working path the ledger must be near-complete; a lossy wifi may drop a few, but
+        // a server that fails to count would show up as massive phantom loss here.
+        assertTrue(received >= sent * 9 / 10, "server counted $received of $sent: $m")
+
+        // The columnar evidence must carry a server timestamp for arrived packets — that column
+        // is what one-way delay math consumes after timesync.
+        val ev = assertNotNull(test.evidence).toString()
+        assertTrue(ev.contains("t_srv_rx_ns"), "no server rx column in evidence")
+    }
+}

echolot-app/core-measurement/src/main/kotlin/app/echo_lot/measurement/Document.kt

@@ -35,9 +35,37 @@ data class Run(
     val device: DeviceInfo,
     val tiers: Tiers,
     @SerialName("profiles_used") val profilesUsed: List<String> = emptyList(),
+    val constraints: Constraints = Constraints(),
     val notes: String? = null,
 )
 
+/**
+ * What limited this run — the counterpart to [Tiers], which records what was available.
+ *
+ * A constrained run is not a failed run, and it is not a normal one either. Without this, a run
+ * taken through a VPN looks exactly like a clean run of a healthy network: the same shape, the
+ * same green verdict, and no way for a reader — or a server aggregating thousands of these — to
+ * know that almost nothing was actually measured.
+ */
+@Serializable
+data class Constraints(
+    /** A VPN held the default route while this ran. */
+    @SerialName("vpn_active") val vpnActive: Boolean = false,
+    /**
+     * Per-network probing was refused by the OS.
+     *
+     * Android blocks `Network.bindSocket()` on the underlying networks whenever a VPN is up, to
+     * stop apps leaking around the tunnel. Every per-network test then measures nothing, so any
+     * conclusion drawn about the wifi or cellular link underneath is unfounded.
+     */
+    @SerialName("per_network_blocked") val perNetworkBlocked: Boolean = false,
+    /** Networks that could not be measured, by id. */
+    @SerialName("unmeasured_networks") val unmeasuredNetworks: List<String> = emptyList(),
+) {
+    /** True when this run's results mean something different from an unconstrained one. */
+    val constrained: Boolean get() = vpnActive || perNetworkBlocked
+}
+
 @Serializable
 enum class Trigger {
     @SerialName("manual") MANUAL,

echolot-app/core-measurement/src/main/kotlin/app/echo_lot/measurement/FindingRegistry.kt

@@ -0,0 +1,321 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package app.echo_lot.measurement
+
+/**
+ * The registry of finding codes (measurement-schema.md §9, open item 1).
+ *
+ * A finding code is the stable, machine-readable half of a result: the prose changes, the code is
+ * what a dashboard groups by and what someone greps a year of archived runs for. That only holds
+ * if a code means exactly one thing forever — which is not something ad-hoc string literals at
+ * fifteen call sites can promise.
+ *
+ * The failure this exists to prevent had already happened by the time it was written. Two
+ * independently-added emitters produced `connectivity.downstream_loss` and
+ * `connectivity.loss_downstream` for the same concept, and nothing anywhere objected. Anyone
+ * aggregating either one would have silently seen half their data.
+ *
+ * So codes are declared here as typed specs, each carrying its category and default severity, and
+ * emitters reference the spec rather than retyping the string. That makes a typo a compile error,
+ * and makes it impossible for two call sites to disagree about which category a finding belongs
+ * to — a disagreement that would otherwise split one fault across two verdict lights.
+ */
+data class FindingSpec(
+    val code: String,
+    val category: Category,
+    /** Severity when nothing about the specific run argues otherwise; emitters may escalate. */
+    val severity: Severity,
+    /** One line: what this finding asserts. Present tense, no hedging. */
+    val meaning: String,
+    /**
+     * What the finding rules *out*, where that is the useful half. "Loss upstream" is worth much
+     * more when it also says the return path is fine, because that halves where to look next.
+     */
+    val rulesOut: String? = null,
+)
+
+object FindingRegistry {
+
+    // ---- connectivity ----------------------------------------------------------------
+
+    // Renamed from nat.* before anything shipped: neither of these is about NAT, and the
+    // prefix is what decides which category - and therefore which verdict light - a finding
+    // rolls up into. A nat.* code landing under connectivity would be a permanent puzzle.
+    val UDP_UNREACHABLE = FindingSpec(
+        "connectivity.udp_unreachable", Category.CONNECTIVITY, Severity.HIGH,
+        "No UDP echo replies came back from the server at all.",
+    )
+
+    val UDP_UNREACHABLE_UPSTREAM = FindingSpec(
+        "connectivity.udp_unreachable_upstream", Category.CONNECTIVITY, Severity.HIGH,
+        "The server received none of the probes, so traffic is dropped on the way out.",
+        rulesOut = "The return path: nothing arrived to be replied to.",
+    )
+
+    val UDP_LOSS = FindingSpec(
+        "connectivity.udp_loss", Category.CONNECTIVITY, Severity.MEDIUM,
+        "A large fraction of round-trip probes were lost, direction unknown.",
+    )
+
+    val LOSS_UPSTREAM = FindingSpec(
+        "connectivity.loss_upstream", Category.CONNECTIVITY, Severity.MEDIUM,
+        "Probes were lost on the way to the server.",
+        rulesOut = "The return path: replies came back for everything that arrived.",
+    )
+
+    /**
+     * The single code for "lost on the return path", whichever measurement found it.
+     *
+     * Two emitters had independently invented `connectivity.downstream_loss` and
+     * `connectivity.loss_downstream` for this, and nothing objected. Anyone aggregating either
+     * one would have silently seen half their data. Paired with [LOSS_UPSTREAM] so the two
+     * directions read as a set.
+     */
+    val LOSS_DOWNSTREAM = FindingSpec(
+        "connectivity.loss_downstream", Category.CONNECTIVITY, Severity.MEDIUM,
+        "Packets were lost on the way back from the server.",
+        rulesOut = "The outbound path: the server received what it was answering.",
+    )
+
+    val DOWNSTREAM_BLOCKED = FindingSpec(
+        "connectivity.downstream_blocked", Category.CONNECTIVITY, Severity.HIGH,
+        "Server-initiated packets never arrive, although round trips work.",
+        rulesOut = "Basic reachability: the path forwards replies, just not unsolicited traffic.",
+    )
+
+    val DOWNSTREAM_REORDER = FindingSpec(
+        "connectivity.downstream_reorder", Category.CONNECTIVITY, Severity.LOW,
+        "Downstream packets arrive in a different order than they were sent.",
+    )
+
+    // MEDIUM, not HIGH: a captive portal is a condition to report, not necessarily a fault - on
+    // hotel or cafe wifi it is exactly what should be there, and logging in clears it. NO_INTERNET
+    // is the HIGH one, because nothing the user does locally fixes that. The registry first said
+    // HIGH; the probe emitting it had always said MEDIUM, and the probe was the considered value.
+    val CAPTIVE_PORTAL = FindingSpec(
+        "connectivity.captive_portal", Category.CONNECTIVITY, Severity.MEDIUM,
+        "A captive portal is intercepting connectivity checks.",
+    )
+
+    val NO_INTERNET = FindingSpec(
+        "connectivity.no_internet", Category.CONNECTIVITY, Severity.HIGH,
+        "Android's own connectivity checks fail on this network.",
+    )
+
+    // ---- mtu -------------------------------------------------------------------------
+
+    val MTU_REDUCED_DOWNSTREAM = FindingSpec(
+        "mtu.reduced_downstream", Category.MTU, Severity.LOW,
+        "The downstream path MTU is below the usual 1500 bytes.",
+    )
+
+    val MTU_DOWNSTREAM_BLACKHOLE = FindingSpec(
+        "mtu.downstream_blackhole", Category.MTU, Severity.MEDIUM,
+        "Datagrams above the path MTU are dropped downstream, fragmented or not.",
+    )
+
+    val FRAGMENTS_BLOCKED = FindingSpec(
+        "mtu.fragments_blocked", Category.MTU, Severity.MEDIUM,
+        "IP fragments do not reach this device even when sent in order.",
+    )
+
+    val FRAGMENT_REORDER_SENSITIVE = FindingSpec(
+        "mtu.fragment_reorder_sensitive", Category.MTU, Severity.LOW,
+        "Fragments are delivered in order but dropped when reordered or delayed.",
+        rulesOut = "Fragmentation itself: in-order fragments arrive fine.",
+    )
+
+    // ---- nat -------------------------------------------------------------------------
+
+    val NAT_UDP_REBINDING = FindingSpec(
+        "nat.udp_rebinding", Category.NAT, Severity.MEDIUM,
+        "A NAT remapped the UDP source port mid-flow.",
+    )
+
+    val NAT_SYMMETRIC = FindingSpec(
+        "nat.symmetric", Category.NAT, Severity.MEDIUM,
+        "The NAT assigns a different external port per destination.",
+    )
+
+    // ---- perf ------------------------------------------------------------------------
+
+    val THROUGHPUT_NO_DELIVERY = FindingSpec(
+        "perf.throughput_no_delivery", Category.PERFORMANCE, Severity.HIGH,
+        "No throughput traffic arrived, although the server sent it.",
+    )
+
+    val THROUGHPUT_BELOW_OFFERED = FindingSpec(
+        "perf.throughput_below_offered", Category.PERFORMANCE, Severity.LOW,
+        "Less throughput arrived than the server sent for the whole run.",
+    )
+
+    // ---- dns -------------------------------------------------------------------------
+
+    val DNS_ANSWER_REWRITTEN = FindingSpec(
+        "dns.answer_rewritten", Category.DNS, Severity.HIGH,
+        "A resolver returned an answer that differs from the authoritative record.",
+    )
+
+    val DNS_AUTHORITATIVE_UNREACHABLE = FindingSpec(
+        "dns.authoritative_unreachable", Category.DNS, Severity.MEDIUM,
+        "The canary zone's authoritative server could not be reached.",
+    )
+
+    // ---- v6 ----------------------------------------------------------------------------
+    //
+    // Prefix is `v6.`, matching the test-type registry (v6.brokenness, v6.happy_eyeballs, ...).
+    // These were `ipv6.*` while declaring Category.IPV6, but the prefix map only knows "v6", so
+    // they silently rolled up under connectivity: the third instance of a prefix disagreeing with
+    // its category and quietly moving a fault to a different verdict light.
+
+    /**
+     * Renamed from `v6.broken`, which claimed more than the evidence supports.
+     *
+     * The only signal behind it is ICMPv6 echo getting no reply — and ICMPv6 echo is widely
+     * filtered on networks where IPv6 otherwise works perfectly. A phone that reported this while
+     * happily loading an IPv6-only site over TCP is what caught it. From here the two cases look
+     * identical, so the finding now says what was observed and names both explanations rather than
+     * picking one.
+     *
+     * It is worth reporting either way: filtered ICMPv6 breaks Path MTU Discovery, which is its
+     * own fault even when IPv6 works.
+     */
+    /**
+     * A global IPv6 address with no default route.
+     *
+     * This is the structural version of the same complaint, and it is worth far more than the
+     * ICMP one because it admits no other explanation: the device has an address it cannot route
+     * with. Nothing is filtered, nothing is inferred — the routing table says so directly, and it
+     * is already in the link snapshot.
+     *
+     * Not always a fault. A VPN that installs host routes to specific destinations produces
+     * exactly this shape on purpose, and it works. What makes it worth reporting either way is
+     * that applications cannot tell: having a global address, they will try IPv6 first and stall
+     * for every destination the routes do not cover.
+     */
+    /**
+     * An IPv6 default route with no global address to use it from — the mirror of
+     * [V6_NO_DEFAULT_ROUTE], and the more common misconfiguration of the two.
+     *
+     * The router is sending RAs that name it as a default gateway, but SLAAC produced no address:
+     * no prefix information option, or a prefix without the autonomous flag, or DHCPv6-only
+     * addressing the device did not complete. The network is announcing IPv6 service it does not
+     * actually deliver.
+     *
+     * This is worth flagging above the ICMP signal because it is both certain and consequential.
+     * Hosts see router advertisements, believe IPv6 is available, and pay a connection-attempt
+     * timeout on every dual-stack destination before falling back to IPv4 — the classic "the
+     * internet feels slow" complaint with no packet loss anywhere to explain it.
+     */
+    val V6_ROUTE_WITHOUT_ADDRESS = FindingSpec(
+        "v6.route_without_address", Category.IPV6, Severity.MEDIUM,
+        "The network advertises an IPv6 default route but the device has no global IPv6 address.",
+        rulesOut = "A working IPv6 setup: SLAAC did not produce a usable address on this link.",
+    )
+
+    /**
+     * A VPN prevented the underlying networks from being measured.
+     *
+     * Reported rather than worked around: Android refuses `Network.bindSocket()` on the networks
+     * beneath a VPN precisely so apps cannot leak around the tunnel, and that is correct
+     * behaviour. What is not acceptable is a run that quietly measures nothing and calls the
+     * result healthy, so this says plainly which networks went unmeasured and why.
+     */
+    /**
+     * The network's DNS server answers, but this device cannot resolve through it.
+     *
+     * Worth separating from every other DNS failure because the remedy is somewhere else entirely.
+     * A name that will not resolve looks identical to a user whatever the cause, and the two causes
+     * pull in opposite directions: a server that does not answer means the network is broken and
+     * the router is the thing to examine, while a server that answers a direct query on a device
+     * that still cannot resolve means the platform resolver has wedged — fixed by toggling wifi,
+     * and nothing to do with the network at all.
+     *
+     * Proven rather than inferred: the probe sends its own UDP query, bypassing the component under
+     * suspicion, and compares that against what the platform returns for the same name.
+     */
+    /**
+     * The network hands out a search domain its DNS server will not answer for.
+     *
+     * A resolver appends search domains to lookups, so every name a client asks about can stall on
+     * a domain the server ignores. The failure mode is silence rather than a negative answer, and
+     * silence is indistinguishable from packet loss: clients retry instead of moving on, and some
+     * give up on the lookup entirely. That makes it look like the device is broken when the
+     * network is.
+     *
+     * Whether it bites depends on the resolver — some try the bare name first and never notice —
+     * which is why two devices on the same network can disagree about whether DNS works.
+     */
+    val DNS_SEARCH_DOMAIN_UNANSWERED = FindingSpec(
+        "dns.search_domain_unanswered", Category.DNS, Severity.HIGH,
+        "The network advertises a DNS search domain that its own server does not answer for.",
+        rulesOut = "A fault on this device: the same server answers ordinary names normally.",
+    )
+
+    val DNS_SYSTEM_RESOLVER_BROKEN = FindingSpec(
+        "dns.system_resolver_broken", Category.DNS, Severity.HIGH,
+        "The network's DNS server answers, but this device cannot resolve names through it.",
+        rulesOut = "A network fault: the server replied to a query sent from this device.",
+    )
+
+    val MEASUREMENT_VPN_CONSTRAINED = FindingSpec(
+        "measurement.vpn_constrained", Category.CONNECTIVITY, Severity.INFO,
+        "A VPN was active, so the networks underneath it could not be measured.",
+        rulesOut = "Nothing — this run says little about the underlying network either way.",
+    )
+
+    val V6_NO_DEFAULT_ROUTE = FindingSpec(
+        "v6.no_default_route", Category.IPV6, Severity.MEDIUM,
+        "The device has a global IPv6 address but no IPv6 default route.",
+        rulesOut = "Guesswork: this is read from the routing table, not inferred from silence.",
+    )
+
+    val V6_NO_ICMP_REPLY = FindingSpec(
+        "v6.no_icmp_reply", Category.IPV6, Severity.LOW,
+        "IPv6 is configured but ICMPv6 echo gets no reply.",
+        rulesOut = "Nothing on its own: IPv6 may work fine with ICMP filtered.",
+    )
+
+    /**
+     * IPv6 is advertised and does not work — the claim `v6.broken` originally made on ICMP
+     * silence alone, now reinstated because it can finally be backed: it is only emitted when a
+     * real IPv6 TCP connection (v6.brokenness) failed on the same network whose ICMPv6 went
+     * unanswered. Two independent transports failing on a network that advertises IPv6 is what
+     * "broken" actually means; either signal alone still gets [V6_NO_ICMP_REPLY].
+     */
+    val V6_BROKEN = FindingSpec(
+        "v6.broken", Category.IPV6, Severity.HIGH,
+        "IPv6 is advertised on this network but carries no traffic.",
+        rulesOut = "ICMP filtering as the benign explanation: a TCP connection over IPv6 failed too.",
+    )
+
+    /**
+     * INFO deliberately, and it needs to stay that way.
+     *
+     * Most networks still do not offer IPv6, and that is not a fault. Reporting it as a warning
+     * lights a yellow verdict on a perfectly healthy network, which teaches people to ignore the
+     * light — the one thing a diagnostic must never do.
+     */
+    val V6_NOT_OFFERED = FindingSpec(
+        "v6.not_offered", Category.IPV6, Severity.INFO,
+        "This network does not offer IPv6.",
+    )
+
+    /** Every registered finding, in declaration order. */
+    val all: List<FindingSpec> = listOf(
+        UDP_UNREACHABLE, UDP_UNREACHABLE_UPSTREAM, UDP_LOSS, LOSS_UPSTREAM, LOSS_DOWNSTREAM,
+        DOWNSTREAM_BLOCKED, DOWNSTREAM_REORDER, CAPTIVE_PORTAL, NO_INTERNET,
+        MTU_REDUCED_DOWNSTREAM, MTU_DOWNSTREAM_BLACKHOLE, FRAGMENTS_BLOCKED,
+        FRAGMENT_REORDER_SENSITIVE,
+        NAT_UDP_REBINDING, NAT_SYMMETRIC,
+        THROUGHPUT_NO_DELIVERY, THROUGHPUT_BELOW_OFFERED,
+        DNS_ANSWER_REWRITTEN, DNS_AUTHORITATIVE_UNREACHABLE,
+        DNS_SEARCH_DOMAIN_UNANSWERED, DNS_SYSTEM_RESOLVER_BROKEN, MEASUREMENT_VPN_CONSTRAINED,
+        V6_NO_DEFAULT_ROUTE, V6_ROUTE_WITHOUT_ADDRESS, V6_NO_ICMP_REPLY, V6_BROKEN, V6_NOT_OFFERED,
+    )
+
+    private val byCode: Map<String, FindingSpec> = all.associateBy { it.code }
+
+    fun byCode(code: String): FindingSpec? = byCode[code]
+}

echolot-app/core-measurement/src/main/kotlin/app/echo_lot/measurement/Network.kt

@@ -18,6 +18,28 @@ data class Network(
     val wifi: Wifi? = null,
     val cellular: Cellular? = null,
     val changes: List<NetworkChange> = emptyList(),
+    @SerialName("system_verdict") val systemVerdict: SystemVerdict? = null,
+)
+
+/**
+ * What Android itself concluded about a network, as opposed to what we measured.
+ *
+ * Recorded because it is the verdict the user can see — the "no internet" warning in the status
+ * bar — and because it is free: the platform has already done the work by the time a run starts.
+ *
+ * Its real value is disagreement. When Android says a network is unusable and our own probes reach
+ * the internet regardless, the fault is in the device rather than the network, and that distinction
+ * is the difference between "fix your router" and "toggle your wifi". Neither number alone can say
+ * that; only the two together.
+ */
+@Serializable
+data class SystemVerdict(
+    /** Android's own connectivity check passed. Null when the platform did not say. */
+    val validated: Boolean? = null,
+    /** Android believes a captive portal is intercepting this network. */
+    @SerialName("captive_portal") val captivePortal: Boolean? = null,
+    /** Some traffic works and some does not — Android's own hedge. */
+    @SerialName("partial_connectivity") val partialConnectivity: Boolean? = null,
 )
 
 @Serializable

echolot-app/core-measurement/src/main/kotlin/app/echo_lot/measurement/Summary.kt

@@ -35,6 +35,9 @@ data class CategorySummary(
  *   (critical|high → red, medium|low → yellow, info/none → green).
  * - A category is `inconclusive` when > 50% of its tests are failed/unsupported.
  * - Overall = the worst category light; `inconclusive` only when ALL categories are.
+ * - A run whose per-network probing was blocked is `inconclusive` outright, whatever the
+ *   categories say. The lights describe what the tests found; when the OS refused to let the
+ *   tests run, a green light would describe nothing at all.
  *
  * The mapping test-type → category comes from [TestType.category]. Only categories that have
  * findings or tests appear in the summary.
@@ -44,7 +47,10 @@ object Verdicts {
     private fun isInconclusiveTest(s: TestStatus) =
         s == TestStatus.FAILED || s == TestStatus.UNSUPPORTED
 
-    fun derive(tests: List<Test>, findings: List<Finding>): Summary {
+    fun derive(tests: List<Test>, findings: List<Finding>): Summary =
+        derive(tests, findings, Constraints())
+
+    fun derive(tests: List<Test>, findings: List<Finding>, constraints: Constraints): Summary {
         val testsByCat = tests.groupBy { TestType.category(it.type) }
         val findingsByCat = findings.groupBy { it.category }
         val categories = (testsByCat.keys + findingsByCat.keys)
@@ -72,7 +78,14 @@ object Verdicts {
             )
         }
 
-        val overall = deriveOverall(perCat.values)
+        // A run that could not measure the networks it was asked about has not found them
+        // healthy; it has found out nothing. Reporting that as green is the single most
+        // misleading thing this function could do, so the constraint outranks the lights.
+        val overall = if (constraints.perNetworkBlocked) {
+            Verdict.INCONCLUSIVE
+        } else {
+            deriveOverall(perCat.values)
+        }
         return Summary(overall = overall, categories = perCat)
     }
 

echolot-app/core-measurement/src/main/kotlin/app/echo_lot/measurement/Test.kt

@@ -89,6 +89,13 @@ object TestType {
     // dns
     const val DNS_RESOLVER_INVENTORY = "dns.resolver_inventory"
     const val DNS_CANARY = "dns.canary"
+    /**
+     * Does this device's own resolver work, as distinct from the network's DNS.
+     *
+     * Registry addition, v1.1. Kept apart from [DNS_CANARY], which asks whether answers are being
+     * tampered with; this asks whether answers arrive at all, and where the failure sits.
+     */
+    const val DNS_RESOLVER = "dns.resolver"
     const val DNS_INTERCEPTION = "dns.interception"
     const val DNS_TTL_INTEGRITY = "dns.ttl_integrity"
     const val DNS_ANSWER_INTEGRITY = "dns.answer_integrity"

echolot-app/core-measurement/src/main/kotlin/app/echo_lot/measurement/V6Analysis.kt

@@ -0,0 +1,73 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package app.echo_lot.measurement
+
+/**
+ * The two ways a network can be half-configured for IPv6, read from the link snapshot.
+ *
+ * Pure model logic rather than something a ViewModel does, because "is this network's IPv6
+ * broken, and in which direction" is exactly the kind of judgement that should be checkable
+ * against a captured routing table without a phone in the loop.
+ */
+object V6Analysis {
+
+    /** Linux tunnel interfaces: WireGuard/Netbird (tun*, wg*), plus the usual VPN names. */
+    private val TUNNEL_IFACE = Regex("""^(tun|tap|wg|ppp|ipsec|utun)\d*$""")
+
+    /** What one network's IPv6 configuration looks like. */
+    data class Shape(
+        val iface: String,
+        /** A global address with no ::/0 route: an address the device cannot route with. */
+        val addressWithoutRoute: Boolean,
+        /** A ::/0 route with no global address: a route the device cannot source from. */
+        val routeWithoutAddress: Boolean,
+        /** The routes belong to a tunnel, so a partial view of IPv6 is likely deliberate. */
+        val tunnel: Boolean,
+    )
+
+    /**
+     * Classifies each network's IPv6 configuration.
+     *
+     * Both shapes are read straight from the link snapshot rather than inferred from silence, so
+     * unlike an ICMP signal there is no competing explanation for what was observed — and both
+     * matter for the same reason: an application cannot tell in advance, so it tries IPv6 first
+     * and waits.
+     *
+     * They differ in what they mean. An address with no route is what a VPN installing host routes
+     * to specific destinations produces on purpose, and it works; calling that a fault would be the
+     * "lack of IPv6 is a yellow condition" mistake in a new costume, so a tunnel downgrades it to
+     * information. A route with no address is the opposite: the router advertised itself as a
+     * default gateway but SLAAC produced nothing usable, so the network is announcing IPv6 service
+     * it does not deliver. That one is a real misconfiguration however it arises.
+     */
+    fun classify(networks: List<Network>): List<Shape> = networks.map { n ->
+        val globalV6 = n.link.addresses.any { isGlobalV6(it.addr) }
+        val v6Routes = n.link.routes.filter { it.dst.contains(':') }
+        val hasDefault = v6Routes.any { it.dst == "::/0" }
+        Shape(
+            iface = n.iface ?: v6Routes.firstOrNull()?.iface.orEmpty(),
+            addressWithoutRoute = globalV6 && !hasDefault,
+            routeWithoutAddress = hasDefault && !globalV6,
+            // Android labels the transport itself, which beats guessing from a name; the regex
+            // stays as a backstop for tunnels Android does not own (a userspace WireGuard, say,
+            // or anything seen through the shell tier).
+            tunnel = n.transport == Transport.VPN ||
+                v6Routes.any { TUNNEL_IFACE.containsMatchIn(it.iface.orEmpty()) },
+        )
+    }
+
+    /**
+     * Whether an address is IPv6 and usable as a source for off-link traffic.
+     *
+     * ULAs count. A ULA is not globally routable, but it is a global-*scope* address the stack
+     * will happily select as a source, which is the property that matters here — an overlay
+     * network handing out fc00::/7 addresses is providing working IPv6 to the destinations it
+     * carries, and treating that as "no address" would misreport every VPN as broken.
+     */
+    private fun isGlobalV6(addr: String): Boolean {
+        if (!addr.contains(':')) return false
+        val a = addr.substringBefore('%').lowercase() // strip any zone index
+        return !a.startsWith("fe80") && a != "::1" && a != "::"
+    }
+}

echolot-app/core-measurement/src/test/kotlin/app/echo_lot/measurement/FindingRegistryTest.kt

@@ -0,0 +1,133 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package app.echo_lot.measurement
+
+import java.io.File
+import kotlin.test.Test
+import kotlin.test.assertEquals
+import kotlin.test.assertTrue
+import kotlin.test.fail
+
+/**
+ * Keeps the finding registry honest.
+ *
+ * The interesting test is the last one: it reads `docs/findings-registry.md` and fails when the
+ * document and the code disagree. Documentation that drifts from its implementation is worse than
+ * none, because it still looks authoritative — and a finding registry is precisely the artifact
+ * other people build tooling against.
+ */
+class FindingRegistryTest {
+
+    @Test
+    fun codesAreUnique() {
+        val dupes = FindingRegistry.all.groupBy { it.code }.filterValues { it.size > 1 }.keys
+        assertTrue(dupes.isEmpty(), "duplicate finding codes: $dupes")
+    }
+
+    @Test
+    fun everyDeclaredSpecIsInTheAllList() {
+        // Reflection over the object's properties: a spec that is declared but left out of `all`
+        // is invisible to the doc check and to any consumer enumerating the registry.
+        val declared = FindingRegistry::class.java.declaredMethods
+            .filter { it.parameterCount == 0 && it.returnType == FindingSpec::class.java }
+            .mapNotNull { runCatching { it.invoke(FindingRegistry) as FindingSpec }.getOrNull() }
+            .map { it.code }
+            .toSet()
+        val listed = FindingRegistry.all.map { it.code }.toSet()
+        assertEquals(declared, listed, "declared specs and the `all` list disagree")
+    }
+
+    // The prefix decides the category, and the category decides which verdict light the finding
+    // rolls up into. A code whose prefix disagrees with its category silently moves a fault to a
+    // different light — the exact bug that got two codes renamed out of nat.*.
+    @Test
+    fun everyPrefixMatchesItsCategory() {
+        for (spec in FindingRegistry.all) {
+            val fromPrefix = TestType.category(spec.code)
+            assertEquals(
+                fromPrefix, spec.category,
+                "${spec.code} is declared as ${spec.category} but its prefix maps to $fromPrefix",
+            )
+        }
+    }
+
+    @Test
+    fun codesFollowTheNamingConvention() {
+        val shape = Regex("^[a-z0-9]+\\.[a-z0-9_]+$")
+        for (spec in FindingRegistry.all) {
+            assertTrue(shape.matches(spec.code), "malformed code: ${spec.code}")
+            assertTrue(spec.meaning.isNotBlank(), "${spec.code} has no meaning")
+            assertTrue(
+                spec.meaning.trimEnd().endsWith("."),
+                "${spec.code}'s meaning should be a sentence: '${spec.meaning}'",
+            )
+        }
+    }
+
+    // Two near-identical codes are how one fault ends up split across two dashboards. This is a
+    // blunt check — it will not catch every synonym — but it catches the shape that already
+    // happened: the same words in a different order.
+    @Test
+    fun noTwoCodesAreAnagramsOfEachOther() {
+        val normalised = FindingRegistry.all.associate { spec ->
+            spec.code to spec.code.substringAfter('.').split('_').sorted().joinToString("_")
+        }
+        val clashes = normalised.entries.groupBy { it.value }.filterValues { it.size > 1 }
+        if (clashes.isNotEmpty()) {
+            fail("codes differing only in word order: ${clashes.values.map { g -> g.map { it.key } }}")
+        }
+    }
+
+    @Test
+    fun theDocumentAndTheRegistryAgree() {
+        val doc = findDoc() ?: run {
+            println("findings-registry.md not found from ${File(".").absolutePath} — skipping")
+            return
+        }
+        val text = doc.readText()
+
+        // Only table rows count as "documented". Prose may legitimately mention a code that no
+        // longer exists — the rules section explains why two were merged — and treating that as
+        // a registry entry would force the document to forget its own history.
+        val documented = text.lines()
+            .filter { it.trimStart().startsWith("|") }
+            .flatMap { row -> Regex("`([a-z0-9]+\\.[a-z0-9_]+)`").findAll(row).map { it.groupValues[1] } }
+            .toSet()
+        val registered = FindingRegistry.all.map { it.code }.toSet()
+
+        val missingFromDoc = registered - documented
+        val missingFromCode = documented - registered
+        assertTrue(
+            missingFromDoc.isEmpty(),
+            "these codes exist in FindingRegistry but not in docs/findings-registry.md: $missingFromDoc",
+        )
+        assertTrue(
+            missingFromCode.isEmpty(),
+            "docs/findings-registry.md documents codes that no longer exist: $missingFromCode",
+        )
+
+        // And the severities must match, or the document is describing a different system.
+        for (spec in FindingRegistry.all) {
+            val row = text.lines().firstOrNull {
+                it.trimStart().startsWith("|") && it.contains("`${spec.code}`")
+            } ?: continue
+            val severity = spec.severity.name.lowercase()
+            assertTrue(
+                row.contains("| $severity |"),
+                "${spec.code} is ${severity} in code but the doc row says otherwise: $row",
+            )
+        }
+    }
+
+    /** Walks up from the test's working directory to find the repo's docs/ folder. */
+    private fun findDoc(): File? {
+        var dir: File? = File(".").absoluteFile
+        repeat(6) {
+            val candidate = File(dir, "docs/findings-registry.md")
+            if (candidate.isFile) return candidate
+            dir = dir?.parentFile
+        }
+        return null
+    }
+}

echolot-app/core-measurement/src/test/kotlin/app/echo_lot/measurement/V6AnalysisTest.kt

@@ -0,0 +1,125 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package app.echo_lot.measurement
+
+import kotlin.test.Test
+import kotlin.test.assertEquals
+import kotlin.test.assertFalse
+import kotlin.test.assertTrue
+
+/**
+ * The fixtures here are a real device's routing table, transcribed from `dumpsys connectivity`
+ * on a OnePlus 15 with a Netbird tunnel up: wifi advertising a default route it cannot source
+ * from, cellular working properly, and a VPN carrying host routes to two destinations.
+ *
+ * Using a captured table rather than invented ones matters, because the bug this guards against
+ * is not "the boolean logic is wrong" — it is "the shapes I imagined are not the shapes real
+ * networks produce".
+ */
+class V6AnalysisTest {
+
+    private fun net(
+        id: String,
+        transport: Transport,
+        iface: String,
+        addrs: List<String>,
+        routes: List<Pair<String, String>>,
+    ) = Network(
+        id = id,
+        transport = transport,
+        iface = iface,
+        link = Link(
+            addresses = addrs.map { Address(addr = it.substringBefore('/'), prefixLen = 64) },
+            routes = routes.map { (dst, dev) -> Route(dst = dst, iface = dev) },
+        ),
+    )
+
+    /** wlan0: an IPv6 default route via a link-local gateway, but SLAAC produced no address. */
+    private val wifi = net(
+        "w", Transport.WIFI, "wlan0",
+        addrs = listOf("fe80::bcf6:edff:fe67:b139", "10.13.102.122"),
+        routes = listOf(
+            "fe80::/64" to "wlan0",
+            "::/0" to "wlan0",
+            "0.0.0.0/0" to "wlan0",
+        ),
+    )
+
+    /** rmnet_data1: a properly configured cellular link — global address and a default route. */
+    private val cellular = net(
+        "c", Transport.CELLULAR, "rmnet_data1",
+        addrs = listOf("2001:4bb8:46a:e724:289d:87ff:feb6:ebd3"),
+        routes = listOf("::/0" to "rmnet_data1", "2001:4bb8:46a:e724::/64" to "rmnet_data1"),
+    )
+
+    /** tun1: Netbird, with a ULA and host routes to exactly two destinations. */
+    private val vpn = net(
+        "v", Transport.VPN, "tun1",
+        addrs = listOf("100.64.158.131", "fdfd:c4fe:c4fe:c4fe:1f3c:98a0:dd66:ac7"),
+        routes = listOf(
+            "2001:1ad0:c4fe:6767::2/128" to "tun1",
+            "2001:1ad0:c4fe:a::136/128" to "tun1",
+            "fdfd:c4fe:c4fe:c4fe::/64" to "tun1",
+        ),
+    )
+
+    @Test
+    fun `wifi advertising a route it cannot source from is reported`() {
+        val s = V6Analysis.classify(listOf(wifi)).single()
+        assertTrue(s.routeWithoutAddress, "::/0 with only a link-local address is the RA-without-SLAAC case")
+        assertFalse(s.addressWithoutRoute)
+        assertFalse(s.tunnel, "wifi is not a tunnel")
+        assertEquals("wlan0", s.iface)
+    }
+
+    @Test
+    fun `a properly configured link produces no finding`() {
+        val s = V6Analysis.classify(listOf(cellular)).single()
+        assertFalse(s.routeWithoutAddress)
+        assertFalse(s.addressWithoutRoute)
+    }
+
+    @Test
+    fun `a tunnel with host routes is deliberate, not broken`() {
+        val s = V6Analysis.classify(listOf(vpn)).single()
+        assertTrue(s.addressWithoutRoute, "a ULA and no ::/0 is an address with nothing to route it")
+        assertTrue(s.tunnel, "so it must be reported as information, not as a fault")
+        assertFalse(s.routeWithoutAddress)
+    }
+
+    @Test
+    fun `each network is judged on its own`() {
+        // The whole point of per-network classification: "IPv6 is broken" is useless advice when
+        // wifi is the broken one and cellular is fine.
+        val shapes = V6Analysis.classify(listOf(wifi, cellular, vpn)).associateBy { it.iface }
+        assertTrue(shapes.getValue("wlan0").routeWithoutAddress)
+        assertFalse(shapes.getValue("rmnet_data1").routeWithoutAddress)
+        assertFalse(shapes.getValue("rmnet_data1").addressWithoutRoute)
+        assertTrue(shapes.getValue("tun1").addressWithoutRoute)
+    }
+
+    @Test
+    fun `a link-local-only network with no v6 route says nothing either way`() {
+        // Plain IPv4-only wifi: no IPv6 offered at all. That is v6.not_offered's business, and
+        // reporting it here as well would double up on a network that is merely legacy, not broken.
+        val v4only = net(
+            "4", Transport.WIFI, "wlan0",
+            addrs = listOf("fe80::1", "192.168.1.5"),
+            routes = listOf("0.0.0.0/0" to "wlan0"),
+        )
+        val s = V6Analysis.classify(listOf(v4only)).single()
+        assertFalse(s.routeWithoutAddress)
+        assertFalse(s.addressWithoutRoute)
+    }
+
+    @Test
+    fun `a zone index does not hide a link-local address`() {
+        val zoned = net(
+            "z", Transport.WIFI, "wlan0",
+            addrs = listOf("fe80::1%wlan0"),
+            routes = listOf("::/0" to "wlan0"),
+        )
+        assertTrue(V6Analysis.classify(listOf(zoned)).single().routeWithoutAddress)
+    }
+}

echolot-app/core-privacy/build.gradle.kts

@@ -20,4 +20,8 @@ kotlin {
 }
 java { sourceCompatibility = JavaVersion.VERSION_17; targetCompatibility = JavaVersion.VERSION_17 }
 
-tasks.test { useJUnitPlatform() }
+tasks.test {
+    useJUnitPlatform()
+    // Opt-in: point this at a captured run to check the anonymizer against real data.
+    System.getenv("ECHOLOT_REAL_RUN")?.let { environment("ECHOLOT_REAL_RUN", it) }
+}

echolot-app/core-privacy/src/main/kotlin/app/echo_lot/privacy/Anonymizer.kt

@@ -108,12 +108,22 @@ class Anonymizer(private val level: PrivacyLevel, private val salt: Salt) {
         is JsonObject -> walkObject(v, path)
         is JsonArray -> JsonArray(v.map { walk(key, it, path) })
         is JsonPrimitive ->
-            if (v.isString) transform(Classification.typeOf(key, path), v.content).let(::JsonPrimitive)
-            else v
+            if (v.isString) {
+                // Name first (it is precise), then shape (it is exhaustive). A field nobody
+                // classified must not be a field that leaks.
+                val type = Classification.typeOf(key, path) ?: Classification.inferFromValue(v.content)
+                JsonPrimitive(transform(type, v.content))
+            } else {
+                v
+            }
     }
 
     private fun transform(type: LogicalType?, value: String): String = when (type) {
-        null -> value
+        // Unclassified strings still get their *embedded* identifiers scrubbed. A whole-value
+        // check cannot see them: raw shell output is one long string that is neither a MAC nor an
+        // address, so it sailed through both the name table and the shape check carrying every
+        // MAC on the user's LAN.
+        null -> scrubEmbedded(value)
         LogicalType.SSID -> pseudo("ssid", value) { "net-" + it.take(6) }
         LogicalType.MAC, LogicalType.BSSID -> macPreservingOui(value)
         LogicalType.IP4 -> ip4(value)
@@ -123,6 +133,38 @@ class Anonymizer(private val level: PrivacyLevel, private val salt: Salt) {
         LogicalType.FREETEXT -> "[removed: may contain identifying text]"
     }
 
+    /**
+     * Replaces addresses and MACs found *inside* a longer string.
+     *
+     * Shizuku probes embed raw command output verbatim — `ip neigh`, `ip route`, `dumpsys` — which
+     * is genuinely valuable evidence and also a complete inventory of every device on the user's
+     * network, with hardware addresses. measurement-schema.md §9 flagged these as "hard to
+     * anonymize" and proposed dropping them from exports.
+     *
+     * Scrubbing beats dropping: the output stays readable and auditable — you can still see the
+     * shape of the neighbour table and how many hosts there were — while the identifiers become
+     * the same pseudonyms used everywhere else in the document. So a MAC appearing both in a
+     * parsed field and in a raw dump still reads as one device.
+     *
+     * Only addresses and MACs are touched, for the same reason as [Classification.inferFromValue]:
+     * they are the patterns that cannot be mistaken for something else in free text.
+     */
+    private fun scrubEmbedded(value: String): String {
+        // Cheap bail-out: the overwhelming majority of strings are short and contain neither.
+        if (value.length < 7 || (!value.contains(':') && !value.contains('.'))) return value
+        // One pass, not three. Sequential passes re-process their own output: after a MAC became
+        // 78:9a:18:xx:yy:zz the IPv6 pattern matched it — six hex groups separated by colons is
+        // exactly an address — and mangled the vendor prefix that the MAC rule had just taken
+        // care to preserve. Ordered alternation resolves each position once, MAC first.
+        return EMBEDDED.replace(value) { m ->
+            when {
+                m.groups[1] != null -> macPreservingOui(m.value)
+                m.groups[2] != null -> ip6(m.value)
+                else -> ip4(m.value)
+            }
+        }
+    }
+
     // ---- per-type transforms -------------------------------------------------------------
 
     /**
@@ -147,6 +189,11 @@ class Anonymizer(private val level: PrivacyLevel, private val salt: Salt) {
      * Public addresses keep only their /16 so the network is still locatable at ISP granularity.
      */
     private fun ip4(value: String): String {
+        // A route destination carries a prefix length; pseudonymize the address and put it back,
+        // or "0.0.0.0/0" turns into nonsense and the routing table becomes unreadable.
+        value.substringAfter('/', "").takeIf { it.isNotEmpty() && value.contains('/') }?.let { len ->
+            return ip4(value.substringBefore('/')) + "/" + len
+        }
         val o = value.split(".")
         if (o.size != 4 || o.any { it.toIntOrNull() == null }) return value
         val n = o.map { it.toInt() }
@@ -167,8 +214,36 @@ class Anonymizer(private val level: PrivacyLevel, private val salt: Salt) {
      * is a device fingerprint, especially with EUI-64.
      */
     private fun ip6(value: String): String {
+        // Dotted quads reach here through the family-agnostic field names (addr, gateway, dst);
+        // hand them to the IPv4 path rather than mangling them as if they were v6.
+        if (value.count { it == ':' } < 2) return ip4(value)
+        if (value.contains('/')) {
+            return ip6(value.substringBefore('/')) + "/" + value.substringAfter('/')
+        }
         val v = value.lowercase(Locale.ROOT)
+        // The unspecified address and the default route are not identities; mangling them would
+        // make a routing table unreadable for no privacy gain.
         if (v == "::1" || v == "::" || v.startsWith("fe80:") || v.startsWith("ff")) return v
+
+        // Unique local addresses (fc00::/7) need the *whole* prefix replaced, not the tail.
+        //
+        // They look like the v6 equivalent of RFC1918, and the first instinct is to keep them for
+        // the same reason: private, topological, says nothing about anyone. That reasoning does
+        // not carry over. An RFC1918 prefix is shared by millions of networks and identifies
+        // none of them; a ULA global ID is 40 *random* bits, unique to one network by
+        // construction (RFC 4193). It is a network fingerprint. Passing the leading groups
+        // through - which is what the general path does - leaked 32 of those 40 bits.
+        //
+        // The prefix is pseudonymized as a unit, so two addresses on the same ULA subnet still
+        // land on the same pseudonymous prefix. "These hosts are on one network" survives;
+        // "this is *that* network" does not.
+        if (v.startsWith("fc") || v.startsWith("fd")) {
+            val groups = v.substringBefore('%').split(":")
+            val prefix = pseudo("ula-prefix", groups.take(3).joinToString(":")) { it }
+            val host = pseudo("ula-host", v) { it }
+            return "fd${prefix.substring(0, 2)}:${prefix.substring(2, 6)}:${prefix.substring(6, 10)}" +
+                "::${host.substring(0, 4)}"
+        }
         val groups = v.substringBefore('%').split(":")
         if (groups.size < 3) return v
         val h = pseudo("ip6", value) { it }
@@ -219,7 +294,7 @@ class Anonymizer(private val level: PrivacyLevel, private val salt: Salt) {
 
     /** Deterministic per (domain, value, salt); memoized so one value maps to one pseudonym. */
     private fun pseudo(domain: String, value: String, shape: (String) -> String): String =
-        cache.getOrPut("$domain$value") {
+        cache.getOrPut("$domain\u0000$value") {
             val md = MessageDigest.getInstance("SHA-256")
             md.update(salt.bytes)
             md.update(domain.toByteArray())
@@ -229,6 +304,21 @@ class Anonymizer(private val level: PrivacyLevel, private val salt: Salt) {
         }
 
     private companion object {
+        /**
+         * MAC | IPv6 | IPv4, in that order — alternation is ordered, so a MAC-shaped token is
+         * claimed by the MAC rule before the IPv6 rule can see it.
+         *
+         * The patterns are deliberately conservative. A missed address is scrubbed by another
+         * rule or not at all; an over-eager one mangles timestamps, version strings and log
+         * prefixes, corrupting evidence to protect nothing.
+         */
+        val EMBEDDED = Regex(
+            // Raw strings: a regex written with escaped escapes is a regex nobody can check.
+            """(\b[0-9a-fA-F]{2}(?:[:-][0-9a-fA-F]{2}){5}\b)""" +
+                """|(\b(?:[0-9a-fA-F]{1,4}:){2,7}(?::|[0-9a-fA-F]{1,4})(?:[0-9a-fA-F:]*))""" +
+                """|(\b(?:\d{1,3}\.){3}\d{1,3}\b)"""
+        )
+
         val publicSuffixes = setOf(
             "local", "lan", "home", "internal", "arpa",
             "com", "net", "org", "io", "app", "dev", "at", "de", "eu", "uk",

echolot-app/core-privacy/src/main/kotlin/app/echo_lot/privacy/Classification.kt

@@ -32,6 +32,16 @@ object Classification {
             "link_local", "ra_source", "prefix",
         ).forEach { put(it, LogicalType.IP6) }
 
+        // Family-agnostic address fields — the names the models actually use (Address.addr,
+        // Route.gateway, Route.dst, DnsConfig.servers). Their absence here was a real leak: the
+        // device's own global IPv6 address went out verbatim at the level whose description
+        // promises addresses are pseudonymized. Typed IP6 because the transform detects the
+        // family from the value, falling through to the IPv4 path for a dotted quad.
+        listOf(
+            "addr", "address", "gateway", "dst", "src", "servers", "server", "resolver",
+            "next_hop", "via", "public_ip", "observed_ip",
+        ).forEach { put(it, LogicalType.IP6) }
+
         listOf("mac", "hw_addr", "gateway_mac", "router_mac", "sender_mac", "peer_mac")
             .forEach { put(it, LogicalType.MAC) }
         listOf("bssid", "ap_mac").forEach { put(it, LogicalType.BSSID) }
@@ -40,6 +50,8 @@ object Classification {
         listOf(
             "fqdn", "hostname", "host", "name", "reverse_dns", "ptr", "domain", "query_name",
             "friendly_name", "server_name", "sni", "cname", "search_domain", "device_name",
+            // Plural and prefixed variants the models actually use.
+            "search_domains", "private_dns_hostname", "domains", "hostnames",
         ).forEach { put(it, LogicalType.FQDN) }
 
         listOf("session_id", "credential", "token", "device_id", "android_id", "serial", "imsi", "iccid")
@@ -78,6 +90,49 @@ object Classification {
         return null
     }
 
+    /**
+     * Last-resort classification from the *value*, when the field name is unrecognised.
+     *
+     * A name table can only protect fields somebody remembered to add, which is the wrong
+     * property for a privacy control: the dangerous field is the one nobody thought of. This
+     * exists because that failed once already — `addresses[].addr` holds the device's own global
+     * IPv6 address, the table had never heard of the name, and it went out verbatim.
+     *
+     * Only addresses and MACs are inferred, because only those have shapes that cannot be
+     * mistaken for something else. Hostnames deliberately are not: `train.udp_updown` is
+     * indistinguishable from a domain by shape, and mangling a test type would corrupt the
+     * document to protect nothing.
+     */
+    fun inferFromValue(value: String): LogicalType? {
+        val v = value.trim()
+        if (v.isEmpty() || v.length > 64) return null
+        if (looksLikeMac(v)) return LogicalType.MAC
+        if (looksLikeIp6(v)) return LogicalType.IP6
+        if (looksLikeIp4(v)) return LogicalType.IP4
+        return null
+    }
+
+    private fun isHex(c: Char) = c in '0'..'9' || c in 'a'..'f' || c in 'A'..'F'
+
+    private fun looksLikeMac(v: String): Boolean {
+        val parts = v.split(':', '-')
+        return parts.size == 6 && parts.all { p -> p.length == 2 && p.all(::isHex) }
+    }
+
+    private fun looksLikeIp4(v: String): Boolean {
+        val parts = v.substringBefore('/').split('.')
+        return parts.size == 4 && parts.all { p ->
+            p.isNotEmpty() && p.length <= 3 && p.all(Char::isDigit) && p.toInt() <= 255
+        }
+    }
+
+    private fun looksLikeIp6(v: String): Boolean {
+        val core = v.substringBefore('/').substringBefore('%')
+        // Two colons minimum, so a time or a MAC fragment does not qualify, and nothing but the
+        // characters an address may contain.
+        return core.count { it == ':' } >= 2 && core.all { it == ':' || isHex(it) }
+    }
+
     fun dropAtBalanced(path: List<String>): Boolean {
         if (path.isNotEmpty() && path.last() in droppedKeys) return true
         return droppedPaths.any { dropped -> dropped.all { path.contains(it) } }

echolot-app/core-privacy/src/test/kotlin/app/echo_lot/privacy/AnonymizerTest.kt

@@ -182,4 +182,47 @@ class AnonymizerTest {
         assertEquals(PrivacyLevel.BALANCED, PrivacyLevel.max(PrivacyLevel.BALANCED, PrivacyLevel.FULL))
         assertEquals(PrivacyLevel.FULL, PrivacyLevel.fromWire("nonsense"))
     }
+
+    // A ULA looks like the v6 RFC1918 and is not. Its global ID is 40 random bits, unique to one
+    // network by construction (RFC 4193), so the prefix IS the identifier - unlike 192.168.x,
+    // which millions of networks share. Passing the leading groups through leaked most of it.
+    @Test
+    fun ulaPrefixesArePseudonymizedWhole() {
+        val doc = json.parseToJsonElement(
+            """{"run":{"id":"r"},"networks":[{"link":{"dns":{"servers":["fda1:3fb1:ff92:6696::2662"]}}}]}"""
+        ).jsonObject
+        val out = flat(anon(PrivacyLevel.BALANCED, doc))
+        assertFalse(out.contains("fda1"), "the ULA global ID survived: $out")
+        assertFalse(out.contains("3fb1"), "part of the ULA global ID survived: $out")
+        assertTrue(out.contains("fd"), "the result should still read as a ULA: $out")
+    }
+
+    // Pseudonymizing the prefix as a unit keeps the one fact that is diagnostically useful:
+    // whether two addresses sit on the same network.
+    @Test
+    fun addressesOnOneUlaSubnetStayRelated() {
+        val doc = json.parseToJsonElement(
+            """{"run":{"id":"r"},"networks":[{"link":{"dns":{"servers":[
+               "fda1:3fb1:ff92:6696::1","fda1:3fb1:ff92:6696::2","fdff:9999:8888:7777::1"]}}}]}"""
+        ).jsonObject
+        val servers = anon(PrivacyLevel.BALANCED, doc)["networks"]!!.jsonArray[0].jsonObject["link"]!!
+            .jsonObject["dns"]!!.jsonObject["servers"]!!.jsonArray.map { it.jsonPrimitive.content }
+        val prefixOf = { s: String -> s.substringBeforeLast("::") }
+        assertEquals(prefixOf(servers[0]), prefixOf(servers[1]),
+            "two addresses on one ULA subnet should share a pseudonymous prefix")
+        assertNotEquals(prefixOf(servers[0]), prefixOf(servers[2]),
+            "a different ULA network must not collide with the first")
+    }
+
+    // RFC1918 stays readable, and this is the contrast that justifies it: a shared, meaningless
+    // prefix is topology; a unique random one is identity.
+    @Test
+    fun rfc1918StaysReadableUnlikeUla() {
+        val doc = json.parseToJsonElement(
+            """{"run":{"id":"r"},"networks":[{"link":{"dns":{"servers":["192.168.1.1","10.13.102.1"]}}}]}"""
+        ).jsonObject
+        val out = flat(anon(PrivacyLevel.BALANCED, doc))
+        assertTrue(out.contains("192.168.1.1"), "RFC1918 should survive: $out")
+        assertTrue(out.contains("10.13.102.1"), "RFC1918 should survive: $out")
+    }
 }

echolot-app/core-privacy/src/test/kotlin/app/echo_lot/privacy/LeakTest.kt

@@ -0,0 +1,177 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package app.echo_lot.privacy
+
+import kotlinx.serialization.json.Json
+import kotlinx.serialization.json.JsonObject
+import kotlinx.serialization.json.jsonArray
+import kotlinx.serialization.json.jsonObject
+import kotlinx.serialization.json.jsonPrimitive
+import kotlin.test.Test
+import kotlin.test.assertFalse
+import kotlin.test.assertTrue
+
+/**
+ * The blunt instrument: build a document with identifying values in every place one can actually
+ * occur, anonymize it, and assert none of them survive.
+ *
+ * [AnonymizerTest] checks that the fields the classification table knows about are handled
+ * correctly. This checks the other half — the fields it does *not* know about. A per-field test
+ * can only fail for a field someone remembered to write a case for, which is exactly the wrong
+ * property for a privacy check: the dangerous field is the one nobody thought of.
+ *
+ * Concretely, this is written the way it is because the schema's own field names disagree with
+ * the classifier's. `Address.addr` carries an IP and is documented as such in
+ * measurement-schema.md §8, but the classifier keys on names like `ip4` and `gateway_ip4` and had
+ * never heard of `addr`.
+ */
+class LeakTest {
+
+    private val json = Json { prettyPrint = false }
+    private val salt = Salt.perRun(ByteArray(32) { 3 })
+
+    /**
+     * Every string here is something that identifies a person, a household or a device, placed
+     * where the real models actually put it (`core-measurement`'s Network/Link/Address/DnsConfig).
+     */
+    private val secrets = listOf(
+        "Rambossek WLAN",              // ssid
+        "78:9a:18:aa:bb:cc",           // bssid
+        "aa:bb:cc:dd:ee:11",           // gateway mac
+        "2001:1ad0:c4fe:6767::150",    // global v6 address on the interface
+        "2a02:1748:dead:beef::1",      // v6 default gateway
+        "203.0.113.77",                // public v4
+        "nas.rambossek.lan",           // private-dns hostname
+        "rambossek.lan",               // search domain
+        "Anna's Chromecast",           // neighbour name
+        "kitchen table",               // free-text note
+    )
+
+    private fun document(): String = """
+        {
+          "schema": "echolot/measurement",
+          "run": {
+            "id": "run-1", "trigger": "manual", "notes": "${secrets[9]}",
+            "device": {"manufacturer": "OnePlus", "model": "CPH2747"}
+          },
+          "networks": [{
+            "id": "net-1", "transport": "wifi",
+            "link": {
+              "mtu": 1500,
+              "addresses": [
+                {"addr": "${secrets[3]}", "prefix_len": 64, "scope": "global"},
+                {"addr": "192.168.1.44", "prefix_len": 24, "scope": "global"}
+              ],
+              "routes": [
+                {"dst": "::/0", "gateway": "${secrets[4]}", "iface": "wlan0"},
+                {"dst": "0.0.0.0/0", "gateway": "192.168.1.1", "iface": "wlan0"}
+              ],
+              "dns": {
+                "servers": ["${secrets[5]}", "192.168.1.1"],
+                "private_dns_hostname": "${secrets[6]}",
+                "search_domains": ["${secrets[7]}"]
+              }
+            },
+            "wifi": {"ssid": "${secrets[0]}", "bssid": "${secrets[1]}"},
+            "neighbors": [{"name": "${secrets[8]}", "mac": "${secrets[2]}"}]
+          }],
+          "tests": [{"id": "t1", "type": "train.udp_updown", "status": "ok",
+                     "metrics": {"rtt_ms_avg": 12.4}}],
+          "findings": [],
+          "summary": {"verdict": "ok"}
+        }
+    """.trimIndent()
+
+    private fun anonymized(level: PrivacyLevel): String =
+        json.encodeToString(
+            kotlinx.serialization.json.JsonObject.serializer(),
+            Anonymizer(level, salt).anonymize(json.parseToJsonElement(document()).jsonObject),
+        )
+
+    @Test
+    fun nothingIdentifyingSurvivesBalanced() {
+        val out = anonymized(PrivacyLevel.BALANCED)
+        val leaked = secrets.filter { out.contains(it) }
+        assertTrue(
+            leaked.isEmpty(),
+            "these identifying values were uploaded verbatim at BALANCED: $leaked\n\n$out",
+        )
+    }
+
+    @Test
+    fun nothingIdentifyingSurvivesStrict() {
+        val out = anonymized(PrivacyLevel.STRICT)
+        val leaked = secrets.filter { out.contains(it) }
+        assertTrue(leaked.isEmpty(), "leaked at STRICT: $leaked\n\n$out")
+    }
+
+    // Private addresses are kept on purpose — they describe the topology and not the person — so
+    // this pins that the leak test above is not passing by accident of over-redaction.
+    @Test
+    fun privateAddressesAreStillReadable() {
+        val out = anonymized(PrivacyLevel.BALANCED)
+        assertTrue(out.contains("192.168.1.1"), "RFC1918 gateway should survive: $out")
+        assertTrue(out.contains("192.168.1.44"), "RFC1918 interface address should survive: $out")
+    }
+
+    /**
+     * Raw shell output embeds a complete inventory of the local network, and neither the field-name
+     * table nor the whole-value shape check can see it: `ip_neigh` is one long string that is
+     * itself neither a MAC nor an address.
+     *
+     * This is not hypothetical. The blob below is (abridged) real output that reached the server
+     * at the `balanced` level from a test device, carrying the hardware address of every host on
+     * the network. measurement-schema.md §9 had flagged raw dumps as "hard to anonymize"; nothing
+     * enforced it.
+     */
+    @Test
+    fun identifiersInsideRawShellOutputAreScrubbed() {
+        // Joined rather than written with escapes, so the fixture stays readable and there is no
+        // chance of an escape being mangled on its way into the JSON below.
+        val dump = listOf(
+            "uid=2000",
+            "10.13.102.5 dev wlan0 lladdr 90:09:d0:1a:83:e4 REACHABLE",
+            "10.13.102.1 dev wlan0 lladdr 78:9a:18:54:b8:f9 REACHABLE",
+            "10.13.102.111 dev wlan0 lladdr dc:a2:66:08:69:95 STALE",
+            "2001:4bb8:46a:e724:289d:87ff:feb6:ebd3 dev wlan0 lladdr b8:be:f4:bc:ca:cf STALE",
+        ).joinToString(" | ")
+        val doc = json.parseToJsonElement(
+            """{"run":{"id":"r"},"tests":[{"id":"t","type":"link.ip_monitor",
+               "evidence":{"ip_neigh":"$dump"}}]}"""
+        ).jsonObject
+        val out = json.encodeToString(
+            kotlinx.serialization.json.JsonObject.serializer(),
+            Anonymizer(PrivacyLevel.BALANCED, salt).anonymize(doc),
+        )
+
+        for (mac in listOf("90:09:d0:1a:83:e4", "78:9a:18:54:b8:f9", "dc:a2:66:08:69:95", "b8:be:f4:bc:ca:cf")) {
+            assertFalse(out.contains(mac), "a neighbour's MAC survived inside the raw dump: $mac")
+        }
+        assertFalse(out.contains("2001:4bb8:46a:e724:289d:87ff:feb6:ebd3"),
+            "a global IPv6 survived inside the raw dump")
+
+        // Scrubbed, not dropped: the evidence must still be readable, or the raw dump stops being
+        // evidence at all. Structure, hostnames of the fields, and RFC1918 addresses stay.
+        assertTrue(out.contains("REACHABLE") && out.contains("STALE"), "the dump lost its structure")
+        assertTrue(out.contains("10.13.102.1"), "RFC1918 addresses should stay readable: $out")
+        assertTrue(out.contains("78:9a:18"), "the vendor prefix should survive for identification")
+    }
+
+    // A MAC in a raw dump and the same MAC in a parsed field must land on the same pseudonym, or
+    // the document stops being internally consistent and one device reads as two.
+    @Test
+    fun theSameIdentifierMatchesAcrossParsedAndRawFields() {
+        val doc = json.parseToJsonElement(
+            """{"run":{"id":"r"},
+               "networks":[{"wifi":{"bssid":"78:9a:18:54:b8:f9"}}],
+               "tests":[{"id":"t","evidence":{"ip_neigh":"gw dev wlan0 lladdr 78:9a:18:54:b8:f9 REACHABLE"}}]}"""
+        ).jsonObject
+        val out = Anonymizer(PrivacyLevel.BALANCED, salt).anonymize(doc)
+        val parsed = out["networks"]!!.jsonArray[0].jsonObject["wifi"]!!.jsonObject["bssid"]!!
+            .jsonPrimitive.content
+        val raw = json.encodeToString(kotlinx.serialization.json.JsonObject.serializer(), out)
+        assertTrue(raw.contains(parsed),
+            "the parsed BSSID pseudonym ($parsed) does not appear in the scrubbed dump")
+    }
+}

echolot-app/core-privacy/src/test/kotlin/app/echo_lot/privacy/RealDocumentTest.kt

@@ -0,0 +1,48 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package app.echo_lot.privacy
+
+import kotlinx.serialization.json.Json
+import kotlinx.serialization.json.JsonObject
+import kotlinx.serialization.json.jsonObject
+import java.io.File
+import kotlin.test.Test
+import kotlin.test.assertTrue
+
+/**
+ * Runs the anonymizer over a real captured document when one is supplied via ECHOLOT_REAL_RUN,
+ * and reports every MAC and public address that survives.
+ *
+ * Fixtures only contain the identifiers somebody thought to put in them. A real run off a real
+ * phone contains whatever the probes actually produce — which is how the raw-shell-output leak was
+ * found in the first place. Self-skips when no document is supplied, so nobody's network ends up
+ * committed to the repository.
+ */
+class RealDocumentTest {
+
+    @Test
+    fun noIdentifiersSurviveInARealDocument() {
+        val path = System.getenv("ECHOLOT_REAL_RUN")
+        if (path.isNullOrBlank() || !File(path).isFile) {
+            println("RealDocumentTest skipped (set ECHOLOT_REAL_RUN to a captured run)"); return
+        }
+        val json = Json { prettyPrint = false }
+        val doc = json.parseToJsonElement(File(path).readText()).jsonObject
+        val out = json.encodeToString(
+            JsonObject.serializer(),
+            Anonymizer(PrivacyLevel.BALANCED, Salt.perRun(ByteArray(32) { 5 })).anonymize(doc),
+        )
+
+        val macs = Regex("""\b[0-9a-fA-F]{2}(?::[0-9a-fA-F]{2}){5}\b""").findAll(out)
+            .map { it.value.lowercase() }
+            .filter { it != "00:00:00:00:00:00" }
+            .toSet()
+        val original = Regex("""\b[0-9a-fA-F]{2}(?::[0-9a-fA-F]{2}){5}\b""")
+            .findAll(File(path).readText()).map { it.value.lowercase() }.toSet()
+
+        val survived = macs intersect original
+        println("MACs in the original: ${original.size}; unchanged after anonymizing: ${survived.size}")
+        assertTrue(survived.isEmpty(), "these real MAC addresses survived anonymization: $survived")
+    }
+}

echolot-app/core-probe/src/main/kotlin/app/echo_lot/probe/ConstraintDetector.kt

@@ -0,0 +1,68 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package app.echo_lot.probe
+
+import android.content.Context
+import android.net.ConnectivityManager
+import android.net.NetworkCapabilities
+import android.system.ErrnoException
+import android.system.OsConstants
+import app.echo_lot.measurement.Constraints
+import app.echo_lot.measurement.Transport
+import java.net.DatagramSocket
+
+/**
+ * Detects what will prevent this run from measuring (measurement-schema.md §3 `constraints`),
+ * before any probe runs and independently of all of them.
+ *
+ * The known case: while a VPN holds the default route, Android refuses `Network.bindSocket()` on
+ * the underlying networks (EPERM) so apps cannot leak around the tunnel. Every per-network test
+ * then silently measures the tunnel or nothing, and the run comes out shaped exactly like a clean
+ * run of a healthy network. Detecting that here — one throwaway bind per network — is what lets
+ * the document say "these networks went unmeasured" instead of leaving the reader to infer it
+ * from a pattern of `attempted: false` scattered across the tests.
+ */
+object ConstraintDetector {
+
+    fun detect(ctx: Context, entries: List<NetworkInventory.Entry>): Constraints {
+        val cm = ctx.getSystemService(Context.CONNECTIVITY_SERVICE) as ConnectivityManager
+        // "A VPN holds the default route" is judged from the ACTIVE network, not from a VPN
+        // network merely existing in the list: a tunnel that was just disconnected lingers in
+        // allNetworks while it tears down, and counting it kept the app claiming "measured
+        // through a VPN" after the VPN was gone.
+        val vpnActive = runCatching {
+            cm.getNetworkCapabilities(cm.activeNetwork)
+                ?.hasTransport(NetworkCapabilities.TRANSPORT_VPN) == true
+        }.getOrDefault(false)
+
+        val refused = ArrayList<String>()
+        for (e in entries) {
+            // The tunnel itself stays bindable — it is the underlying networks the OS walls off.
+            if (e.model.transport == Transport.VPN) continue
+            val err = try {
+                DatagramSocket().use { s -> e.handle.bindSocket(s) }
+                null
+            } catch (t: Throwable) {
+                t
+            }
+            // Only the OS *refusing* counts as blocked (EPERM: the VPN wall). A network that
+            // happens to die mid-snapshot fails its bind too, but with a different errno, and
+            // calling that "per-network probing blocked" would flip a whole healthy run to
+            // INCONCLUSIVE over one network going away — the probes already record
+            // attempted:false for that case.
+            if (err != null && isPermissionRefusal(err)) refused.add(e.model.id)
+        }
+        return Constraints(
+            vpnActive = vpnActive,
+            perNetworkBlocked = refused.isNotEmpty(),
+            unmeasuredNetworks = refused,
+        )
+    }
+
+    private fun isPermissionRefusal(t: Throwable): Boolean =
+        generateSequence(t) { it.cause }.any {
+            (it is ErrnoException && it.errno == OsConstants.EPERM) ||
+                (it.message?.contains("EPERM") == true)
+        }
+}

echolot-app/core-probe/src/main/kotlin/app/echo_lot/probe/DnsResolverProbe.kt

@@ -0,0 +1,227 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package app.echo_lot.probe
+
+import android.content.Context
+import app.echo_lot.measurement.Test
+import app.echo_lot.measurement.TestStatus
+import app.echo_lot.measurement.TestType
+import app.echo_lot.measurement.Tier
+import kotlinx.coroutines.Dispatchers
+import kotlinx.coroutines.withContext
+import kotlinx.serialization.json.JsonObject
+import kotlinx.serialization.json.buildJsonObject
+import kotlinx.serialization.json.put
+import java.net.DatagramPacket
+import java.net.DatagramSocket
+import java.net.InetAddress
+import java.net.InetSocketAddress
+import java.util.Random
+
+/**
+ * Asks the network's own DNS servers directly, then asks Android to resolve the same name, and
+ * compares.
+ *
+ * The comparison is the point. A name that fails to resolve looks the same to a user whatever the
+ * cause, but the causes want opposite responses: if the server does not answer, the network is
+ * broken and the router is the thing to look at; if the server answers a raw query while the
+ * platform still cannot resolve, the device's own resolver has wedged and toggling wifi fixes it in
+ * seconds. Nothing else on a phone will tell you which of those you have.
+ *
+ * This is deliberately not a general DNS test — no recursion checks, no DNSSEC, no rewriting
+ * detection; [DnsCanaryProbe] covers interception. This one answers a single question: is the
+ * resolver on this device doing its job.
+ */
+class DnsResolverProbe(
+    private val entries: List<NetworkInventory.Entry>,
+    /** Resolved directly rather than through any cache; any name with a stable answer will do. */
+    private val probeName: String = "one.one.one.one",
+) : Probe {
+    override val type = TestType.DNS_RESOLVER
+    override val tier = Tier.APP
+    // A query per server with a 3s ceiling, plus one getaddrinfo that may sit out its own timeout.
+    override val estimatedMs = 6_000L
+
+    override suspend fun run(ctx: Context, ids: ProbeIds): Test = withContext(Dispatchers.IO) {
+        val b = TestBuilder(type, tier, ids)
+        val perNetwork = LinkedHashMap<String, Pair<String, JsonObject>>()
+
+        for (e in entries) {
+            val servers = e.model.link.dns?.servers.orEmpty()
+            if (servers.isEmpty()) continue
+            val label = "${e.model.transport.name.lowercase()}:${e.model.id}"
+
+            // Directly: does the configured server answer at all?
+            var direct: Boolean? = null
+            var directDetail = "no server answered"
+            for (s in servers) {
+                val r = try {
+                    queryDirect(s, probeName)
+                } catch (e: DnsRefused) {
+                    // Distinguished deliberately: a server that replies with a failure is a
+                    // working server saying no, which points at the network rather than here.
+                    direct = false
+                    directDetail = "$s ${e.why}"
+                    break
+                }
+                if (r != null) {
+                    direct = true
+                    directDetail = "$s answered in ${r}ms"
+                    break
+                }
+                direct = false
+                directDetail = "$s did not answer"
+            }
+
+            // The search domains the network handed out, asked about separately.
+            //
+            // A resolver appends these to a lookup, so a search domain the server will not answer
+            // for stalls every name a client asks about — and it fails as silence, which is
+            // indistinguishable from packet loss, so clients retry rather than moving on. Asking
+            // about a name that cannot exist is deliberate: the answer wanted here is NXDOMAIN,
+            // and what matters is only whether anything comes back at all.
+            val searchDomains = e.model.link.dns?.searchDomains.orEmpty()
+            var searchAnswered: Boolean? = null
+            var searchDetail = ""
+            for (d in searchDomains) {
+                val nonce = "echolot-probe-" + java.util.UUID.randomUUID().toString().take(8)
+                val answered = servers.any { srv ->
+                    runCatching { queryDirect(srv, "$nonce.$d") != null }
+                        .getOrElse { it is DnsRefused } // a refusal is still an answer
+                }
+                if (!answered) {
+                    searchAnswered = false
+                    searchDetail = "$d is not answered at all — queries under it vanish"
+                    break
+                }
+                searchAnswered = true
+                searchDetail = "$d answers"
+            }
+
+            // Through the platform: what an app actually gets.
+            val viaSystem = runCatching {
+                e.handle.getAllByName(probeName).isNotEmpty()
+            }.getOrElse { false }
+
+            perNetwork[label] = e.model.id to buildJsonObject {
+                put("network_ref", e.model.id)
+                put("servers", servers.joinToString(","))
+                direct?.let { put("direct_answer", it) }
+                put("direct_detail", directDetail)
+                if (searchDomains.isNotEmpty()) {
+                    put("search_domains", searchDomains.joinToString(","))
+                    searchAnswered?.let { put("search_answered", it) }
+                    put("search_detail", searchDetail)
+                }
+                put("system_resolves", viaSystem)
+                // Named here rather than left for a finding to infer, because the pairing is the
+                // whole observation and splitting it across two places invites reading one alone.
+                put(
+                    "verdict",
+                    when {
+                        // Ordered by which component is at fault, most specific first. A search
+                        // domain that swallows queries explains a failure that would otherwise be
+                        // blamed on the device, so it has to be tested before that conclusion.
+                        searchAnswered == false -> "search domain swallows queries"
+                        viaSystem -> "resolver working"
+                        direct == true -> "server answers, device resolver does not"
+                        direct == false -> "server does not answer"
+                        else -> "not determined"
+                    },
+                )
+            }
+        }
+
+        if (perNetwork.isEmpty()) {
+            return@withContext b.build(
+                TestStatus.SKIPPED,
+                evidence = buildJsonObject { put("reason", "no network advertised a DNS server") },
+            )
+        }
+        val evidence = buildJsonObject {
+            put("name", probeName)
+            for ((label, v) in perNetwork) put(label, v.second)
+        }
+        // OK means the measurement ran, not that DNS is healthy — the finding says that.
+        b.build(TestStatus.OK, evidence = evidence)
+    }
+
+    /**
+     * Sends one A query straight to [server] over UDP. Returns the round trip in ms, or null.
+     *
+     * Hand-rolled rather than via any resolver API on purpose: the entire point is to bypass the
+     * component under suspicion. Anything that goes through the platform resolver would inherit
+     * exactly the fault this is trying to detect.
+     */
+    private fun queryDirect(server: String, name: String): Long? {
+        return try {
+            queryDirectOrThrow(server, name)
+        } catch (e: DnsRefused) {
+            throw e
+        } catch (t: Throwable) {
+            null
+        }
+    }
+
+    private fun queryDirectOrThrow(server: String, name: String): Long? = run {
+        val id = Random().nextInt(0xFFFF)
+        val query = buildQuery(id, name)
+        DatagramSocket().use { sock ->
+            sock.soTimeout = 3000
+            val addr = InetAddress.getByName(server) // a literal from DHCP; no lookup happens
+            val t0 = System.nanoTime()
+            sock.send(DatagramPacket(query, query.size, InetSocketAddress(addr, 53)))
+            val buf = ByteArray(512)
+            val reply = DatagramPacket(buf, buf.size)
+            sock.receive(reply)
+            val ms = (System.nanoTime() - t0) / 1_000_000
+            // A reply is not an answer. Counting any packet as success would let a REFUSED or
+            // SERVFAIL — both perfectly well-formed responses — be reported as "the server
+            // answers", and this probe's whole output is the claim that the server is fine and
+            // the device is not. That would be an accusation pointed at the wrong component,
+            // stated with confidence.
+            val replyId = ((buf[0].toInt() and 0xFF) shl 8) or (buf[1].toInt() and 0xFF)
+            val rcode = if (reply.length >= 4) buf[3].toInt() and 0x0F else -1
+            val answers = if (reply.length >= 8) {
+                ((buf[6].toInt() and 0xFF) shl 8) or (buf[7].toInt() and 0xFF)
+            } else 0
+            when {
+                replyId != id || reply.length < 12 -> null
+                rcode != 0 -> throw DnsRefused(rcodeName(rcode))
+                answers == 0 -> throw DnsRefused("answered with no records")
+                else -> ms
+            }
+        }
+    }
+
+    /** The server replied, but with a failure — which is a network fault, not a device one. */
+    private class DnsRefused(val why: String) : Exception(why)
+
+    private fun rcodeName(rcode: Int): String = when (rcode) {
+        1 -> "rejected the query as malformed"
+        2 -> "reported its own failure (SERVFAIL)"
+        3 -> "said the name does not exist (NXDOMAIN)"
+        4 -> "does not implement this query"
+        5 -> "refused the query (REFUSED)"
+        else -> "returned rcode $rcode"
+    }
+
+    /** A minimal DNS query: one question, class IN, type A, recursion desired. */
+    private fun buildQuery(id: Int, name: String): ByteArray {
+        val labels = name.split('.').filter { it.isNotEmpty() }
+        val out = ArrayList<Byte>(32)
+        out.add((id shr 8).toByte()); out.add(id.toByte())
+        out.add(0x01); out.add(0x00)   // recursion desired
+        out.add(0x00); out.add(0x01)   // one question
+        repeat(6) { out.add(0x00) }    // no answers, authority or additional
+        for (l in labels) {
+            out.add(l.length.toByte())
+            for (c in l.toByteArray(Charsets.US_ASCII)) out.add(c)
+        }
+        out.add(0x00)                  // root label
+        out.add(0x00); out.add(0x01)   // type A
+        out.add(0x00); out.add(0x01)   // class IN
+        return out.toByteArray()
+    }
+}

echolot-app/core-probe/src/main/kotlin/app/echo_lot/probe/IcmpProbe.kt

@@ -40,24 +40,37 @@ class IcmpProbe(
 
     override suspend fun run(ctx: Context, ids: ProbeIds): Test = withContext(Dispatchers.IO) {
         val b = TestBuilder(type, tier, ids)
-        val perNetwork = LinkedHashMap<String, String>()
+        val perNetwork = LinkedHashMap<String, Pair<String?, Attempt>>()
         var anyOk = false
         val rtts = ArrayList<Double>()
 
         // Default network first, then each active network explicitly.
-        attempt(null).let { (ok, detail, rtt) ->
-            perNetwork["default"] = detail; if (ok) { anyOk = true; rtt?.let(rtts::add) }
+        attempt(null).let { a ->
+            perNetwork["default"] = null to a
+            if (a.ok) { anyOk = true; a.rttMs?.let(rtts::add) }
         }
         for (e in entries) {
             val label = "${e.model.transport.name.lowercase()}:${e.model.id}"
-            val (ok, detail, rtt) = attempt(e.handle)
-            perNetwork[label] = detail
-            if (ok) { anyOk = true; rtt?.let(rtts::add) }
+            val a = attempt(e.handle)
+            perNetwork[label] = e.model.id to a
+            if (a.ok) { anyOk = true; a.rttMs?.let(rtts::add) }
         }
 
+        // Per-network results are recorded structurally, not just as prose. The aggregate status
+        // can only say "some network answered"; a finding needs to know *which* network failed,
+        // and recovering that by parsing a human-readable detail string would be a trap waiting to
+        // spring the first time the wording changes.
         val evidence: JsonObject = buildJsonObject {
             put("target", target)
-            for ((k, v) in perNetwork) put(k, v)
+            for ((label, r) in perNetwork) {
+                val (netId, a) = r
+                put(label, buildJsonObject {
+                    netId?.let { put("network_ref", it) }
+                    put("ok", a.ok)
+                    put("attempted", a.attempted)
+                    put("detail", a.detail)
+                })
+            }
         }
         val metrics: JsonObject = buildJsonObject {
             put("networks_ok", rtts.size)
@@ -69,14 +82,31 @@ class IcmpProbe(
         b.build(status, evidence = evidence, metrics = metrics)
     }
 
-    private data class Attempt(val ok: Boolean, val detail: String, val rttMs: Double?)
+    /**
+     * One network's result.
+     *
+     * [attempted] separates "we sent an echo request and heard nothing" from "we never got as far
+     * as sending one". Both leave [ok] false, and collapsing them is how a probe ends up asserting
+     * something about a network it never touched: binding to a non-default network can fail with
+     * EPERM, and reporting that as ICMPv6 silence blames the carrier for the app's own inability
+     * to use the interface.
+     */
+    private data class Attempt(
+        val ok: Boolean,
+        val attempted: Boolean,
+        val detail: String,
+        val rttMs: Double?,
+    )
 
     private fun attempt(network: Network?): Attempt {
         var fd: FileDescriptor? = null
+        var sent = false
         return try {
             val proto = if (v6) OsConstants.IPPROTO_ICMPV6 else OsConstants.IPPROTO_ICMP
             val family = if (v6) OsConstants.AF_INET6 else OsConstants.AF_INET
             fd = Os.socket(family, OsConstants.SOCK_DGRAM, proto)
+            // Everything up to and including sendto is setup. A failure here means the test did
+            // not run on this network — not that the network stayed silent.
             network?.bindSocket(fd)
             Os.setsockoptTimeval(fd, OsConstants.SOL_SOCKET, OsConstants.SO_RCVTIMEO, StructTimeval.fromMillis(3000))
             val addr = network?.getByName(target) ?: InetAddress.getByName(target)
@@ -85,14 +115,16 @@ class IcmpProbe(
             val packet = buildEchoRequest(v6, ident.toShort(), 1)
             val t0 = System.nanoTime()
             Os.sendto(fd, packet, 0, packet.size, 0, addr, 0)
+            sent = true
             val buf = ByteBuffer.allocate(1500)
             val received = Os.recvfrom(fd, buf, 0, null)
             val rttMs = (System.nanoTime() - t0) / 1_000_000.0
             val replyType = if (received > 0) buf.get(0).toInt() and 0xFF else -1
             val ok = replyType == (if (v6) 129 else 0)
-            Attempt(ok, "reply type=$replyType rtt_ms=${"%.1f".format(Locale.ROOT, rttMs)} bytes=$received", if (ok) rttMs else null)
+            Attempt(ok, true, "reply type=$replyType rtt_ms=${"%.1f".format(Locale.ROOT, rttMs)} bytes=$received", if (ok) rttMs else null)
         } catch (e: Throwable) {
-            Attempt(false, "error: ${e.message ?: e.javaClass.simpleName}", null)
+            // A timeout after a successful send is a real "no reply"; anything before it is not.
+            Attempt(false, sent, "error: ${e.message ?: e.javaClass.simpleName}", null)
         } finally {
             fd?.let { runCatching { Os.close(it) } }
         }

echolot-app/core-probe/src/main/kotlin/app/echo_lot/probe/MdnsInventoryProbe.kt

@@ -0,0 +1,107 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package app.echo_lot.probe
+
+import android.content.Context
+import android.net.nsd.NsdManager
+import android.net.nsd.NsdServiceInfo
+import android.net.wifi.WifiManager
+import app.echo_lot.measurement.Test
+import app.echo_lot.measurement.TestStatus
+import app.echo_lot.measurement.TestType
+import app.echo_lot.measurement.Tier
+import kotlinx.coroutines.Dispatchers
+import kotlinx.coroutines.delay
+import kotlinx.coroutines.withContext
+import kotlinx.serialization.json.buildJsonObject
+import kotlinx.serialization.json.put
+import kotlinx.serialization.json.putJsonObject
+import java.util.Collections
+
+/**
+ * local.mdns_inventory — what answers mDNS on this network (MulticastLock + NSD discovery).
+ * The service inventory doubles as the VLAN-leakage detector: a chromecast answering on the
+ * guest wifi is a segmentation fault made visible. Folded from the prober, validated on both
+ * known devices (4 services each).
+ *
+ * Two hardware-bought lessons are load-bearing here:
+ *  - The `_services._dns-sd._udp.` meta-query returned 0 on BOTH devices while concrete types
+ *    found live services — NsdManager's meta-query support is unreliable across builds, so the
+ *    concrete types are the measurement and the meta-query result is itself evidence.
+ *  - 4 s of listening missed services that 10 s catches; mDNS answers straggle.
+ */
+class MdnsInventoryProbe : Probe {
+    override val type = TestType.LOCAL_MDNS_INVENTORY
+    override val tier = Tier.APP
+    override val estimatedMs = 10_500L
+
+    /** Meta-query + common concrete types (HTTP covers HA/printers/NAS; googlecast is ubiquitous). */
+    private val queries = listOf(
+        "meta" to "_services._dns-sd._udp.",
+        "http" to "_http._tcp.",
+        "googlecast" to "_googlecast._tcp.",
+    )
+
+    private class Recorder : NsdManager.DiscoveryListener {
+        val names: MutableList<String> = Collections.synchronizedList(mutableListOf())
+        @Volatile var started = false
+        @Volatile var startFailCode: Int? = null
+        override fun onStartDiscoveryFailed(t: String?, code: Int) { startFailCode = code }
+        override fun onStopDiscoveryFailed(t: String?, code: Int) {}
+        override fun onDiscoveryStarted(t: String?) { started = true }
+        override fun onDiscoveryStopped(t: String?) {}
+        override fun onServiceFound(s: NsdServiceInfo?) { s?.serviceName?.let { names.add(it) } }
+        override fun onServiceLost(s: NsdServiceInfo?) {}
+    }
+
+    override suspend fun run(ctx: Context, ids: ProbeIds): Test = withContext(Dispatchers.IO) {
+        val b = TestBuilder(type, tier, ids)
+        val wifi = ctx.getSystemService(WifiManager::class.java)
+        val lock = wifi?.createMulticastLock("echolot")?.apply {
+            setReferenceCounted(false)
+            runCatching { acquire() }
+        }
+        val nsd = ctx.getSystemService(NsdManager::class.java)
+            ?: return@withContext b.build(
+                TestStatus.UNSUPPORTED,
+                evidence = buildJsonObject { put("reason", "NsdManager unavailable") },
+            )
+
+        val recorders = queries.map { (label, type) ->
+            val r = Recorder()
+            runCatching { nsd.discoverServices(type, NsdManager.PROTOCOL_DNS_SD, r) }
+                .onFailure { r.startFailCode = -1 }
+            Triple(label, type, r)
+        }
+        try {
+            delay(10_000)
+            var total = 0
+            var anyStarted = false
+            val evidence = buildJsonObject {
+                put("multicast_lock", lock?.isHeld == true)
+                for ((label, type, r) in recorders) {
+                    runCatching { nsd.stopServiceDiscovery(r) }
+                    anyStarted = anyStarted || r.started
+                    val names = r.names.distinct()
+                    total += names.size
+                    putJsonObject(label) {
+                        put("query", type)
+                        put("started", r.started)
+                        r.startFailCode?.let { put("start_fail_code", it) }
+                        put("found", names.size)
+                        if (names.isNotEmpty()) put("names", names.joinToString(", ").take(300))
+                    }
+                }
+            }
+            val metrics = buildJsonObject { put("services_found", total) }
+            // Zero services on a started discovery is a legitimate result (an empty or properly
+            // isolated network), not a failure — only discovery refusing to start is one.
+            b.build(if (anyStarted) TestStatus.OK else TestStatus.FAILED,
+                evidence = evidence, metrics = metrics)
+        } finally {
+            recorders.forEach { (_, _, r) -> runCatching { nsd.stopServiceDiscovery(r) } }
+            runCatching { lock?.release() }
+        }
+    }
+}

echolot-app/core-probe/src/main/kotlin/app/echo_lot/probe/NetworkInventory.kt

@@ -39,6 +39,32 @@ object NetworkInventory {
         return out
     }
 
+    /**
+     * Android's own verdict on the network, read straight from the capabilities it already has.
+     *
+     * PARTIAL_CONNECTIVITY only exists from API 28 and CAPTIVE_PORTAL from 23, so both are read
+     * defensively: an older platform that cannot answer should leave the field null rather than
+     * assert a false.
+     */
+    private fun systemVerdict(caps: NetworkCapabilities): app.echo_lot.measurement.SystemVerdict =
+        app.echo_lot.measurement.SystemVerdict(
+            validated = caps.hasCapability(NetworkCapabilities.NET_CAPABILITY_VALIDATED),
+            captivePortal = runCatching {
+                caps.hasCapability(NetworkCapabilities.NET_CAPABILITY_CAPTIVE_PORTAL)
+            }.getOrNull(),
+            // NET_CAPABILITY_PARTIAL_CONNECTIVITY is @SystemApi, so the constant is not in the
+            // public SDK even though the platform sets it from API 28. The number is stable —
+            // changing it would break every system app that reads it — but this is a value the
+            // SDK does not promise us, so it is asked for defensively and reported as unknown
+            // rather than as false if anything about it is not as expected.
+            partialConnectivity = runCatching {
+                caps.hasCapability(NET_CAPABILITY_PARTIAL_CONNECTIVITY)
+            }.getOrNull(),
+        )
+
+    /** @SystemApi NetworkCapabilities.NET_CAPABILITY_PARTIAL_CONNECTIVITY, API 28+. */
+    private const val NET_CAPABILITY_PARTIAL_CONNECTIVITY = 24
+
     private fun toModel(id: String, caps: NetworkCapabilities, lp: LinkProperties): MNetwork {
         val transport = when {
             caps.hasTransport(NetworkCapabilities.TRANSPORT_WIFI) -> Transport.WIFI
@@ -72,6 +98,7 @@ object NetworkInventory {
         return MNetwork(
             id = id, transport = transport, iface = lp.interfaceName,
             link = Link(mtu = lp.mtu.takeIf { it > 0 }, addresses = addresses, routes = routes, dns = dns),
+            systemVerdict = systemVerdict(caps),
         )
     }
 }

echolot-app/core-probe/src/main/kotlin/app/echo_lot/probe/OsAbi.kt

@@ -0,0 +1,70 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package app.echo_lot.probe
+
+import android.system.Os
+import java.io.FileDescriptor
+
+/**
+ * Linux socket-option ABI numbers that android.system.OsConstants does NOT reliably expose.
+ * Stable across Android's supported ABIs at the IP/IPv6 protocol levels, which is why they can
+ * be hardcoded: if setsockoptInt with one of these succeeds, the kernel accepted the option; if
+ * it throws ErrnoException, it did not. Either outcome is data. Do not "fix" these to
+ * OsConstants names — they don't exist there (validated in the prober; see its OsAbi.kt).
+ *
+ * Measured fact worth keeping: `Os.getsockoptInt` is absent on both known devices (OnePlus 15
+ * A16, Lenovo TB330FU A15), so path-MTU values must be read from the errqueue (`ee_info`), never
+ * from getsockopt(IP_MTU).
+ */
+object OsAbi {
+    // IP level
+    const val IP_TTL = 2
+    const val IP_MTU_DISCOVER = 10
+    const val IP_MTU = 14
+    const val IP_RECVERR = 11
+    const val IP_PMTUDISC_DO = 2      // set DF, honor PMTU
+    const val IP_PMTUDISC_PROBE = 3   // set DF, ignore PMTU (for probing)
+
+    // IPv6 level
+    const val IPV6_MTU_DISCOVER = 23
+    const val IPV6_MTU = 24
+    const val IPV6_RECVERR = 25
+    const val IPV6_UNICAST_HOPS = 16
+    const val IPV6_PMTUDISC_PROBE = 3
+
+    // recv flags — not in OsConstants on any current API level
+    const val MSG_ERRQUEUE = 0x2000
+    const val MSG_DONTWAIT = 0x40
+
+    // struct sock_extended_err (uapi/linux/errqueue.h), fixed layout on all Android ABIs:
+    // u32 ee_errno; u8 ee_origin; u8 ee_type; u8 ee_code; u8 ee_pad; u32 ee_info; u32 ee_data;
+    // followed directly by the offender sockaddr (SO_EE_OFFENDER).
+    const val SOCK_EE_SIZE = 16
+    const val SO_EE_ORIGIN_ICMP = 2
+    const val ICMP_TIME_EXCEEDED = 11
+    const val ICMP_DEST_UNREACH = 3
+
+    /** Try setsockoptInt; return null on success, or the errno name on failure. */
+    fun trySetIntOpt(fd: FileDescriptor, level: Int, opt: Int, value: Int): String? =
+        try {
+            Os.setsockoptInt(fd, level, opt, value)
+            null
+        } catch (e: Throwable) {
+            e.message ?: e.javaClass.simpleName
+        }
+
+    /**
+     * getsockoptInt is not part of the stable public Os surface on every API level, so it is
+     * reached via reflection; callers must treat failure as "unreadable", not as an error.
+     */
+    fun tryGetIntOpt(fd: FileDescriptor, level: Int, opt: Int): Result<Int> = runCatching {
+        val m = Os::class.java.getMethod(
+            "getsockoptInt",
+            FileDescriptor::class.java,
+            Int::class.javaPrimitiveType,
+            Int::class.javaPrimitiveType,
+        )
+        m.invoke(null, fd, level, opt) as Int
+    }
+}

echolot-app/core-probe/src/main/kotlin/app/echo_lot/probe/Probe.kt

@@ -55,9 +55,10 @@ class TestBuilder(
         evidence: JsonObject? = null,
         metrics: JsonObject? = null,
         error: TestError? = null,
+        params: JsonObject? = null,
     ): Test = Test(
         id = id, type = type, networkRef = networkRef, sessionRef = sessionRef, tier = tier,
         startedMonoNs = startedMonoNs, endedMonoNs = ids.monoNs(),
-        status = status, error = error, evidence = evidence, metrics = metrics,
+        status = status, error = error, params = params, evidence = evidence, metrics = metrics,
     )
 }

echolot-app/core-probe/src/main/kotlin/app/echo_lot/probe/StunProbe.kt

@@ -60,6 +60,15 @@ class StunProbe(
 
     override suspend fun run(ctx: Context, ids: ProbeIds): Test = withContext(Dispatchers.IO) {
         val b = TestBuilder(type, tier, ids)
+        // Without a server there is nothing to ask. Skipped rather than failed: "the STUN test
+        // failed" reads as a finding about the network, when the truth is that this device is
+        // not enrolled anywhere and no packet was ever sent.
+        if (serverHost.isBlank()) {
+            return@withContext b.build(
+                TestStatus.SKIPPED,
+                evidence = buildJsonObject { put("reason", "no server configured to ask") },
+            )
+        }
         DatagramSocket().use { sock ->
             sock.soTimeout = 3000
             val localPort = sock.localPort

echolot-app/core-probe/src/main/kotlin/app/echo_lot/probe/TracerouteProbe.kt

@@ -0,0 +1,255 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package app.echo_lot.probe
+
+import android.content.Context
+import android.system.Os
+import android.system.OsConstants
+import app.echo_lot.measurement.Flow
+import app.echo_lot.measurement.Hop
+import app.echo_lot.measurement.HopProbe
+import app.echo_lot.measurement.Test
+import app.echo_lot.measurement.TestError
+import app.echo_lot.measurement.TestStatus
+import app.echo_lot.measurement.TestType
+import app.echo_lot.measurement.Tier
+import app.echo_lot.measurement.TracerouteEvidence
+import app.echo_lot.measurement.toEvidence
+import kotlinx.coroutines.Dispatchers
+import kotlinx.coroutines.delay
+import kotlinx.coroutines.withContext
+import kotlinx.serialization.json.buildJsonObject
+import kotlinx.serialization.json.put
+import java.io.FileDescriptor
+import java.net.InetAddress
+import java.net.InetSocketAddress
+import java.nio.ByteBuffer
+import java.nio.ByteOrder
+
+/**
+ * traceroute.udp4 — UDP traceroute reading ICMP time-exceeded off the socket error queue via
+ * Os.recvmsg(MSG_ERRQUEUE): no root, no raw socket, no native code. Folded from the prober,
+ * which validated real hop addresses on both known devices (6 hops on the OnePlus 15, 5 on the
+ * Lenovo) and thereby retired the planned C-over-JNI errqueue shim.
+ *
+ * StructMsghdr/StructCmsghdr/recvmsg are reached via reflection (repo convention for uncertain
+ * OS paths): present since roughly API 34, absent before, and the probe must run — and report —
+ * on both. An absent API is UNSUPPORTED with the reason, never a crash.
+ */
+class TracerouteProbe(
+    private val targetHost: String = "1.1.1.1",
+    private val maxHops: Int = 6,
+) : Probe {
+    override val type = TestType.TRACEROUTE_UDP4
+    override val tier = Tier.APP
+    // Validated wall clock is ~250 ms on a healthy path; the ceiling is maxHops silent hops at
+    // 900 ms each, which only a blackholing path produces.
+    override val estimatedMs = 1_500L
+
+    private companion object {
+        const val BASE_PORT = 33434
+        /** ICMP errors take one RTT to surface on the errqueue; poll briefly, never block. */
+        const val HOP_DEADLINE_NS = 900_000_000L
+        const val POLL_INTERVAL_MS = 40L
+    }
+
+    override suspend fun run(ctx: Context, ids: ProbeIds): Test = withContext(Dispatchers.IO) {
+        val b = TestBuilder(type, tier, ids)
+        val params = buildJsonObject {
+            put("target", targetHost); put("max_hops", maxHops); put("base_port", BASE_PORT)
+        }
+
+        val api = ErrqueueApi.resolve()
+            ?: return@withContext b.build(
+                TestStatus.UNSUPPORTED,
+                params = params,
+                error = TestError(
+                    "no_recvmsg",
+                    "StructMsghdr/Os.recvmsg not on this API level — errqueue unreadable",
+                ),
+            )
+
+        var fd: FileDescriptor? = null
+        try {
+            fd = Os.socket(OsConstants.AF_INET, OsConstants.SOCK_DGRAM, OsConstants.IPPROTO_UDP)
+            OsAbi.trySetIntOpt(fd, OsConstants.IPPROTO_IP, OsAbi.IP_RECVERR, 1)?.let {
+                return@withContext b.build(
+                    TestStatus.UNSUPPORTED,
+                    params = params,
+                    error = TestError("ip_recverr_rejected", it),
+                )
+            }
+            val target = InetAddress.getByName(targetHost)
+
+            val hops = ArrayList<Hop>(maxHops)
+            var hopsSeen = 0
+            var reachedTarget = false
+            var srcPort = 0
+            for (ttl in 1..maxHops) {
+                OsAbi.trySetIntOpt(fd, OsConstants.IPPROTO_IP, OsAbi.IP_TTL, ttl)
+                val t0 = System.nanoTime()
+                val sent = runCatching {
+                    Os.sendto(fd, ByteArray(32), 0, 32, 0, target, BASE_PORT + ttl)
+                }
+                if (sent.isFailure) {
+                    hops.add(Hop(ttl, listOf(HopProbe(icmp = "sendto failed: " +
+                        (sent.exceptionOrNull()?.message ?: "?")))))
+                    continue
+                }
+                if (srcPort == 0) {
+                    // Only readable after the implicit bind the first send performs.
+                    srcPort = runCatching {
+                        (Os.getsockname(fd) as? InetSocketAddress)?.port ?: 0
+                    }.getOrDefault(0)
+                }
+
+                var hop: ErrqueueApi.ErrEvent? = null
+                val deadline = System.nanoTime() + HOP_DEADLINE_NS
+                while (hop == null && System.nanoTime() < deadline) {
+                    hop = api.pollErrqueue(fd)
+                    if (hop == null) delay(POLL_INTERVAL_MS)
+                }
+                val rttNs = System.nanoTime() - t0
+                when {
+                    hop == null -> hops.add(Hop(ttl, listOf(HopProbe()))) // silent hop: all null
+                    hop.parseError != null ->
+                        hops.add(Hop(ttl, listOf(HopProbe(icmp = "unparsed: ${hop.parseError}"))))
+                    else -> {
+                        hopsSeen++
+                        hops.add(Hop(ttl, listOf(HopProbe(
+                            replyFrom = hop.offender,
+                            rttNs = rttNs,
+                            icmp = when (hop.icmpType) {
+                                OsAbi.ICMP_TIME_EXCEEDED -> "time_exceeded"
+                                OsAbi.ICMP_DEST_UNREACH -> "dest_unreachable"
+                                else -> "type_${hop.icmpType}"
+                            },
+                        ))))
+                        if (hop.icmpType == OsAbi.ICMP_DEST_UNREACH) reachedTarget = true
+                    }
+                }
+                if (reachedTarget) break
+            }
+
+            // dst_port varies per TTL (classic traceroute, and what was validated on hardware),
+            // so this flow is explicitly NOT fixed-tuple; base_port is in params.
+            val evidence = TracerouteEvidence(
+                flow = Flow(srcPort = srcPort, dstPort = BASE_PORT, fixedTuple = false),
+                hops = hops,
+            ).toEvidence()
+            val metrics = buildJsonObject {
+                put("hops_seen", hopsSeen)
+                put("reached_target", reachedTarget)
+            }
+            val status = when {
+                hopsSeen > 0 -> TestStatus.OK
+                // API present, sends succeeded, nothing surfaced: a fact about this path or
+                // kernel, not proof the mechanism is missing.
+                else -> TestStatus.PARTIAL
+            }
+            b.build(status, params = params, evidence = evidence, metrics = metrics)
+        } catch (e: Throwable) {
+            b.build(
+                TestStatus.FAILED,
+                params = params,
+                error = TestError("uncaught", e.message ?: e.javaClass.simpleName),
+            )
+        } finally {
+            fd?.let { runCatching { Os.close(it) } }
+        }
+    }
+}
+
+/**
+ * Reflection facade over android.system.{StructMsghdr, StructCmsghdr, Os.recvmsg}.
+ * Resolved once; null if any piece is missing on this API level.
+ */
+internal class ErrqueueApi private constructor(
+    private val msghdrCtor: java.lang.reflect.Constructor<*>,
+    private val recvmsg: java.lang.reflect.Method,
+    private val cmsgLevel: java.lang.reflect.Field,
+    private val cmsgType: java.lang.reflect.Field,
+    private val cmsgData: java.lang.reflect.Field,
+    private val msgControl: java.lang.reflect.Field,
+) {
+    class ErrEvent(
+        val offender: String?,
+        val icmpType: Int,
+        val origin: Int,
+        val parseError: String? = null,
+    )
+
+    /** One non-blocking MSG_ERRQUEUE read; null when the queue is empty. */
+    fun pollErrqueue(fd: FileDescriptor): ErrEvent? {
+        return try {
+            val iov = arrayOf(ByteBuffer.allocate(512))
+            // (SocketAddress msg_name, ByteBuffer[] msg_iov, StructCmsghdr[] msg_control, flags)
+            val msghdr = msghdrCtor.newInstance(
+                InetSocketAddress(0), iov, null, 0,
+            )
+            recvmsg.invoke(null, fd, msghdr, OsAbi.MSG_ERRQUEUE or OsAbi.MSG_DONTWAIT)
+            val control = msgControl.get(msghdr) as? Array<*>
+                ?: return ErrEvent(null, -1, -1, "msg_control empty after recvmsg")
+            for (cmsg in control.filterNotNull()) {
+                val level = cmsgLevel.getInt(cmsg)
+                val type = cmsgType.getInt(cmsg)
+                if (level == OsConstants.IPPROTO_IP && type == OsAbi.IP_RECVERR) {
+                    return parseSockExtendedErr(cmsgData.get(cmsg))
+                }
+            }
+            ErrEvent(null, -1, -1, "no IP_RECVERR cmsg among ${control.size}")
+        } catch (e: Throwable) {
+            // The single most load-bearing line: reflection wraps errno in
+            // InvocationTargetException, and EAGAIN there means "queue empty", not failure.
+            val cause = (e as? java.lang.reflect.InvocationTargetException)?.cause ?: e
+            val msg = cause.message ?: cause.javaClass.simpleName
+            if ("EAGAIN" in msg || "EWOULDBLOCK" in msg) null
+            else ErrEvent(null, -1, -1, msg)
+        }
+    }
+
+    /** cmsg_data = struct sock_extended_err + offender sockaddr_in (see OsAbi). */
+    private fun parseSockExtendedErr(data: Any?): ErrEvent {
+        val bytes: ByteArray = when (data) {
+            is ByteArray -> data
+            is ByteBuffer -> ByteArray(data.remaining()).also { data.duplicate().get(it) }
+            else -> return ErrEvent(null, -1, -1, "cmsg_data is ${data?.javaClass?.name}")
+        }
+        if (bytes.size < OsAbi.SOCK_EE_SIZE) {
+            return ErrEvent(null, -1, -1, "cmsg_data too short: ${bytes.size}")
+        }
+        val origin = bytes[4].toInt() and 0xFF
+        val icmpType = bytes[5].toInt() and 0xFF
+        // SO_EE_OFFENDER: sockaddr_in directly after the fixed struct; family is in native
+        // byte order, sin_addr at offset +4 within the sockaddr.
+        val offender = if (bytes.size >= OsAbi.SOCK_EE_SIZE + 8) {
+            val family = ByteBuffer.wrap(bytes, OsAbi.SOCK_EE_SIZE, 2)
+                .order(ByteOrder.nativeOrder()).short.toInt()
+            if (family == OsConstants.AF_INET) {
+                val a = bytes.copyOfRange(OsAbi.SOCK_EE_SIZE + 4, OsAbi.SOCK_EE_SIZE + 8)
+                InetAddress.getByAddress(a).hostAddress
+            } else null
+        } else null
+        return ErrEvent(offender, icmpType, origin)
+    }
+
+    companion object {
+        fun resolve(): ErrqueueApi? = runCatching {
+            val msghdrCls = Class.forName("android.system.StructMsghdr")
+            val cmsghdrCls = Class.forName("android.system.StructCmsghdr")
+            ErrqueueApi(
+                // Picked by shape, not by position: the 4-arg form is
+                // (SocketAddress, ByteBuffer[], StructCmsghdr[], int) on every level that has it.
+                msghdrCtor = msghdrCls.constructors.first { it.parameterCount == 4 },
+                recvmsg = Os::class.java.getMethod(
+                    "recvmsg", FileDescriptor::class.java, msghdrCls, Int::class.javaPrimitiveType,
+                ),
+                cmsgLevel = cmsghdrCls.getField("cmsg_level"),
+                cmsgType = cmsghdrCls.getField("cmsg_type"),
+                cmsgData = cmsghdrCls.getField("cmsg_data"),
+                msgControl = msghdrCls.getField("msg_control"),
+            )
+        }.getOrNull()
+    }
+}

echolot-app/core-probe/src/main/kotlin/app/echo_lot/probe/V6ConnectProbe.kt

@@ -0,0 +1,131 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package app.echo_lot.probe
+
+import android.content.Context
+import app.echo_lot.measurement.Test
+import app.echo_lot.measurement.TestStatus
+import app.echo_lot.measurement.TestType
+import app.echo_lot.measurement.Tier
+import app.echo_lot.measurement.Network as MNetwork
+import kotlinx.coroutines.Dispatchers
+import kotlinx.coroutines.withContext
+import kotlinx.serialization.json.JsonObject
+import kotlinx.serialization.json.buildJsonObject
+import kotlinx.serialization.json.put
+import java.net.Inet6Address
+import java.net.InetSocketAddress
+import java.util.Locale
+
+/**
+ * v6.brokenness — does IPv6 actually carry traffic, asked with a real TCP connection.
+ *
+ * This exists to corroborate (or refute) the ICMPv6 silence that icmp.ping6 observes. ICMPv6 echo
+ * is widely filtered on networks where IPv6 works fine, so silence alone cannot distinguish
+ * "IPv6 is broken" from "ping is filtered" — a phone that reported v6.broken while happily
+ * loading IPv6-only sites is what proved the point. A TCP connect over IPv6 to the configured
+ * server settles it: if it succeeds, IPv6 works and the ICMP silence is filtering; if it fails
+ * too, on a network that advertises IPv6, the brokenness claim finally has evidence behind it.
+ *
+ * Only networks that claim to offer IPv6 (a global address or a v6 default route) are attempted:
+ * connecting over v6 on an IPv4-only network fails by design, and recording that as evidence
+ * would manufacture the exact false positive this probe exists to kill.
+ */
+class V6ConnectProbe(
+    private val entries: List<NetworkInventory.Entry>,
+    private val serverHost: String,
+    private val port: Int = 443,
+) : Probe {
+    override val type = TestType.V6_BROKENNESS
+    override val tier = Tier.APP
+    // One 3s connect timeout per v6-provisioned network, at most.
+    override val estimatedMs = 4_000L
+
+    override suspend fun run(ctx: Context, ids: ProbeIds): Test = withContext(Dispatchers.IO) {
+        val b = TestBuilder(type, tier, ids)
+        // Same rule as the STUN and canary probes: with no server there is no target, and
+        // borrowing someone else's infrastructure to get one is not this app's call to make.
+        if (serverHost.isBlank()) {
+            return@withContext b.build(
+                TestStatus.SKIPPED,
+                evidence = buildJsonObject { put("reason", "no server configured to connect to") },
+            )
+        }
+        val candidates = entries.filter { ipv6Provisioned(it.model) }
+        if (candidates.isEmpty()) {
+            return@withContext b.build(
+                TestStatus.SKIPPED,
+                evidence = buildJsonObject {
+                    put("reason", "no active network claims to offer IPv6")
+                },
+            )
+        }
+
+        var okCount = 0
+        var attemptedCount = 0
+        val evidence: JsonObject = buildJsonObject {
+            put("target", "$serverHost:$port")
+            for (e in candidates) {
+                val label = "${e.model.transport.name.lowercase()}:${e.model.id}"
+                val a = attempt(e)
+                if (a.attempted) attemptedCount++
+                if (a.ok) okCount++
+                put(label, buildJsonObject {
+                    put("network_ref", e.model.id)
+                    put("ok", a.ok)
+                    put("attempted", a.attempted)
+                    put("detail", a.detail)
+                })
+            }
+        }
+        val status = when {
+            attemptedCount == 0 -> TestStatus.SKIPPED // resolution/binding never got that far
+            okCount == attemptedCount -> TestStatus.OK
+            okCount > 0 -> TestStatus.PARTIAL
+            else -> TestStatus.FAILED
+        }
+        b.build(status, evidence = evidence)
+    }
+
+    /** Same attempted/ok separation as IcmpProbe: a connect we never sent proves nothing. */
+    private data class Attempt(val ok: Boolean, val attempted: Boolean, val detail: String)
+
+    private fun attempt(e: NetworkInventory.Entry): Attempt {
+        // Resolved through this network's own resolver; a v6 address obtained over another
+        // network would still be connected to over this one, which is what matters.
+        val addr = runCatching {
+            e.handle.getAllByName(serverHost).filterIsInstance<Inet6Address>().firstOrNull()
+        }.getOrNull()
+            ?: return Attempt(false, false, "no AAAA answer for $serverHost via this network")
+
+        // createSocket() binds to the network at creation; failing here means the app could not
+        // use the interface at all (e.g. EPERM under a VPN) — nothing was sent, nothing is known.
+        val socket = try {
+            e.handle.socketFactory.createSocket()
+        } catch (t: Throwable) {
+            return Attempt(false, false, "socket unavailable: ${t.message ?: t.javaClass.simpleName}")
+        }
+        return try {
+            val t0 = System.nanoTime()
+            socket.connect(InetSocketAddress(addr, port), 3000)
+            val rttMs = (System.nanoTime() - t0) / 1_000_000.0
+            Attempt(true, true, "connected to [${addr.hostAddress}]:$port " +
+                "rtt_ms=${"%.1f".format(Locale.ROOT, rttMs)}")
+        } catch (t: Throwable) {
+            // A refused connection would still prove the path forwards IPv6, but against our own
+            // server's 443 the realistic failures are timeout and unreachable — both silence.
+            Attempt(false, true, "error: ${t.message ?: t.javaClass.simpleName}")
+        } finally {
+            runCatching { socket.close() }
+        }
+    }
+
+    /** The network claims IPv6: a global (non-link-local) address or a v6 default route. */
+    private fun ipv6Provisioned(n: MNetwork): Boolean =
+        n.link.addresses.any { a ->
+            a.addr.contains(':') &&
+                !a.addr.startsWith("fe80", ignoreCase = true) &&
+                !a.addr.startsWith("::1")
+        } || n.link.routes.any { it.dst == "::/0" }
+}

echolot-app/core-protocol/src/main/kotlin/app/echo_lot/protocol/ControlClient.kt

@@ -35,13 +35,57 @@ class ControlClient(
     private val controlUrl: String,
     pins: Set<String>,
     private val appVersion: String = "",
+    /**
+     * Addresses to fall back to when the server's name will not resolve, learned from its profile.
+     *
+     * A measurement tool that cannot report from a broken network is useless exactly when it
+     * matters, and a wedged resolver is one of the faults it is built to find — it should not also
+     * be the thing that stops the finding being delivered.
+     *
+     * Safe because the pin is the trust and the name is not part of it: the server presents the
+     * same certificate whether it was reached by name or by address, and a wrong address fails the
+     * pin like anything else would.
+     */
+    private val fallbackAddrs: List<String> = emptyList(),
 ) {
 
     private val json = Json { ignoreUnknownKeys = true }
     private val socketFactory = Pinning.sslContext(pins).socketFactory
 
+    /**
+     * The base URL to use, substituting a cached address only when the name genuinely fails.
+     *
+     * Resolved once per client and only on failure, so a working network pays nothing and never
+     * silently drifts onto an address that may be stale.
+     */
+    private val base: String by lazy { resolveBase() }
+
+    private fun resolveBase(): String {
+        if (fallbackAddrs.isEmpty()) return controlUrl
+        val uri = runCatching { java.net.URI(controlUrl) }.getOrNull() ?: return controlUrl
+        val host = uri.host ?: return controlUrl
+        if (runCatching { java.net.InetAddress.getByName(host) }.isSuccess) return controlUrl
+
+        val port = if (uri.port > 0) uri.port else 443
+        for (ip in fallbackAddrs) {
+            // Checked rather than assumed: on a v4-only network a v6 address would otherwise be
+            // chosen and fail slowly, which is the wrong answer delivered late.
+            val reachable = runCatching {
+                java.net.Socket().use { sock ->
+                    sock.connect(java.net.InetSocketAddress(ip, port), 4000)
+                    true
+                }
+            }.getOrDefault(false)
+            if (reachable) {
+                val literal = if (ip.contains(':')) "[$ip]" else ip
+                return uri.scheme + "://" + literal + ":" + port
+            }
+        }
+        return controlUrl
+    }
+
     private fun open(path: String, method: String, credential: String?): HttpsURLConnection {
-        val conn = URL(controlUrl.trimEnd('/') + path).openConnection() as HttpsURLConnection
+        val conn = URL(base.trimEnd('/') + path).openConnection() as HttpsURLConnection
         conn.sslSocketFactory = socketFactory
         conn.setHostnameVerifier { _, _ -> true } // pin is the trust, not the name
         conn.requestMethod = method
@@ -184,6 +228,33 @@ class ControlClient(
         open("/v1/runs/$runId", "DELETE", credential).responseCode
     }
 
+    /**
+     * Ties this device to the person the ID token identifies.
+     *
+     * The device credential proves *which device*, the token proves *which person*; the server
+     * requires both. Returns the raw JSON reply (account id and display name).
+     */
+    fun linkAccount(credential: String, idToken: String): String {
+        val conn = open("/v1/account/link", "POST", credential)
+        writeJson(conn, """{"id_token":${jstr(idToken)}}""")
+        val text = body(conn)
+        check(conn.responseCode in 200..299) { "sign-in failed: ${conn.responseCode} $text" }
+        return text
+    }
+
+    /** Signs out on this device. The device stays enrolled. */
+    fun unlinkAccount(credential: String) {
+        open("/v1/account/link", "DELETE", credential).responseCode
+    }
+
+    /** Whether anyone is signed in on this device, and who. */
+    fun accountStatus(credential: String): String {
+        val conn = open("/v1/account", "GET", credential)
+        val text = body(conn)
+        check(conn.responseCode == 200) { "account status failed: ${conn.responseCode} $text" }
+        return text
+    }
+
     fun observations(credential: String, sessionId: String): String {
         val conn = open("/v1/sessions/$sessionId/observations", "GET", credential)
         val text = body(conn)

echolot-app/core-protocol/src/main/kotlin/app/echo_lot/protocol/Enrollment.kt

@@ -45,11 +45,22 @@ data class EnrollmentLink(
      * the reason the pin travels in the link at all.
      */
     fun redeem(deviceName: String? = null, appVersion: String = ""): Enrolled {
-        val client = ControlClient(controlUrl, setOf(pin), appVersion)
+        // The link may name the server's public address rather than its control endpoint, so that
+        // a person is handed a name they recognise. Ask where to actually connect.
+        //
+        // Only the address comes from here. The pin still comes from the link, because a pin
+        // fetched over an ordinary TLS connection would be worth exactly what the certificate
+        // authorities are worth — and pinning exists to survive one the operator does not
+        // control, such as a root injected by corporate device management. An intercepted
+        // discovery can therefore send this device to the wrong host, where the pin will not
+        // match: an outage, not a compromise.
+        val endpoint = discover(controlUrl) ?: controlUrl
+        val client = ControlClient(endpoint, setOf(pin), appVersion)
         val response = client.enroll(token, deviceName)
         val profile = client.profile(response.credential)
         return Enrolled(
-            controlUrl = controlUrl,
+            controlUrl = endpoint,
+            publicUrl = controlUrl,
             pin = pin,
             credential = response.credential,
             deviceId = response.deviceId,
@@ -57,6 +68,29 @@ data class EnrollmentLink(
         )
     }
 
+    /**
+     * Asks a server where its control plane lives. Null when it does not say, or cannot be asked.
+     *
+     * Deliberately forgiving: a server that predates this, or one whose link already names the
+     * control endpoint directly, simply answers nothing and the link's own URL is used. Enrollment
+     * must not start failing because an optional lookup did.
+     */
+    private fun discover(publicUrl: String): String? = runCatching {
+        val conn = (java.net.URL(publicUrl.trimEnd('/') + "/v1/discover").openConnection()
+            as java.net.HttpURLConnection).apply {
+            connectTimeout = 8_000
+            readTimeout = 8_000
+            setRequestProperty("Accept", "application/json")
+        }
+        if (conn.responseCode !in 200..299) return null
+        val body = conn.inputStream.bufferedReader().use { it.readText() }
+        kotlinx.serialization.json.Json { ignoreUnknownKeys = true }
+            .parseToJsonElement(body)
+            .let { (it as kotlinx.serialization.json.JsonObject)["control_url"] }
+            ?.let { (it as kotlinx.serialization.json.JsonPrimitive).content }
+            ?.takeIf { it.isNotBlank() }
+    }.getOrNull()
+
     companion object {
         const val SCHEME = "echolot"
         const val HOST = "enroll"
@@ -111,7 +145,15 @@ data class EnrollmentLink(
 
 /** A server this device is now enrolled with, ready to be stored in settings. */
 data class Enrolled(
+    /** Where this device connects: the endpoint whose certificate the pin matches. */
     val controlUrl: String,
+    /**
+     * The address a person was given, kept for display.
+     *
+     * Shown instead of [controlUrl] because the endpoint is plumbing — it exists to select a
+     * certificate — while this is the name the operator handed out and would recognise.
+     */
+    val publicUrl: String,
     val pin: String,
     val credential: String,
     val deviceId: String,

echolot-app/core-protocol/src/main/kotlin/app/echo_lot/protocol/Model.kt

@@ -29,6 +29,15 @@ data class Target(
     val id: String,
     val ip4: String? = null,
     val ip6: String? = null,
+    /**
+     * The second address, which RFC 5780 behaviour discovery redirects to.
+     *
+     * Worth surfacing rather than treating as an implementation detail: a report that says "the
+     * server did not answer" means something different depending on which of its addresses was
+     * asked, and an operator reading one needs to be able to tell.
+     */
+    @SerialName("ip4_alt") val ip4Alt: String? = null,
+    @SerialName("ip6_alt") val ip6Alt: String? = null,
     @SerialName("udp_port") val udpPort: Int = 0,
     @SerialName("tcp_port") val tcpPort: Int = 0,
     @SerialName("stun_port") val stunPort: Int = 0,
@@ -74,6 +83,25 @@ data class CompatInfo(
     @SerialName("app_max") val appMax: String = "",
 )
 
+/**
+ * How to sign in to this server's identity provider, advertised so the app can offer the button
+ * only when there is something behind it — and drive the flow without anyone typing an issuer URL.
+ */
+@Serializable
+data class AuthInfo(
+    val enabled: Boolean = false,
+    val issuer: String = "",
+    @SerialName("client_id") val clientId: String = "",
+    val flow: String = "",
+    @SerialName("redirect_uri") val redirectUri: String = "",
+    val scopes: String = "openid profile email",
+    @SerialName("authorization_endpoint") val authorizationEndpoint: String = "",
+    @SerialName("token_endpoint") val tokenEndpoint: String = "",
+    @SerialName("end_session_endpoint") val endSessionEndpoint: String = "",
+    /** Present when the server has an issuer configured but could not reach it. */
+    @SerialName("discovery_error") val discoveryError: String? = null,
+)
+
 @Serializable
 data class Profile(
     @SerialName("profile_version") val profileVersion: Int = 0,
@@ -86,6 +114,7 @@ data class Profile(
     val pins: List<String> = emptyList(),
     val uploads: UploadPolicy = UploadPolicy(),
     val compat: CompatInfo = CompatInfo(),
+    val auth: AuthInfo = AuthInfo(),
 ) {
     fun supports(capability: String) = capability in capabilities
 }

echolot-app/core-protocol/src/main/kotlin/app/echo_lot/protocol/OidcLogin.kt

@@ -0,0 +1,145 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package app.echo_lot.protocol
+
+import kotlinx.serialization.json.Json
+import kotlinx.serialization.json.jsonObject
+import kotlinx.serialization.json.jsonPrimitive
+import java.io.IOException
+import java.net.HttpURLConnection
+import java.net.URL
+import java.net.URLEncoder
+import java.security.MessageDigest
+import java.security.SecureRandom
+import java.util.Base64
+
+/**
+ * Sign-in for the app: authorization code with PKCE (RFC 7636).
+ *
+ * The app is a *public* client — it ships to devices, so any secret compiled into it can be read
+ * out of the APK with `unzip` and `strings`. PKCE is what replaces the client secret, and it
+ * defends a specific attack that matters here more than most places: the redirect comes back
+ * through a custom URI scheme, and on Android *any* app may register `echolot://`. A malicious one
+ * could intercept the callback and take the authorization code. Because the code can only be
+ * exchanged by presenting the verifier — which never left this process and cannot be derived from
+ * the challenge that did — a stolen code is worth nothing.
+ *
+ * Nothing from the IdP is kept afterwards. The ID token is used once, to prove to the server who
+ * is signing in, and then discarded: the device credential is what authenticates every later
+ * request. So there are no access tokens to store, no refresh tokens to rotate, and no token
+ * lifetime for the app to manage.
+ */
+object OidcLogin {
+
+    /** A started sign-in. [verifier] and [state] must survive until the callback returns. */
+    data class Pending(val authorizationUrl: String, val verifier: String, val state: String)
+
+    /**
+     * Builds the authorization URL and the secrets that must be held until the callback.
+     *
+     * Everything comes from the server's profile rather than being compiled in, so pointing the
+     * app at a different server with a different IdP is configuration, not a rebuild.
+     */
+    fun begin(auth: AuthInfo, random: SecureRandom = SecureRandom()): Pending {
+        require(auth.enabled && auth.authorizationEndpoint.isNotBlank()) {
+            "this server has no identity provider configured"
+        }
+        val verifier = randomUrlSafe(random)
+        val state = randomUrlSafe(random)
+        val challenge = b64(MessageDigest.getInstance("SHA-256").digest(verifier.toByteArray()))
+
+        val q = buildString {
+            append("response_type=code")
+            append("&client_id=").append(enc(auth.clientId))
+            append("&redirect_uri=").append(enc(auth.redirectUri))
+            append("&scope=").append(enc(auth.scopes))
+            append("&state=").append(enc(state))
+            append("&code_challenge=").append(enc(challenge))
+            append("&code_challenge_method=S256")
+        }
+        val sep = if (auth.authorizationEndpoint.contains('?')) "&" else "?"
+        return Pending(auth.authorizationEndpoint + sep + q, verifier, state)
+    }
+
+    /** What came back on the `echolot://auth` redirect. */
+    data class Callback(val code: String?, val state: String?, val error: String?)
+
+    /** Parses the redirect URI the browser handed back to the app. */
+    fun parseCallback(uri: String): Callback {
+        val q = uri.substringAfter('?', "")
+        var code: String? = null
+        var state: String? = null
+        var error: String? = null
+        for (pair in q.split('&')) {
+            val k = pair.substringBefore('=')
+            val v = dec(pair.substringAfter('=', ""))
+            when (k) {
+                "code" -> code = v
+                "state" -> state = v
+                "error" -> error = v
+                "error_description" -> if (error != null) error = "$error: $v"
+            }
+        }
+        return Callback(code, state, error)
+    }
+
+    /** The sign-in failed in a way worth showing someone, rather than a transport error. */
+    class LoginFailed(message: String) : Exception(message)
+
+    /**
+     * Exchanges the code for an ID token.
+     *
+     * The state is compared before anything else happens. A callback whose state does not match
+     * the one this process generated did not come from a flow this process started — which is
+     * precisely how an attacker gets a victim to complete *their* login — so it is refused before
+     * the code is spent.
+     */
+    fun complete(auth: AuthInfo, pending: Pending, callbackUri: String): String {
+        val cb = parseCallback(callbackUri)
+        if (cb.error != null) throw LoginFailed(cb.error)
+        if (cb.state.isNullOrEmpty() || cb.state != pending.state) {
+            throw LoginFailed("this sign-in did not start on this device — start again")
+        }
+        val code = cb.code ?: throw LoginFailed("the identity provider returned no authorization code")
+
+        val body = buildString {
+            append("grant_type=authorization_code")
+            append("&code=").append(enc(code))
+            append("&redirect_uri=").append(enc(auth.redirectUri))
+            append("&client_id=").append(enc(auth.clientId))
+            append("&code_verifier=").append(enc(pending.verifier))
+        }
+        val conn = (URL(auth.tokenEndpoint).openConnection() as HttpURLConnection).apply {
+            requestMethod = "POST"
+            doOutput = true
+            connectTimeout = 15_000
+            readTimeout = 15_000
+            setRequestProperty("Content-Type", "application/x-www-form-urlencoded")
+            setRequestProperty("Accept", "application/json")
+        }
+        conn.outputStream.use { it.write(body.toByteArray()) }
+        val text = try {
+            val stream = if (conn.responseCode in 200..299) conn.inputStream else conn.errorStream
+            stream?.bufferedReader()?.use { it.readText() } ?: ""
+        } catch (e: IOException) {
+            throw LoginFailed("could not reach the identity provider: ${e.message}")
+        }
+        if (conn.responseCode !in 200..299) {
+            throw LoginFailed("the identity provider refused the sign-in (${conn.responseCode})")
+        }
+        val idToken = runCatching {
+            Json.parseToJsonElement(text).jsonObject["id_token"]?.jsonPrimitive?.content
+        }.getOrNull()
+        return idToken?.takeIf { it.isNotBlank() }
+            ?: throw LoginFailed("the identity provider returned no id_token")
+    }
+
+    private fun randomUrlSafe(random: SecureRandom): String =
+        ByteArray(32).also(random::nextBytes).let(::b64)
+
+    private fun b64(b: ByteArray): String = Base64.getUrlEncoder().withoutPadding().encodeToString(b)
+    private fun enc(s: String): String = URLEncoder.encode(s, "UTF-8")
+    private fun dec(s: String): String =
+        runCatching { java.net.URLDecoder.decode(s, "UTF-8") }.getOrDefault(s)
+}

echolot-app/core-protocol/src/main/kotlin/app/echo_lot/protocol/ProbeSession.kt

@@ -109,6 +109,192 @@ class ProbeSession(
         return out
     }
 
+    /**
+     * Sends paced upstream traffic for [durationMs] and reports what was put on the wire.
+     *
+     * Paced rather than flat out, for the same reason the server paces: an unpaced burst measures
+     * the local NIC and the first queue it meets, then collapses into loss that reads as a network
+     * fault. The schedule is absolute rather than sleep-per-packet, which accumulates the
+     * scheduler's error and drifts the achieved rate below target over a multi-second run.
+     *
+     * Nothing comes back — the server counts and stays silent — so the result here is only the
+     * send side. The measurement is the gap between this and the server's tally.
+     */
+    fun sendThroughput(durationMs: Long, kbps: Int, sizeBytes: Int = 1200): Sent {
+        val size = sizeBytes.coerceIn(Wire.HEADER_SIZE + 16, 1472)
+        val payload = ByteArray(size - Wire.HEADER_SIZE)
+        val perPacketNs = (size.toLong() * 8 * 1_000_000 / kbps.coerceAtLeast(1)).coerceAtLeast(1_000)
+
+        val start = System.nanoTime()
+        val deadline = start + durationMs * 1_000_000
+        var next = start
+        var packets = 0
+        var bytes = 0L
+        while (System.nanoTime() < deadline) {
+            val pkt = Wire.build(Wire.TYPE_THROUGHPUT_UP, prefix, ++seq, nowNs(), key, payload)
+            try {
+                socket.send(DatagramPacket(pkt, pkt.size, server))
+            } catch (e: java.io.IOException) {
+                // A local send failure is our condition, not the path's. Stop and report what
+                // actually left, rather than counting the remainder as loss on the network.
+                break
+            }
+            packets++
+            bytes += pkt.size
+            next += perPacketNs
+            val sleepNs = next - System.nanoTime()
+            if (sleepNs > 0) Thread.sleep(sleepNs / 1_000_000, (sleepNs % 1_000_000).toInt())
+        }
+        val elapsedMs = (System.nanoTime() - start) / 1_000_000
+        return Sent(packets, bytes, elapsedMs, if (elapsedMs > 0) (bytes * 8 / elapsedMs).toInt() else 0)
+    }
+
+    /** What one upstream run put on the wire locally. */
+    data class Sent(val packets: Int, val bytes: Long, val durationMs: Long, val kbps: Int)
+
+    // ---- upstream trains (spec §3.2, types 0x03-0x05) --------------------------------------
+
+    /** One TRAIN_DATA packet as sent: its wire seq, local tx time and size. */
+    data class TrainPacket(val seq: Int, val tTxNs: Long, val sizeBytes: Int)
+
+    /** One row of the server's received view. 255 in ttl/dscp/ecn means "not observed". */
+    data class TrainRow(
+        val seq: Int, val tRxNs: Long, val sizeBytes: Int,
+        val ttl: Int, val dscp: Int, val ecn: Int,
+    )
+
+    /**
+     * The server's account of one train. [received] counts every packet that arrived, buffered
+     * or not; [truncated] mirrors the wire flag (rows beyond the server's cap were counted but
+     * not kept). [partsExpected]/[partsReceived] make a lossy report path visible instead of
+     * letting missing rows masquerade as train loss.
+     */
+    data class TrainReport(
+        val trainId: Int,
+        val received: Int,
+        val truncated: Boolean,
+        val rows: List<TrainRow>,
+        val partsExpected: Int,
+        val partsReceived: Int,
+    )
+
+    /**
+     * Sends one paced upstream train. Nothing comes back per packet by design; pair with
+     * [trainReport] to learn what arrived. The absolute schedule (not sleep-per-packet) is the
+     * same anti-drift choice as [sendThroughput].
+     */
+    fun sendTrain(trainId: Int, count: Int, sizeBytes: Int = 200, interPacketMs: Long = 5): List<TrainPacket> {
+        val size = sizeBytes.coerceIn(Wire.HEADER_SIZE + 4, 1472)
+        val out = ArrayList<TrainPacket>(count)
+        val start = System.nanoTime()
+        var next = start
+        for (i in 0 until count) {
+            val payload = ByteArray(size - Wire.HEADER_SIZE)
+            payload[0] = (trainId ushr 24).toByte(); payload[1] = (trainId ushr 16).toByte()
+            payload[2] = (trainId ushr 8).toByte(); payload[3] = trainId.toByte()
+            val tTx = nowNs()
+            val pkt = Wire.build(Wire.TYPE_TRAIN_DATA, prefix, ++seq, tTx, key, payload)
+            try {
+                socket.send(DatagramPacket(pkt, pkt.size, server))
+            } catch (e: java.io.IOException) {
+                // A local send failure is our condition, not the path's: report what actually
+                // left rather than letting the server's report read as loss.
+                break
+            }
+            out.add(TrainPacket(seq, tTx, pkt.size))
+            next += interPacketMs * 1_000_000
+            val sleepNs = next - System.nanoTime()
+            if (sleepNs > 0) Thread.sleep(sleepNs / 1_000_000, (sleepNs % 1_000_000).toInt())
+        }
+        return out
+    }
+
+    /**
+     * Fetches the server's received view of a train (one REPORT_REQ, N REPORT datagrams).
+     *
+     * Returns null when no report arrives at all — indistinguishable between "report lost" and
+     * "server predates trains", and the caller must say so rather than choose. Missing parts of
+     * a multi-part report are tolerated and visible via partsReceived < partsExpected.
+     */
+    fun trainReport(trainId: Int, timeoutMs: Long = 3_000): TrainReport? {
+        val req = ByteArray(4)
+        req[0] = (trainId ushr 24).toByte(); req[1] = (trainId ushr 16).toByte()
+        req[2] = (trainId ushr 8).toByte(); req[3] = trainId.toByte()
+        val pkt = Wire.build(Wire.TYPE_TRAIN_REPORT_REQ, prefix, ++seq, nowNs(), key, req)
+        socket.send(DatagramPacket(pkt, pkt.size, server))
+
+        var received = 0
+        var truncated = false
+        var partsExpected = -1
+        val seenParts = HashSet<Int>()
+        val rows = ArrayList<TrainRow>()
+        val deadline = System.nanoTime() + timeoutMs * 1_000_000
+        val buf = ByteArray(2048)
+        val prevTimeout = socket.soTimeout
+        try {
+            while (partsExpected < 0 || seenParts.size < partsExpected) {
+                val remainMs = ((deadline - System.nanoTime()) / 1_000_000).toInt()
+                if (remainMs <= 0) break
+                socket.soTimeout = remainMs.coerceAtMost(1000)
+                val dp = DatagramPacket(buf, buf.size)
+                try {
+                    socket.receive(dp)
+                } catch (e: java.net.SocketTimeoutException) {
+                    continue
+                }
+                val p = Wire.parseVerified(buf, dp.length, key) ?: continue
+                if (p.type != Wire.TYPE_TRAIN_REPORT) continue
+                val part = parseReportPart(p.payload, trainId) ?: continue
+                if (!seenParts.add(part.part)) continue
+                received = part.received
+                truncated = truncated || part.truncated
+                partsExpected = part.parts
+                rows.addAll(part.rows)
+            }
+        } finally {
+            socket.soTimeout = prevTimeout
+        }
+        if (seenParts.isEmpty()) return null
+        rows.sortBy { it.seq }
+        return TrainReport(trainId, received, truncated, rows, partsExpected, seenParts.size)
+    }
+
+    private class ReportPart(
+        val part: Int, val parts: Int, val received: Int,
+        val truncated: Boolean, val rows: List<TrainRow>,
+    )
+
+    /** Mirrors the server's columnar layout (dataplane/train.go buildTrainReport). */
+    private fun parseReportPart(b: ByteArray, wantId: Int): ReportPart? {
+        if (b.size < 16) return null
+        fun u16(off: Int) = ((b[off].toInt() and 0xFF) shl 8) or (b[off + 1].toInt() and 0xFF)
+        fun u32(off: Int) = ((b[off].toLong() and 0xFF) shl 24) or ((b[off + 1].toLong() and 0xFF) shl 16) or
+            ((b[off + 2].toLong() and 0xFF) shl 8) or (b[off + 3].toLong() and 0xFF)
+        if (u32(0).toInt() != wantId) return null
+        val received = u32(4).toInt()
+        val part = u16(8)
+        val parts = u16(10)
+        val truncated = (b[12].toInt() and 0x01) != 0
+        val n = u16(14)
+        if (b.size < 16 + n * 17) return null
+        val rows = ArrayList<TrainRow>(n)
+        var off = 16
+        val seqs = IntArray(n) { u32(off + it * 4).toInt() }; off += n * 4
+        val tRx = LongArray(n) {
+            var v = 0L
+            for (j in 0 until 8) v = (v shl 8) or (b[off + it * 8 + j].toLong() and 0xFF)
+            v
+        }; off += n * 8
+        val sizes = IntArray(n) { u16(off + it * 2) }; off += n * 2
+        val ttls = IntArray(n) { b[off + it].toInt() and 0xFF }; off += n
+        val dscps = IntArray(n) { b[off + it].toInt() and 0xFF }; off += n
+        val ecns = IntArray(n) { b[off + it].toInt() and 0xFF }
+        for (i in 0 until n) {
+            rows.add(TrainRow(seqs[i], tRx[i], sizes[i], ttls[i], dscps[i], ecns[i]))
+        }
+        return ReportPart(part, parts, received, truncated, rows)
+    }
+
     /** One packet received from the server, with the wire size actually delivered. */
     data class Received(val type: Int, val seq: Int, val sizeBytes: Int, val tRxNs: Long)
 

echolot-app/core-protocol/src/main/kotlin/app/echo_lot/protocol/Wire.kt

@@ -23,6 +23,14 @@ object Wire {
 
     const val TYPE_ECHO_REQ: Int = 0x01
     const val TYPE_ECHO_RESP: Int = 0x02
+    /**
+     * Upstream train (spec §3.2): DATA is deliberately unanswered — a per-packet reply would
+     * double the traffic and drag the return path into a measurement of the outbound one. The
+     * server's received view comes back afterwards via REPORT_REQ → one or more REPORTs.
+     */
+    const val TYPE_TRAIN_DATA: Int = 0x03
+    const val TYPE_TRAIN_REPORT_REQ: Int = 0x04
+    const val TYPE_TRAIN_REPORT: Int = 0x05
     const val TYPE_TIMESYNC_REQ: Int = 0x07
     const val TYPE_TIMESYNC_RSP: Int = 0x08
     const val TYPE_MTU_PROBE: Int = 0x09
@@ -38,6 +46,16 @@ object Wire {
      */
     const val TYPE_FRAG_DATA: Int = 0x0D
 
+    /** One packet of a sustained-rate downstream run. */
+    const val TYPE_THROUGHPUT_DATA: Int = 0x0E
+
+    /**
+     * One packet of a client-driven upstream run. The server counts it and does not answer:
+     * a reply would double the traffic and drag the return path into a measurement that is
+     * specifically about the outbound one.
+     */
+    const val TYPE_THROUGHPUT_UP: Int = 0x0F
+
     /** The 8-byte on-the-wire prefix = first 16 hex chars of the session id, decoded. */
     fun wirePrefix(sessionId: String): ByteArray {
         require(sessionId.length >= 16) { "session id too short" }

echolot-app/core-protocol/src/test/kotlin/app/echo_lot/protocol/OidcLoginTest.kt

@@ -0,0 +1,100 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package app.echo_lot.protocol
+
+import java.security.SecureRandom
+import kotlin.test.Test
+import kotlin.test.assertEquals
+import kotlin.test.assertFailsWith
+import kotlin.test.assertNotEquals
+import kotlin.test.assertTrue
+
+class OidcLoginTest {
+
+    private val auth = AuthInfo(
+        enabled = true,
+        issuer = "https://id.example.net/application/o/echolot-app/",
+        clientId = "the-client",
+        redirectUri = "echolot://auth",
+        scopes = "openid profile email",
+        authorizationEndpoint = "https://id.example.net/application/o/authorize/",
+        tokenEndpoint = "https://id.example.net/application/o/token/",
+    )
+
+    @Test
+    fun theAuthorizationUrlCarriesEverythingTheIdPNeeds() {
+        val p = OidcLogin.begin(auth)
+        val url = p.authorizationUrl
+        assertTrue(url.startsWith(auth.authorizationEndpoint + "?"), url)
+        for (part in listOf(
+            "response_type=code",
+            "client_id=the-client",
+            "redirect_uri=echolot%3A%2F%2Fauth",
+            "code_challenge_method=S256",
+            "scope=openid+profile+email",
+        )) {
+            assertTrue(url.contains(part), "missing $part in $url")
+        }
+        assertTrue(url.contains("code_challenge="), url)
+        // The verifier itself must never appear in the URL — that is the entire point of PKCE.
+        assertTrue(!url.contains(p.verifier), "the code verifier leaked into the authorize URL")
+    }
+
+    // Two sign-ins must not share a verifier or state, or one intercepted flow compromises the next.
+    @Test
+    fun everySignInGetsFreshSecrets() {
+        val a = OidcLogin.begin(auth, SecureRandom())
+        val b = OidcLogin.begin(auth, SecureRandom())
+        assertNotEquals(a.verifier, b.verifier)
+        assertNotEquals(a.state, b.state)
+        assertTrue(a.verifier.length >= 43, "verifier is shorter than RFC 7636 allows")
+    }
+
+    @Test
+    fun parsesTheRedirectTheBrowserHandsBack() {
+        val cb = OidcLogin.parseCallback("echolot://auth?code=abc123&state=xyz")
+        assertEquals("abc123", cb.code)
+        assertEquals("xyz", cb.state)
+    }
+
+    @Test
+    fun parsesAnErrorRedirect() {
+        val cb = OidcLogin.parseCallback("echolot://auth?error=access_denied&error_description=User%20said%20no")
+        assertEquals("access_denied", cb.error?.substringBefore(":"))
+        assertTrue(cb.code == null)
+    }
+
+    // A callback whose state does not match is how an attacker gets someone to complete *their*
+    // sign-in. It must be refused before the code is spent, without any network call.
+    @Test
+    fun aMismatchedStateIsRefusedBeforeTheCodeIsSpent() {
+        val p = OidcLogin.begin(auth)
+        val e = assertFailsWith<OidcLogin.LoginFailed> {
+            OidcLogin.complete(auth, p, "echolot://auth?code=stolen&state=not-ours")
+        }
+        assertTrue(e.message!!.contains("did not start on this device"), e.message!!)
+    }
+
+    @Test
+    fun aMissingStateIsRefused() {
+        val p = OidcLogin.begin(auth)
+        assertFailsWith<OidcLogin.LoginFailed> {
+            OidcLogin.complete(auth, p, "echolot://auth?code=abc")
+        }
+    }
+
+    @Test
+    fun anErrorRedirectSurfacesTheReason() {
+        val p = OidcLogin.begin(auth)
+        val e = assertFailsWith<OidcLogin.LoginFailed> {
+            OidcLogin.complete(auth, p, "echolot://auth?error=access_denied&state=${p.state}")
+        }
+        assertTrue(e.message!!.contains("access_denied"))
+    }
+
+    @Test
+    fun refusesToStartWhenTheServerHasNoIdentityProvider() {
+        assertFailsWith<IllegalArgumentException> { OidcLogin.begin(AuthInfo(enabled = false)) }
+    }
+}

echolot-app/core-shizuku/build.gradle.kts

@@ -32,4 +32,8 @@ dependencies {
     implementation(libs.shizuku.provider)
     implementation(libs.kotlinx.coroutines.android)
     implementation(libs.kotlinx.serialization.json)
+    // JVM unit tests for the pure dump parsers (DumpParsers.kt) against the archived
+    // vendor fixtures — no device, no Android runtime.
+    testImplementation(libs.kotlin.test.junit)
+    testImplementation(libs.junit4)
 }

echolot-app/core-shizuku/src/main/kotlin/app/echo_lot/shizuku/DumpParsers.kt

@@ -0,0 +1,174 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package app.echo_lot.shizuku
+
+/**
+ * Parsed view of one IPv6 default route from `ip -6 route show table all`.
+ *
+ * [table] stays a string: Android's per-network route tables use ids past Int range (the Lenovo
+ * TB330FU prints `table 1000000015`), and `table local` is not a number at all — parsing to a
+ * numeric type either overflows or silently drops rows, and the id is only ever compared, never
+ * computed with.
+ */
+data class V6DefaultRoute(
+    /** Link-local address of the advertising router; null for gateway-less defaults (dummy0). */
+    val gateway: String?,
+    val dev: String,
+    val table: String?, // null = main table (`ip` omits the token there)
+    val proto: String?, // "ra" marks a route installed from a Router Advertisement
+    val metric: Long?,
+    /** Remaining RA route lifetime (`expires NNNsec`); null when the route does not age out. */
+    val expiresSec: Long?,
+)
+
+/** One `ip neigh show` row. [lladdr] is null for FAILED/INCOMPLETE entries — the kernel tried to
+ *  resolve and has nothing, which is itself signal. */
+data class NeighborEntry(
+    val ip: String,
+    val dev: String?,
+    val lladdr: String?,
+    val state: String?, // REACHABLE/STALE/FAILED/... — kept verbatim, the kernel's vocabulary
+    val router: Boolean,
+)
+
+/** A NEIGH transition seen inside the `ip monitor` window. */
+data class NeighborEvent(val entry: NeighborEntry, val deleted: Boolean)
+
+/**
+ * Pure-string parsers for the shell battery's `ip` command outputs. No Android imports on
+ * purpose: these run (and are unit-tested) on the JVM against the real vendor dumps archived
+ * from the prober, which is the only way to catch a vendor format drift before it ships.
+ *
+ * All parsers degrade to an empty result on missing or unrecognized input — the battery's
+ * captures are best-effort (the Lenovo's `ip monitor` times out under newProcess, the
+ * UserService path prepends a stray `uid=2000` line, evidence strings are trimmed mid-line
+ * at 1200 chars), so an exception here would turn a degraded capture into a lost test.
+ */
+object DumpParsers {
+
+    /** True when [raw] is real command output rather than an executor error sentinel. */
+    fun captureUsable(raw: String?): Boolean {
+        if (raw.isNullOrBlank()) return false
+        val t = raw.trimStart()
+        return !t.startsWith("SHIZUKU_") && !t.startsWith("EXEC_") && !t.startsWith("NEWPROCESS_")
+    }
+
+    /** Extracts every `default …` route from `ip -6 route show table all` output. */
+    fun parseV6DefaultRoutes(raw: String?): List<V6DefaultRoute> {
+        if (!captureUsable(raw)) return emptyList()
+        val routes = ArrayList<V6DefaultRoute>()
+        for (line in raw!!.lineSequence()) {
+            val tok = line.trim().split(WS)
+            if (tok.firstOrNull() != "default") continue
+            var gateway: String? = null; var dev: String? = null; var table: String? = null
+            var proto: String? = null; var metric: Long? = null; var expires: Long? = null
+            var i = 1
+            while (i < tok.size - 1) {
+                when (tok[i]) {
+                    "via" -> gateway = tok[i + 1]
+                    "dev" -> dev = tok[i + 1]
+                    "table" -> table = tok[i + 1]
+                    "proto" -> proto = tok[i + 1]
+                    "metric" -> metric = tok[i + 1].toLongOrNull()
+                    // `expires 1269sec` — the unit is glued to the number.
+                    "expires" -> expires = tok[i + 1].removeSuffix("sec").toLongOrNull()
+                }
+                i++
+            }
+            // A default route without a device is not something `ip` prints; treat it as a
+            // truncated/garbled line rather than fabricating a partial route.
+            if (dev != null) routes.add(V6DefaultRoute(gateway, dev, table, proto, metric, expires))
+        }
+        return routes
+    }
+
+    /**
+     * Maps interface name → link-layer address from `ip addr show`. Only `link/ether` counts:
+     * loopback/ipip/gre pseudo-addresses are not identities, and the RA-source cross-reference
+     * this feeds compares Ethernet MACs.
+     */
+    fun parseInterfaceMacs(raw: String?): Map<String, String> {
+        if (!captureUsable(raw)) return emptyMap()
+        val macs = LinkedHashMap<String, String>()
+        var current: String? = null
+        for (line in raw!!.lineSequence()) {
+            val header = STANZA_HEADER.find(line)
+            if (header != null) {
+                // "5: tunl0@NONE:" — the name is the part before an optional @suffix.
+                current = header.groupValues[1].substringBefore('@')
+                continue
+            }
+            val dev = current ?: continue
+            val tok = line.trim().split(WS)
+            if (tok.size >= 2 && tok[0] == "link/ether" && MAC.matches(tok[1])) {
+                macs.putIfAbsent(dev, tok[1])
+            }
+        }
+        return macs
+    }
+
+    /** Parses `ip neigh show` output into entries; non-neighbor lines (uid noise) are skipped. */
+    fun parseNeighbors(raw: String?): List<NeighborEntry> {
+        if (!captureUsable(raw)) return emptyList()
+        return raw!!.lineSequence()
+            .mapNotNull { parseNeighborTokens(it.trim().split(WS)) }
+            .toList()
+    }
+
+    /**
+     * Extracts NEIGH transitions from an `ip monitor all` capture. Each event line carries a
+     * `[NEIGH]` label (other families — ROUTE, ADDR, LINK — are ignored) and deletions are
+     * printed as `Deleted <entry>`. An empty result is normal: a quiet 5 s window sees nothing.
+     */
+    fun parseNeighborEvents(raw: String?): List<NeighborEvent> {
+        if (!captureUsable(raw)) return emptyList()
+        val events = ArrayList<NeighborEvent>()
+        for (line in raw!!.lineSequence()) {
+            val m = MONITOR_LABEL.find(line.trim()) ?: continue
+            if (!m.groupValues[1].equals("NEIGH", ignoreCase = true)) continue
+            var rest = line.trim().removeRange(m.range).trim()
+            val deleted = rest.startsWith("Deleted ", ignoreCase = true)
+            if (deleted) rest = rest.substring("Deleted ".length)
+            parseNeighborTokens(rest.split(WS))?.let { events.add(NeighborEvent(it, deleted)) }
+        }
+        return events
+    }
+
+    /**
+     * (ip → lladdr) for every neighbor that has one. This is the comparison surface the future
+     * gateway-MAC-change finding diffs across runs, so it is computed here — in the tested,
+     * pure layer — rather than re-derived from JSON by each consumer.
+     */
+    fun lladdrByIp(neighbors: List<NeighborEntry>): Map<String, String> =
+        neighbors.mapNotNull { n -> n.lladdr?.let { n.ip to it } }.toMap()
+
+    /** One neighbor row: `<ip> dev <if> [lladdr <mac>] [router] [proxy] <STATE>`. */
+    private fun parseNeighborTokens(tok: List<String>): NeighborEntry? {
+        val ip = tok.firstOrNull() ?: return null
+        // The first token must look like an address — this is what drops the UserService path's
+        // stray "uid=2000" line and any grep noise without needing to know every noise shape.
+        if (!IP_LIKE.matches(ip) || (!ip.contains('.') && !ip.contains(':'))) return null
+        var dev: String? = null; var lladdr: String? = null; var state: String? = null
+        var router = false
+        var i = 1
+        while (i < tok.size) {
+            when (tok[i]) {
+                "dev" -> { dev = tok.getOrNull(i + 1); i++ }
+                "lladdr" -> { lladdr = tok.getOrNull(i + 1); i++ }
+                "router" -> router = true
+                "proxy" -> {} // recorded nowhere: proxy entries have no bearing on ARP watching
+                else -> if (STATE.matches(tok[i])) state = tok[i]
+            }
+            i++
+        }
+        return NeighborEntry(ip, dev, lladdr, state, router)
+    }
+
+    private val WS = Regex("\\s+")
+    private val STANZA_HEADER = Regex("^\\d+:\\s+([^:\\s]+):")
+    private val MAC = Regex("^[0-9a-fA-F]{2}(:[0-9a-fA-F]{2}){5}$")
+    private val IP_LIKE = Regex("^[0-9a-fA-F:.]+(%[\\w-]+)?$")
+    private val STATE = Regex("^(REACHABLE|STALE|DELAY|PROBE|FAILED|INCOMPLETE|PERMANENT|NOARP|NONE)$")
+    private val MONITOR_LABEL = Regex("^\\[(\\w+)]")
+}

echolot-app/core-shizuku/src/main/kotlin/app/echo_lot/shizuku/ShizukuAvailability.kt

@@ -98,20 +98,32 @@ object ShizukuAvailability {
             .addFlags(android.content.Intent.FLAG_ACTIVITY_NEW_TASK)
 
     /**
-     * Reports the state now and on every binder transition. Returns a function that removes the
-     * listeners again (call it from onCleared).
+     * Reports the state now, on every binder transition, and when a permission request is
+     * answered. Returns a function that removes the listeners again (call it from onCleared).
+     *
+     * The permission listener matters as much as the binder ones: granting permission does not
+     * make the binder arrive or die, so without it the banner still read "running but not
+     * authorised" after the user had just authorised it — the one moment they are looking for
+     * confirmation that it worked.
+     *
+     * It is still not sufficient on its own. Permission can be granted inside Shizuku's own app,
+     * where nothing calls back into this process at all, so callers should re-check on resume as
+     * well; see [current].
      */
     fun observe(context: Context, onChange: (State) -> Unit): () -> Unit {
         val app = context.applicationContext
         val received = Shizuku.OnBinderReceivedListener { onChange(current(app)) }
         val dead = Shizuku.OnBinderDeadListener { onChange(current(app)) }
+        val permission = Shizuku.OnRequestPermissionResultListener { _, _ -> onChange(current(app)) }
         // "Sticky" fires immediately if the binder already arrived before we registered.
         runCatching { Shizuku.addBinderReceivedListenerSticky(received) }
         runCatching { Shizuku.addBinderDeadListener(dead) }
+        runCatching { Shizuku.addRequestPermissionResultListener(permission) }
         onChange(current(app))
         return {
             runCatching { Shizuku.removeBinderReceivedListener(received) }
             runCatching { Shizuku.removeBinderDeadListener(dead) }
+            runCatching { Shizuku.removeRequestPermissionResultListener(permission) }
         }
     }
 }

echolot-app/core-shizuku/src/main/kotlin/app/echo_lot/shizuku/ShizukuProbe.kt

@@ -10,15 +10,24 @@ import app.echo_lot.measurement.TestType
 import app.echo_lot.measurement.Tier
 import kotlinx.coroutines.Dispatchers
 import kotlinx.coroutines.withContext
+import kotlinx.serialization.json.JsonObject
 import kotlinx.serialization.json.buildJsonObject
 import kotlinx.serialization.json.put
+import kotlinx.serialization.json.putJsonArray
+import kotlinx.serialization.json.putJsonObject
+import kotlinx.serialization.json.addJsonObject
 
 /**
  * The Shizuku shell-tier probe: runs the privileged command battery (neighbor table, RA routes
  * with lifetimes, netlink monitor, IpClient DHCP logs, wifi dump) that the app UID cannot, and
- * captures the real per-device dump formats the production parsers must handle. Emitted as a
- * shizuku-tier `link.ip_monitor` test (the representative shell-tier link test); `exec_path`
- * records whether the UserService or the newProcess fallback carried it.
+ * captures the real per-device dump formats the production parsers must handle. Emits three
+ * shizuku-tier tests from the one battery:
+ *  - `link.ip_monitor` — the raw captures (the shell tier's ground truth), `exec_path` records
+ *    whether the UserService or the newProcess fallback carried it;
+ *  - `link.ra_source` — parsed from the v6 route table + `ip addr`: who advertises IPv6 here;
+ *  - `sec.arp_watch` — parsed from the neighbor table + monitor window: (ip → lladdr) pairs for
+ *    gateway-MAC-change detection.
+ * The battery runs once; the derived tests parse its captures, so they share its time window.
  */
 class ShizukuProbe {
     val type = TestType.LINK_IP_MONITOR
@@ -34,24 +43,34 @@ class ShizukuProbe {
         "wifi_dump" to "dumpsys wifi 2>/dev/null | grep -iA1 -m 20 -e 'mDhcpResults' -e 'Gateway' -e 'DNS' || true",
     )
 
-    /** Runs the battery and returns a Test. [uuid]/[monoNs] come from the run's id/clock source. */
-    suspend fun run(context: Context, uuid: () -> String, monoNs: () -> Long): Test = withContext(Dispatchers.IO) {
-        val id = uuid()
+    /**
+     * Runs the battery and returns the three tests, battery first. [uuid]/[monoNs] come from the
+     * run's id/clock source. When the shell tier is unavailable all three come back UNSUPPORTED —
+     * one silent test would leave the other two types missing from the document, which reads as
+     * "never attempted" rather than "tier absent".
+     */
+    suspend fun run(context: Context, uuid: () -> String, monoNs: () -> Long): List<Test> = withContext(Dispatchers.IO) {
         val started = monoNs()
         val runner = ShizukuRunner(context)
         val st = runner.status()
 
-        fun envelope(status: TestStatus, evidence: kotlinx.serialization.json.JsonObject, metrics: kotlinx.serialization.json.JsonObject? = null) =
-            Test(id = id, type = type, tier = tier, startedMonoNs = started, endedMonoNs = monoNs(),
+        fun envelope(type: String, status: TestStatus, evidence: JsonObject, metrics: JsonObject? = null) =
+            Test(id = uuid(), type = type, tier = tier, startedMonoNs = started, endedMonoNs = monoNs(),
                 status = status, evidence = evidence, metrics = metrics)
 
+        fun allUnsupported(evidence: JsonObject) = listOf(
+            envelope(TestType.LINK_IP_MONITOR, TestStatus.UNSUPPORTED, evidence),
+            envelope(TestType.LINK_RA_SOURCE, TestStatus.UNSUPPORTED, evidence),
+            envelope(TestType.SEC_ARP_WATCH, TestStatus.UNSUPPORTED, evidence),
+        )
+
         if (!st.binderAlive) {
-            return@withContext envelope(TestStatus.UNSUPPORTED, buildJsonObject {
+            return@withContext allUnsupported(buildJsonObject {
                 put("binder_alive", false); put("detail", "Shizuku not running")
             })
         }
         if (!st.permissionGranted && !runner.requestPermission()) {
-            return@withContext envelope(TestStatus.UNSUPPORTED, buildJsonObject {
+            return@withContext allUnsupported(buildJsonObject {
                 put("binder_alive", true); put("permission", false)
             })
         }
@@ -77,6 +96,125 @@ class ShizukuProbe {
             ok >= 1 -> TestStatus.PARTIAL
             else -> TestStatus.FAILED
         }
-        envelope(status, evidence, metrics)
+        listOf(
+            envelope(type, status, evidence, metrics),
+            raSourceTest(batch, ::envelope),
+            arpWatchTest(batch, ::envelope),
+        )
+    }
+
+    /** `link.ra_source` from the battery's `ip -6 route` / `ip addr` / `ip neigh` captures. */
+    private fun raSourceTest(
+        batch: ShizukuRunner.BatchResult,
+        envelope: (String, TestStatus, JsonObject, JsonObject?) -> Test,
+    ): Test {
+        val routeRaw = batch.results["ip6_route"]
+        if (!DumpParsers.captureUsable(routeRaw)) {
+            // The source command failed (executor sentinel or empty) — say so instead of
+            // presenting "no default routes" as a measurement of the network.
+            return envelope(TestType.LINK_RA_SOURCE, TestStatus.SKIPPED, buildJsonObject {
+                put("exec_path", batch.execPath)
+                put("reason", "ip -6 route capture unavailable: ${(routeRaw ?: "absent").take(80)}")
+            }, null)
+        }
+        val routes = DumpParsers.parseV6DefaultRoutes(routeRaw)
+        val macs = DumpParsers.parseInterfaceMacs(batch.results["ip_addr"])
+        // The RA sender's own identity: its link-local gateway address resolved through the
+        // neighbor table gives the router's MAC, which is what survives address renumbering.
+        val neighMacs = DumpParsers.lladdrByIp(DumpParsers.parseNeighbors(batch.results["ip_neigh"]))
+
+        val evidence = buildJsonObject {
+            put("exec_path", batch.execPath)
+            putJsonArray("default_routes") {
+                for (r in routes) addJsonObject {
+                    r.gateway?.let { put("gateway", it) }
+                    put("dev", r.dev)
+                    r.table?.let { put("table", it) }
+                    r.proto?.let { put("proto", it) }
+                    r.metric?.let { put("metric", it) }
+                    r.expiresSec?.let { put("expires_sec", it) }
+                    r.gateway?.let { gw -> neighMacs[gw]?.let { put("gateway_lladdr", it) } }
+                }
+            }
+            putJsonObject("interface_mac") {
+                // Only interfaces that actually carry a default route: the full MAC inventory
+                // belongs to the raw capture, not to this test's claim.
+                for (dev in routes.map { it.dev }.distinct()) macs[dev]?.let { put(dev, it) }
+            }
+        }
+        val metrics = buildJsonObject {
+            put("routes_total", routes.size)
+            put("routes_ra", routes.count { it.proto == "ra" })
+        }
+        val status = when {
+            routes.any { it.proto == "ra" } -> TestStatus.OK
+            // Routes parsed but none RA-installed, or a capture we couldn't parse a single
+            // default from: could be a genuinely RA-less link, could be vendor format drift —
+            // PARTIAL keeps it visible either way instead of quietly claiming success.
+            else -> TestStatus.PARTIAL
+        }
+        return envelope(TestType.LINK_RA_SOURCE, status, evidence, metrics)
+    }
+
+    /** `sec.arp_watch` from the battery's `ip neigh` snapshot + `ip monitor` window. */
+    private fun arpWatchTest(
+        batch: ShizukuRunner.BatchResult,
+        envelope: (String, TestStatus, JsonObject, JsonObject?) -> Test,
+    ): Test {
+        val neighRaw = batch.results["ip_neigh"]
+        val monitorRaw = batch.results["ip_monitor"]
+        val neighUsable = DumpParsers.captureUsable(neighRaw)
+        // The monitor window is best-effort (EXEC_TIMEOUT under newProcess on the Lenovo); the
+        // snapshot alone still yields the (ip → lladdr) pairs the MAC-change finding diffs.
+        val monitorRan = DumpParsers.captureUsable(monitorRaw)
+        if (!neighUsable && !monitorRan) {
+            return envelope(TestType.SEC_ARP_WATCH, TestStatus.SKIPPED, buildJsonObject {
+                put("exec_path", batch.execPath)
+                put("reason", "ip neigh capture unavailable: ${(neighRaw ?: "absent").take(80)}")
+            }, null)
+        }
+        val neighbors = DumpParsers.parseNeighbors(neighRaw)
+        val events = DumpParsers.parseNeighborEvents(monitorRaw)
+
+        val evidence = buildJsonObject {
+            put("exec_path", batch.execPath)
+            putJsonArray("neighbors") {
+                for (n in neighbors) addJsonObject {
+                    put("ip", n.ip)
+                    n.dev?.let { put("dev", it) }
+                    n.lladdr?.let { put("lladdr", it) }
+                    n.state?.let { put("state", it) }
+                    if (n.router) put("router", true)
+                }
+            }
+            // The comparison surface, precomputed: a MAC-change finding diffs this map between
+            // runs without re-walking the neighbor array.
+            putJsonObject("lladdr_by_ip") {
+                for ((ip, mac) in DumpParsers.lladdrByIp(neighbors)) put(ip, mac)
+            }
+            put("monitor_ran", monitorRan)
+            if (!monitorRan) put("monitor_reason", (monitorRaw ?: "absent").take(80))
+            putJsonArray("monitor_events") {
+                for (e in events) addJsonObject {
+                    put("ip", e.entry.ip)
+                    e.entry.dev?.let { put("dev", it) }
+                    e.entry.lladdr?.let { put("lladdr", it) }
+                    e.entry.state?.let { put("state", it) }
+                    if (e.deleted) put("deleted", true)
+                }
+            }
+        }
+        val metrics = buildJsonObject {
+            put("neighbors_total", neighbors.size)
+            put("neighbors_with_lladdr", neighbors.count { it.lladdr != null })
+            put("monitor_events", events.size)
+        }
+        val status = when {
+            neighbors.isNotEmpty() -> TestStatus.OK
+            // A snapshot that parsed to nothing (or a monitor-only capture) is thin evidence:
+            // usable command output with zero entries is unusual enough to flag, not to fail.
+            else -> TestStatus.PARTIAL
+        }
+        return envelope(TestType.SEC_ARP_WATCH, status, evidence, metrics)
     }
 }

echolot-app/core-shizuku/src/test/kotlin/app/echo_lot/shizuku/DumpParsersTest.kt

@@ -0,0 +1,303 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package app.echo_lot.shizuku
+
+import kotlin.test.Test
+import kotlin.test.assertEquals
+import kotlin.test.assertFalse
+import kotlin.test.assertNull
+import kotlin.test.assertTrue
+
+/**
+ * The fixtures below are the REAL shell-battery captures from the two archived prober reports
+ * (echolot-prober/reports/CPH2747-android16-sdk36-build5.json — OnePlus 15, UserService path;
+ * TB330FU-android15-sdk35-build5.json — Lenovo TB330FU, newProcess fallback), trimmed to the
+ * relevant lines but otherwise verbatim. That includes their warts on purpose: the UserService
+ * path's stray `uid=2000` first line, the 1200-char evidence trim cutting the last line mid-word,
+ * the Lenovo's 10-digit route table ids and its `EXEC_TIMEOUT(newProcess)` monitor sentinel.
+ * A parser that only survives clean textbook output has not been tested.
+ */
+class DumpParsersTest {
+
+    // ---- OnePlus 15 (CPH2747, Android 16) — UserService exec path ----
+
+    private val onePlusIp6Route = """
+        uid=2000
+        fe80::/64 dev wlan0 table 1028 proto kernel metric 256 pref medium
+        fe80::/64 dev wlan0 table 1028 proto static metric 1024 pref medium
+        default via fe80::7a9a:18ff:fe54:b8f9 dev wlan0 table 1028 proto ra metric 1024 expires 1269sec pref medium
+        fe80::/64 dev vgate0 table 1031 proto kernel metric 256 pref medium
+        2001:4bb8:417:bd78::/64 dev rmnet_data4 table 1032 proto kernel metric 256 pref medium
+        2001:4bb8:417:bd78::/64 dev rmnet_data4 table 1032 proto static metric 1024 pref medium
+        fe80::/64 dev rmnet_data4 table 1032 proto kernel metric 256 pref medium
+        default via fe80::246f:12be:21ef:1b54 dev rmnet_data4 table 1032 proto ra metric 1024 expires 64373sec hoplimit 255 pref medium
+        2001:4bb8:2fb:fe4c::/64 dev rmnet_data2 table 1000000022 proto static metric 1024 pref medium
+        fe80::/64 dev wlan0 table 1000000028 proto static metric 1024 pref medium
+        2001:4bb8:417:bd78::/64 dev rmnet_data4 table 1000000032 proto static metric 1024 pref medium
+        fe80::/64 dev dummy0 table 1002 proto kernel metric 256 pref medium
+        default dev dummy0 table 1002 proto static metric 1024 pref medium
+        fe80::/64 dev ifb0 table 1003 proto kernel metric 256 pref medium
+        fe80::/64 dev ifb1 table 1004 proto kerne
+    """.trimIndent()
+
+    private val onePlusIpNeigh = """
+        uid=2000
+        10.13.102.111 dev wlan0  FAILED
+        10.13.102.50 dev wlan0 lladdr 50:57:9c:4f:7a:3c STALE
+        10.13.102.31 dev wlan0 lladdr 98:5f:d3:f6:f1:75 STALE
+        10.13.102.116 dev wlan0 lladdr 0c:08:b4:03:68:0e STALE
+        10.13.102.120 dev wlan0 lladdr 0e:d8:14:58:6c:8b STALE
+        10.13.102.5 dev wlan0 lladdr 90:09:d0:1a:83:e4 STALE
+        10.13.102.1 dev wlan0 lladdr 78:9a:18:54:b8:f9 REACHABLE
+        10.13.102.21 dev wlan0 lladdr c8:7f:54:01:94:7c STALE
+        fe80::babe:f4ff:febc:caf9 dev wlan0 lladdr b8:be:f4:bc:ca:f9 REACHABLE
+        fe80::7a9a:18ff:fe54:b8f9 dev wlan0 lladdr 78:9a:18:54:b8:f9 router STALE
+        fe80::babe:f4ff:febc:cacf dev wlan0 lladdr b8:be:f4:bc:ca:cf REACHABLE
+        fe80::babe:f4ff:fec2:bf14 dev wlan0 lladdr b8:be:f4:c2:bf:14 REACHABLE
+    """.trimIndent()
+
+    // The 1200-char trim cut this capture off before wlan0's stanza — so the real archived
+    // evidence has NO MAC for the interface that carries the default route. The parser must
+    // yield what is there and nothing else; the probe records the gap instead of inventing one.
+    private val onePlusIpAddr = """
+        uid=2000
+        1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group default qlen 1000
+            link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
+            inet 127.0.0.1/8 scope host lo
+               valid_lft forever preferred_lft forever
+            inet6 ::1/128 scope host
+               valid_lft forever preferred_lft forever
+        2: dummy0: <BROADCAST,NOARP,UP,LOWER_UP> mtu 1500 qdisc noqueue state UNKNOWN group default qlen 1000
+            link/ether be:3d:e2:93:78:b9 brd ff:ff:ff:ff:ff:ff
+            inet6 fe80::bc3d:e2ff:fe93:78b9/64 scope link
+               valid_lft forever preferred_lft forever
+        3: ifb0: <BROADCAST,NOARP,UP,LOWER_UP> mtu 1500 qdisc htb state UNKNOWN group default qlen 1000
+            link/ether ba:6e:46:b5:3d:bb brd ff:ff:ff:ff:ff:ff
+        4: ifb1: <BROADCAST,NOARP,UP,LOWER_UP> mtu 1500 qdisc htb state UNKNOWN group default qlen 1000
+            link/ether d6:2a:e2:f5:93:8f brd ff:ff:ff:ff:ff:ff
+        5: tunl0@NONE: <NOARP> mtu 1480 qdisc noop state DOWN group default qlen 1000
+            link/ipip 0.0.0.0 brd 0.0.0.0
+        6: gre0@NONE: <NO
+    """.trimIndent()
+
+    // A monitor window that ran but saw nothing: the capture is "usable", just empty of events.
+    private val onePlusIpMonitor = "uid=2000"
+
+    // ---- Lenovo TB330FU (Android 15) — newProcess fallback ----
+
+    private val lenovoIp6Route = """
+        fe80::/64 dev wlan0 table 1000000015 proto static metric 1024 pref medium
+        fe80::/64 dev dummy0 table 1002 proto kernel metric 256 pref medium
+        default dev dummy0 table 1002 proto static metric 1024 pref medium
+        fe80::/64 dev wlan0 table 1015 proto kernel metric 256 pref medium
+        fe80::/64 dev wlan0 table 1015 proto static metric 1024 pref medium
+        default via fe80::7a9a:18ff:fe54:b8f9 dev wlan0 table 1015 proto ra metric 1024 expires 1622sec pref medium
+        local ::1 dev lo table local proto kernel metric 0 pref medium
+        local fe80::416:b9ff:feac:5b65 dev wlan0 table local proto kernel metric 0 pref medium
+        local fe80::1450:43ff:feec:93c4 dev dummy0 table local proto kernel metric 0 pref medium
+        multicast ff00::/8 dev dummy0 table local proto kernel metric 256 pref medium
+        multicast ff00::/8 dev wlan0 table local proto kernel metric 256 pref medium
+    """.trimIndent()
+
+    private val lenovoIpNeigh = """
+        10.13.102.21 dev wlan0 lladdr c8:7f:54:01:94:7c STALE
+        10.13.102.64 dev wlan0 lladdr b8:be:f4:c2:bf:14 STALE
+        10.13.102.5 dev wlan0 lladdr 90:09:d0:1a:83:e4 STALE
+        10.13.102.1 dev wlan0 lladdr 78:9a:18:54:b8:f9 STALE
+        10.13.102.79 dev wlan0 lladdr 02:11:32:25:63:bb STALE
+        fe80::babe:f4ff:febc:cacf dev wlan0 lladdr b8:be:f4:bc:ca:cf STALE
+        fe80::7a9a:18ff:fe54:b8f9 dev wlan0 lladdr 78:9a:18:54:b8:f9 router STALE
+        fe80::babe:f4ff:fec2:bf14 dev wlan0 lladdr b8:be:f4:c2:bf:14 STALE
+        fe80::babe:f4ff:febc:caf9 dev wlan0 lladdr b8:be:f4:bc:ca:f9 STALE
+    """.trimIndent()
+
+    private val lenovoIpAddr = """
+        1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group default qlen 1000
+            link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
+            inet 127.0.0.1/8 scope host lo
+               valid_lft forever preferred_lft forever
+        2: dummy0: <BROADCAST,NOARP,UP,LOWER_UP> mtu 1500 qdisc noqueue state UNKNOWN group default qlen 1000
+            link/ether 16:50:43:ec:93:c4 brd ff:ff:ff:ff:ff:ff
+            inet6 fe80::1450:43ff:feec:93c4/64 scope link
+               valid_lft forever preferred_lft forever
+        3: ifb0: <BROADCAST,NOARP> mtu 1500 qdisc noop state DOWN group default qlen 32
+            link/ether f6:d4:d4:9b:51:9c brd ff:ff:ff:ff:ff:ff
+        4: ifb1: <BROADCAST,NOARP> mtu 1500 qdisc noop state DOWN group default qlen 32
+            link/ether fe:16:ea:60:a2:d1 brd ff:ff:ff:ff:ff:ff
+        5: tunl0@NONE: <NOARP> mtu 1480 qdisc noop state DOWN group default qlen 1000
+            link/ipip 0.0.0.0 brd 0.0.0.0
+        7: gretap0@NONE: <BROADCAST,MULTICAST> mtu 1462 qdisc noop state DOWN group default qlen 1000
+            link/ether 00:00:00:00:00:00 brd ff:ff:ff:ff:f
+    """.trimIndent()
+
+    // On the Lenovo the 5 s monitor window exceeds the newProcess exec timeout — the executor's
+    // sentinel is all we get, and the arp_watch test must still stand on the snapshot alone.
+    private val lenovoIpMonitor = "EXEC_TIMEOUT(newProcess)"
+
+    // ---- link.ra_source: v6 default routes ----
+
+    @Test
+    fun onePlusDefaultRoutesParsed() {
+        val routes = DumpParsers.parseV6DefaultRoutes(onePlusIp6Route)
+        assertEquals(3, routes.size)
+
+        val wlan = routes.single { it.dev == "wlan0" }
+        assertEquals("fe80::7a9a:18ff:fe54:b8f9", wlan.gateway)
+        assertEquals("1028", wlan.table)
+        assertEquals("ra", wlan.proto)
+        assertEquals(1024L, wlan.metric)
+        assertEquals(1269L, wlan.expiresSec)
+
+        // The cellular default: `hoplimit 255` sits between expires and pref and must not derail
+        // the token walk.
+        val rmnet = routes.single { it.dev == "rmnet_data4" }
+        assertEquals("fe80::246f:12be:21ef:1b54", rmnet.gateway)
+        assertEquals("1032", rmnet.table)
+        assertEquals(64373L, rmnet.expiresSec)
+
+        // Android's gateway-less dummy0 default is a real route; it is the proto that tells a
+        // consumer it is not an RA.
+        val dummy = routes.single { it.dev == "dummy0" }
+        assertNull(dummy.gateway)
+        assertEquals("static", dummy.proto)
+        assertNull(dummy.expiresSec)
+    }
+
+    @Test
+    fun lenovoNumberedTablesAllCaptured() {
+        val routes = DumpParsers.parseV6DefaultRoutes(lenovoIp6Route)
+        // Two default routes: the RA one in table 1015 and the dummy0 one in 1002. The 10-digit
+        // table 1000000015 and the `table local` rows carry no default and must neither appear
+        // nor break parsing.
+        assertEquals(setOf("1002", "1015"), routes.map { it.table }.toSet())
+
+        val ra = routes.single { it.proto == "ra" }
+        assertEquals("fe80::7a9a:18ff:fe54:b8f9", ra.gateway)
+        assertEquals("wlan0", ra.dev)
+        assertEquals("1015", ra.table)
+        assertEquals(1622L, ra.expiresSec)
+    }
+
+    @Test
+    fun tenDigitTableIdOnADefaultRouteSurvives() {
+        // Not seen on a default route in the wild yet, but the Lenovo proves vendors put routes
+        // in tables past Int range — the day one holds a default, it must not overflow away.
+        val routes = DumpParsers.parseV6DefaultRoutes(
+            "default via fe80::1 dev wlan0 table 1000000015 proto ra metric 1024 expires 100sec pref medium"
+        )
+        assertEquals(1, routes.size)
+        assertEquals("1000000015", routes[0].table)
+        assertEquals(100L, routes[0].expiresSec)
+    }
+
+    // ---- link.ra_source: interface MACs ----
+
+    @Test
+    fun onePlusInterfaceMacsParsed() {
+        val macs = DumpParsers.parseInterfaceMacs(onePlusIpAddr)
+        assertEquals("be:3d:e2:93:78:b9", macs["dummy0"])
+        assertEquals("ba:6e:46:b5:3d:bb", macs["ifb0"])
+        // link/loopback and link/ipip are not identities.
+        assertFalse("lo" in macs)
+        assertFalse("tunl0" in macs)
+        // The capture is cut mid-stanza-header ("6: gre0@NONE: <NO") — no exception, no entry.
+        assertNull(macs["gre0"])
+    }
+
+    @Test
+    fun lenovoInterfaceMacsParsed() {
+        val macs = DumpParsers.parseInterfaceMacs(lenovoIpAddr)
+        assertEquals("16:50:43:ec:93:c4", macs["dummy0"])
+        assertEquals("fe:16:ea:60:a2:d1", macs["ifb1"])
+        // The @-suffixed stanza name resolves to the bare interface name.
+        assertEquals("00:00:00:00:00:00", macs["gretap0"])
+    }
+
+    // ---- sec.arp_watch: neighbor snapshot ----
+
+    @Test
+    fun onePlusNeighborsParsed() {
+        val n = DumpParsers.parseNeighbors(onePlusIpNeigh)
+        assertEquals(12, n.size) // the `uid=2000` noise line is not a neighbor
+
+        val failed = n.single { it.ip == "10.13.102.111" }
+        assertNull(failed.lladdr)
+        assertEquals("FAILED", failed.state)
+        assertEquals("wlan0", failed.dev)
+
+        val gw = n.single { it.ip == "10.13.102.1" }
+        assertEquals("78:9a:18:54:b8:f9", gw.lladdr)
+        assertEquals("REACHABLE", gw.state)
+
+        val v6gw = n.single { it.ip == "fe80::7a9a:18ff:fe54:b8f9" }
+        assertTrue(v6gw.router)
+        assertEquals("78:9a:18:54:b8:f9", v6gw.lladdr)
+    }
+
+    @Test
+    fun lenovoNeighborsParsed() {
+        val n = DumpParsers.parseNeighbors(lenovoIpNeigh)
+        assertEquals(9, n.size)
+        assertTrue(n.all { it.lladdr != null && it.state == "STALE" })
+        assertEquals(1, n.count { it.router })
+    }
+
+    @Test
+    fun lladdrByIpIsTheComparisonSurface() {
+        val pairs = DumpParsers.lladdrByIp(DumpParsers.parseNeighbors(onePlusIpNeigh))
+        // 12 neighbors, 11 with a MAC — the FAILED entry must drop out, or a diff against a
+        // later run would flag "null → MAC" as a gateway change.
+        assertEquals(11, pairs.size)
+        assertEquals("78:9a:18:54:b8:f9", pairs["10.13.102.1"])
+        assertFalse("10.13.102.111" in pairs)
+    }
+
+    // ---- sec.arp_watch: monitor window ----
+
+    @Test
+    fun monitorSentinelIsUnusableAndYieldsNoEvents() {
+        assertFalse(DumpParsers.captureUsable(lenovoIpMonitor))
+        assertTrue(DumpParsers.parseNeighborEvents(lenovoIpMonitor).isEmpty())
+    }
+
+    @Test
+    fun quietMonitorWindowIsUsableButEmpty() {
+        // OnePlus: the monitor ran (only the uid noise line came back) — "ran and saw nothing"
+        // must stay distinguishable from "never ran".
+        assertTrue(DumpParsers.captureUsable(onePlusIpMonitor))
+        assertTrue(DumpParsers.parseNeighborEvents(onePlusIpMonitor).isEmpty())
+    }
+
+    @Test
+    fun monitorNeighEventsParsedFromLabeledLines() {
+        // Synthetic, in `ip monitor all` label format — neither archived run caught a live
+        // transition, but the format is fixed by iproute2's print_neigh/print_headers.
+        val sample = """
+            [NEIGH]10.13.102.1 dev wlan0 lladdr 78:9a:18:54:b8:f9 REACHABLE
+            [NEIGH]Deleted 10.13.102.5 dev wlan0 lladdr 90:09:d0:1a:83:e4 STALE
+            [ROUTE]default via 10.13.102.1 dev wlan0 table 1015
+            [NEIGH]fe80::7a9a:18ff:fe54:b8f9 dev wlan0 lladdr 78:9a:18:54:b8:f9 router STALE
+        """.trimIndent()
+        val events = DumpParsers.parseNeighborEvents(sample)
+        assertEquals(3, events.size) // the ROUTE line belongs to a different family
+        assertEquals("10.13.102.1", events[0].entry.ip)
+        assertFalse(events[0].deleted)
+        assertTrue(events[1].deleted)
+        assertEquals("90:09:d0:1a:83:e4", events[1].entry.lladdr)
+        assertTrue(events[2].entry.router)
+    }
+
+    // ---- degradation: missing or garbage input ----
+
+    @Test
+    fun missingAndGarbageInputYieldsEmptyResultsNotExceptions() {
+        for (bad in listOf(null, "", "   \n ", "EXEC_TIMEOUT(newProcess)", "SHIZUKU_BINDER_DEAD",
+            "NEWPROCESS_UNAVAILABLE", "total garbage\nno routes here at all\ndefault", "default")) {
+            assertTrue(DumpParsers.parseV6DefaultRoutes(bad).isEmpty(), "routes from: $bad")
+            assertTrue(DumpParsers.parseNeighbors(bad).isEmpty(), "neighbors from: $bad")
+            assertTrue(DumpParsers.parseInterfaceMacs(bad).isEmpty(), "macs from: $bad")
+            assertTrue(DumpParsers.parseNeighborEvents(bad).isEmpty(), "events from: $bad")
+        }
+    }
+}

echolot-app/gradle/libs.versions.toml

@@ -8,6 +8,7 @@ lifecycle = "2.8.7"
 activityCompose = "1.9.3"
 composeBom = "2024.10.01"
 shizuku = "13.1.5"
+junit4 = "4.13.2"
 
 [libraries]
 kotlinx-serialization-json = { group = "org.jetbrains.kotlinx", name = "kotlinx-serialization-json", version.ref = "kotlinxSerialization" }
@@ -23,6 +24,10 @@ androidx-ui-tooling = { group = "androidx.compose.ui", name = "ui-tooling" }
 androidx-ui-tooling-preview = { group = "androidx.compose.ui", name = "ui-tooling-preview" }
 androidx-material3 = { group = "androidx.compose.material3", name = "material3" }
 shizuku-api = { group = "dev.rikka.shizuku", name = "api", version.ref = "shizuku" }
+# Android-module unit tests run on JUnit 4 (AGP's default); the JVM modules use kotlin("test")
+# with the JUnit Platform instead — that helper isn't available under AGP 9's built-in Kotlin.
+kotlin-test-junit = { group = "org.jetbrains.kotlin", name = "kotlin-test-junit", version.ref = "kotlin" }
+junit4 = { group = "junit", name = "junit", version.ref = "junit4" }
 shizuku-provider = { group = "dev.rikka.shizuku", name = "provider", version.ref = "shizuku" }
 
 [plugins]

echolot-app/scripts/enroll-link.sh

@@ -5,8 +5,13 @@
 # Mints an enrollment link on the probe server and prints it — as text, as a QR code if
 # `qrencode` is around, and as an adb command if a device is attached.
 #
-# The admin listener is localhost-only by design, so this goes over SSH. The link carries a
-# single-use bearer token: treat it like a password until it is redeemed.
+# The link is minted by the server binary on the host rather than over HTTP. The admin API this
+# used to call is gone: the admin UI that replaced it is authenticated, as it should be, and
+# adding a second unauthenticated door on loopback is what briefly exposed the old one to the
+# network. A root shell on the host needs no authentication anyway — whoever has one already has
+# every privilege the server has.
+#
+# The link carries a single-use bearer token: treat it like a password until it is redeemed.
 #
 # Usage: echolot-app/scripts/enroll-link.sh [note]
 set -euo pipefail
@@ -14,13 +19,18 @@ set -euo pipefail
 SSH_HOST="${ECHOLOT_SSH:-claude-echolot}"
 NOTE="${1:-manual}"
 
-MINTED=$(ssh -o BatchMode=yes "$SSH_HOST" \
-  "curl -s -X POST 'http://127.0.0.1:8444/admin/enroll-tokens?note=$NOTE'")
+# The env file is sourced rather than assumed: the state directory and the public URL live there,
+# and minting against the wrong state directory would produce a token the running server has
+# never heard of.
+REMOTE='set -a; . /etc/echolot/server.env; set +a;
+        exec /usr/local/bin/echolot-server --mint-enroll-token'
+RAW=$(ssh -o BatchMode=yes "$SSH_HOST" "sudo sh -c \"$REMOTE '$NOTE'\"" 2>/dev/null || true)
+URI=$(printf '%s' "$RAW" | tr -d '\r' | grep -m1 '^echolot://enroll' || true)
 
-URI=$(printf '%s' "$MINTED" | python -c 'import json,sys;print(json.load(sys.stdin).get("enroll_uri",""))')
 if [ -z "$URI" ]; then
-  echo "server returned no enroll_uri (needs server-v0.5.4+):" >&2
-  echo "$MINTED" >&2
+  echo "could not mint a link — needs a server with --mint-enroll-token (v0.9.7+)." >&2
+  echo "raw response:" >&2
+  printf '%s\n' "$RAW" >&2
   exit 1
 fi
 

server/Dockerfile

@@ -22,4 +22,7 @@ VOLUME ["/state"]
 # the data plane must see real client source addresses/TTLs, and Docker's
 # userland NAT would falsify exactly what this server exists to observe.
 EXPOSE 8441/tcp 8442/udp 8443/tcp
+# The verb is explicit here too, so `docker run <image>` serves and `docker run <image> --help`
+# still works by overriding the command.
 ENTRYPOINT ["/echolot-server"]
+CMD ["--serve"]

server/README.md

@@ -106,9 +106,7 @@ ECHOLOT_UDP_LISTEN=203.0.113.10:8442,203.0.113.11:8442,[2001:db8::10]:8442,[2001
 
 Passing `--self-update-api` to `--install-systemd` additionally installs a daily randomized
 self-update timer (`echolot-server-update.timer`) that restarts the service after a successful
-update. Updates are checksum-verified against the release's `SHA256SUMS` (integrity, not
-authenticity — signature verification remains TODO before treating the update source as
-untrusted).
+update.
 
 ### Self-update (opt-in, native only)
 
@@ -119,14 +117,24 @@ echolot-server --self-update \
 
 Fetches the newest `server-v*` release asset for this OS/arch and atomically replaces the
 binary; systemd's `Restart=` brings up the new version. Run it from a systemd timer for
-unattended updates. TODO before enabling anywhere untrusted: signature verification of the
-downloaded asset.
+unattended updates.
+
+Releases are trusted by signature, not by host: CI signs `SHA256SUMS` with an ed25519 key that
+exists only in its secret store (`RELEASE_SIGNING_KEY`), and the updater verifies
+`SHA256SUMS.sig` against the public key baked into the binary before believing any checksum —
+an unsigned or re-signed release is refused, so a compromised Gitea can withhold updates but not
+inject one. Running your own release pipeline? Mint a keypair with
+`go run ./cmd/release-sign -gen`, set the secret, and point `ECHOLOT_SELF_UPDATE_PUBKEY` (or
+`--self-update-pubkey`) at your public key.
 
 ## First contact
 
 ```sh
-# 1. mint an enrollment token (admin listener is loopback-only)
-curl -s -X POST 'http://127.0.0.1:8444/admin/enroll-tokens?note=phone'
+# 1. mint an enrollment token (admin listener is loopback-only; authenticates as the
+#    break-glass admin — set that once with --set-admin-password)
+curl -s -u admin:<password> -H 'Accept: application/json' \
+  -X POST 'http://127.0.0.1:8444/admin/enroll-tokens?note=phone'
+# → { "token": "…", "expires_in_s": 86400, "enroll_uri": "echolot://enroll?…" }
 # 2. device enrolls with it (normally via the echolot:// QR code)
 curl -sk -X POST https://<host>:8443/v1/enroll -H 'Authorization: Bearer <token>'
 # 3. device fetches its profile
@@ -144,5 +152,43 @@ go vet ./...
 
 CI (`.gitea/workflows/build-server.yml`): tests on every push touching `server/`;
 tagging `server-v1.2.3` builds + pushes the container image to the Gitea registry and
-attaches static linux amd64/arm64 binaries (+ SHA256SUMS) to a release — the same
+attaches static linux amd64/arm64 binaries (+ signed SHA256SUMS) to a release — the same
 artifacts `--self-update` consumes.
+
+## TLS for the admin UI
+
+The binary terminates TLS itself; there is no reverse proxy in the design. It already serves TLS
+for the control plane, so this is reuse rather than new machinery, and it keeps the "one process,
+one config file" property. A proxy would also invite someone to eventually front the control plane
+too — which would break SPKI pinning, because clients pin *that* certificate's key.
+
+```
+ECHOLOT_ADMIN_LISTEN=[2001:db8::2]:443
+ECHOLOT_ADMIN_TLS_CERT=/etc/echolot/admin.pem
+ECHOLOT_ADMIN_TLS_KEY=/etc/echolot/admin.key
+ECHOLOT_ADMIN_BASE_URL=https://admin.example.net
+```
+
+Certificates come from any ACME client. **DNS-01 is the one to use here**: it needs no inbound
+port 80, which matters on a host where 80 is awkward or already spoken for.
+
+```sh
+acme.sh --issue --dns dns_cf -d admin.example.net \
+  --key-file       /etc/echolot/admin.key \
+  --fullchain-file /etc/echolot/admin.pem
+```
+
+**No reload hook is needed.** The certificate is re-read when the files change, so a renewal that
+drops new files in place is picked up on the next handshake. That is deliberate: a reload hook is
+the part of a renewal setup that quietly stops working, months later, and is noticed only once the
+certificate has already expired. A torn write — renewal tools write cert and key separately — keeps
+the previous certificate rather than failing the listener.
+
+Serving the admin UI in plaintext on a non-loopback address is refused: the session cookie is a
+bearer credential for everything the server can do, and the OIDC authorization code arrives in a
+URL. Bind to loopback and use an SSH tunnel (`ssh -L 8444:localhost:8444 host`), supply a
+certificate, or set `ECHOLOT_ADMIN_INSECURE=1` if you mean it.
+
+The control-plane certificate is deliberately *not* hot-reloaded. Clients pin its public key, so
+replacing it is a rotation an operator should have to think about, not something that happens
+because a file changed.

server/cmd/echolot-server/main.go

@@ -11,6 +11,7 @@
 package main
 
 import (
+	"bufio"
 	"context"
 	"crypto/ecdsa"
 	"crypto/elliptic"
@@ -18,7 +19,6 @@ import (
 	"crypto/tls"
 	"crypto/x509"
 	"crypto/x509/pkix"
-	"encoding/json"
 	"encoding/pem"
 	"errors"
 	"fmt"
@@ -36,11 +36,17 @@ import (
 	"syscall"
 	"time"
 
+	"echo-lot.app/server/internal/acmehttp"
+	"echo-lot.app/server/internal/adminauth"
+	"echo-lot.app/server/internal/adminui"
 	"echo-lot.app/server/internal/canarydns"
+	"echo-lot.app/server/internal/certreload"
 	"echo-lot.app/server/internal/compat"
 	"echo-lot.app/server/internal/config"
 	"echo-lot.app/server/internal/control"
 	"echo-lot.app/server/internal/dataplane"
+	"echo-lot.app/server/internal/oidc"
+	"echo-lot.app/server/internal/ratelimit"
 	"echo-lot.app/server/internal/runs"
 	"echo-lot.app/server/internal/selftest"
 	"echo-lot.app/server/internal/selfupdate"
@@ -69,6 +75,30 @@ func run() error {
 	control.Version = Version
 
 	switch {
+	case actions.Help:
+		// Compatibility shim for one release.
+		//
+		// Serving became an explicit verb, but self-update is run by the *old* binary — so the
+		// repair added to the updater cannot fix the very update that installs the new one. A
+		// unit written before this change starts us with no arguments, and without this branch
+		// the service would simply stop working, unattended, on a host nobody is watching.
+		//
+		// Only when systemd started us: INVOCATION_ID is set by systemd for every service
+		// invocation and by nothing else, so a person at a terminal still gets usage. Remove
+		// this once no deployment predates --serve.
+		if os.Getenv("INVOCATION_ID") != "" {
+			slog.Warn("started by systemd with no verb — this unit predates --serve; " +
+				"repairing it and serving anyway")
+			if repaired, err := system.RepairExecStart(); err != nil {
+				slog.Error("could not repair the unit; fix ExecStart by hand", "err", err)
+			} else if repaired {
+				slog.Info("systemd unit updated to pass --serve")
+			}
+			return serve(cfg)
+		}
+		config.Usage(os.Stderr)
+		os.Exit(2)
+		return nil
 	case actions.Version:
 		fmt.Println(Version)
 		return nil
@@ -79,17 +109,87 @@ func run() error {
 		return system.InstallSystemd(cfg.SelfUpdateAPI)
 	case actions.UninstallSystemd:
 		return system.UninstallSystemd()
+	case actions.SetAdminPassword:
+		return setAdminPassword(cfg)
+	case actions.MintEnrollToken != "":
+		return mintEnrollToken(cfg, actions.MintEnrollToken)
 	case actions.SelfUpdate:
-		return selfupdate.Run(cfg.SelfUpdateAPI, Version)
+		return selfupdate.Run(cfg.SelfUpdateAPI, cfg.SelfUpdatePubKey, Version)
 	}
 	return serve(cfg)
 }
 
+// mintEnrollToken prints a §2.1 bootstrap link for a new device.
+//
+// The link is assembled here rather than by hand because it has to carry the public URL and the
+// base64 SPKI pin percent-encoded correctly, and a pin wrong by one character fails later as an
+// inscrutable TLS error rather than as a bad pin.
+func mintEnrollToken(cfg *config.Config, note string) error {
+	st, err := store.Open(cfg.StateDir)
+	if err != nil {
+		return fmt.Errorf("state store: %w", err)
+	}
+	cert, err := loadOrCreateCert(cfg)
+	if err != nil {
+		return fmt.Errorf("tls: %w", err)
+	}
+	pin, err := control.SpkiPinB64(cert)
+	if err != nil {
+		return fmt.Errorf("pin: %w", err)
+	}
+	tok, err := st.NewEnrollToken(24*time.Hour, note)
+	if err != nil {
+		return err
+	}
+	base := cfg.PublicControlURL
+	if base == "" {
+		return fmt.Errorf("set ECHOLOT_PUBLIC_URL so the link can say where to connect")
+	}
+	fmt.Println(control.EnrollmentURI(base, pin, tok))
+	// stderr, so piping the command somewhere yields the link alone.
+	fmt.Fprintln(os.Stderr, "\nSingle use, valid 24 hours. Treat it like a password until spent.")
+	return nil
+}
+
+// controlURL is the address devices connect to: the hostname that selects the pinned certificate.
+//
+// Falls back to the public URL when no separate control hostname is configured, so a server that
+// does not share the admin port keeps answering discovery with something usable.
+func controlURL(cfg *config.Config) string {
+	if cfg.ControlHostname == "" {
+		return cfg.PublicControlURL
+	}
+	return "https://" + cfg.ControlHostname
+}
+
 func serve(cfg *config.Config) error {
 	slog.Info("echolot-server starting", "version", Version, "mode",
 		map[bool]string{true: "container", false: "native"}[cfg.Docker],
 		"state_dir", cfg.StateDir)
 
+	// The reserved addresses' proof is only as good as 80/443 actually being free there.
+	// CheckReserved already keeps OUR listeners away, but a process outside this config pollutes
+	// them just as silently — the adb-beacon receiver on 0.0.0.0:443 did exactly that. So ask
+	// the OS, not the config. A hard stop for the same reason CheckReserved is one: the failure
+	// is invisible, and its first symptom is a measurement calling an intercepted network clean.
+	if reserved := cfg.ReservedIPs(); len(reserved) > 0 {
+		occupied, unverifiable := selftest.ReservedWebPortsFree(reserved)
+		if len(occupied) > 0 {
+			return fmt.Errorf(
+				"refusing to start: something outside this server is listening on reserved "+
+					"measurement address(es) %s\n"+
+					"The interception proof those addresses exist for is void while anything "+
+					"answers there.\nFind it with `ss -tlnp | grep -E ':(80|443) '`, stop it, "+
+					"or remove the address from ECHOLOT_RESERVED_ADDRS if it is no longer reserved",
+				strings.Join(occupied, ", "))
+		}
+		for _, u := range unverifiable {
+			// Not fatal: an address with a typo, or one this host no longer carries, is a
+			// config problem — refusing to serve over it would take the whole instrument down.
+			slog.Warn("could not verify a reserved web port is free", "addr", u)
+		}
+	}
+
 	st, err := store.Open(cfg.StateDir)
 	if err != nil {
 		return fmt.Errorf("state store: %w", err)
@@ -106,12 +206,22 @@ func serve(cfg *config.Config) error {
 
 	sessions := session.NewManager(15 * time.Minute)
 	dp := &dataplane.Server{Sessions: sessions}
+	// Spec §2.5 ceilings. Control plane answers 429; the data plane drops silently. 0 = off.
+	if cfg.RateUDPPps > 0 {
+		pps := float64(cfg.RateUDPPps)
+		// Burst of two seconds' worth: a 5000-packet train arrives as one burst by design.
+		dp.PacketRate = ratelimit.New(pps, 2*pps)
+	}
+	if cfg.RateUDPKbps > 0 {
+		bytesPerSec := float64(cfg.RateUDPKbps) * 125 // kbps -> bytes/s
+		dp.ByteRate = ratelimit.New(bytesPerSec, bytesPerSec)
+	}
 	// TCP echo shares the control cert for its elt-echo TLS variant.
 	tcpSrv := &tcpecho.Server{
 		TLSConfig: &tls.Config{Certificates: []tls.Certificate{cert}, MinVersion: tls.VersionTLS12},
 	}
 
-	caps := []string{"udp-probe", "delayed-echo", "connect-back", "http-echo", "downtrain", "big-send"}
+	caps := []string{"udp-probe", "delayed-echo", "connect-back", "http-echo", "downtrain", "big-send", "throughput"}
 	// Crafted fragments need a raw socket. Advertised only when one can actually be opened —
 	// a capability we cannot deliver turns a missing feature into a failed measurement.
 	rawFrag := dataplane.RawFragSupported()
@@ -150,7 +260,43 @@ func serve(cfg *config.Config) error {
 	slog.Info("client compatibility", "accepts_app", appRange.String(),
 		"protocol", control.ProtocolVersion, "schema", control.SchemaVersion)
 
+	// Identity is optional. Without an issuer the server simply has no sign-in, and
+	// uploads=account can never be satisfied — which is the honest outcome, not a silent
+	// downgrade to anonymous.
+	// One verifier per issuer. An IdP may mint a distinct issuer per application — Authentik
+	// derives it from the application slug — and a token's `iss` must match whoever signed it.
+	// Each verifier accepts only the client belonging to its own issuer, so a token minted for
+	// the phone cannot be replayed at the admin login and vice versa.
+	var idp, adminIdP *oidc.Verifier
+	appIssuer := cfg.OIDCAppIssuer
+	if appIssuer == "" {
+		appIssuer = cfg.OIDCIssuer // IdPs with one global issuer
+	}
+	if appIssuer != "" && cfg.OIDCAppClientID != "" {
+		idp = oidc.New(oidc.Config{
+			Issuer: appIssuer, AppClientID: cfg.OIDCAppClientID, AdminGroup: cfg.OIDCAdminGroup,
+		}, nil)
+		slog.Info("identity: app client", "issuer", appIssuer, "client_id", cfg.OIDCAppClientID)
+	}
+	if cfg.OIDCIssuer != "" && cfg.OIDCClientID != "" {
+		adminIdP = oidc.New(oidc.Config{
+			Issuer: cfg.OIDCIssuer, ClientID: cfg.OIDCClientID, AdminGroup: cfg.OIDCAdminGroup,
+		}, nil)
+		slog.Info("identity: admin client", "issuer", cfg.OIDCIssuer,
+			"client_id", cfg.OIDCClientID, "admin_group", cfg.OIDCAdminGroup)
+		if cfg.OIDCAdminGroup == "" {
+			slog.Warn("no admin group set: nobody will be an admin via OIDC " +
+				"(set ECHOLOT_OIDC_ADMIN_GROUP)")
+		}
+	}
+	if idp == nil && adminIdP == nil && cfg.UploadsMode == string(runs.ModeAccount) {
+		slog.Warn("uploads=account but no identity provider is configured — " +
+			"every upload will be refused")
+	}
+
+	ip4, ip6, ip4Alt, ip6Alt := cfg.MeasurementAddrs()
 	ctl := &control.Server{
+		IP4: ip4, IP6: ip6, IP4Alt: ip4Alt, IP6Alt: ip6Alt,
 		Store: st, Sessions: sessions, Name: cfg.Name,
 		UDPPort: mustPort(firstAddr(cfg.UDPListen)), TCPPort: mustPort(firstAddr(cfg.TCPListen)),
 		StunPort: mustPort(firstAddr(cfg.StunListen)), PinB64: pin, CertChain: cert.Certificate,
@@ -161,12 +307,21 @@ func serve(cfg *config.Config) error {
 		Runs:             runStore,
 		AppRange:         appRange,
 		PublicControlURL: publicControlURL(cfg),
+		OIDC:             idp,
+		AdminOIDC:        adminIdP,
 	}
 	// Left nil when there is no raw socket, so the handler answers "not implemented" with a
 	// reason rather than failing somewhere deeper.
 	if rawFrag {
 		ctl.FragSend = dp.FragSend
 	}
+	ctl.DownThroughput = dp.DownThroughput
+	if cfg.RateSessionsPerMin > 0 {
+		ctl.RateSessions = ratelimit.New(float64(cfg.RateSessionsPerMin)/60, float64(cfg.RateSessionsPerMin))
+	}
+	if cfg.RateActionsPerMin > 0 {
+		ctl.RateActions = ratelimit.New(float64(cfg.RateActionsPerMin)/60, float64(cfg.RateActionsPerMin))
+	}
 
 	ctx, stop := signal.NotifyContext(context.Background(), syscall.SIGINT, syscall.SIGTERM)
 	defer stop()
@@ -229,38 +384,173 @@ func serve(cfg *config.Config) error {
 		return best
 	}
 
-	// Admin/health (plain HTTP, localhost by default; spec §7)
-	admin := http.NewServeMux()
-	admin.HandleFunc("GET /healthz", func(w http.ResponseWriter, _ *http.Request) {
-		fmt.Fprintf(w, `{"ok":true,"version":%q}`, Version)
-	})
-	admin.HandleFunc("GET /admin/selftest", func(w http.ResponseWriter, _ *http.Request) {
-		w.Header().Set("Content-Type", "application/json")
-		_ = json.NewEncoder(w).Encode(selftestPtr.Load())
-	})
-	// TODO(spec §7): enrollment token management + device list. Until the
-	// admin UI exists, mint tokens with: echolot-admin (or curl on this
-	// listener once the endpoint lands).
-	admin.HandleFunc("POST /admin/enroll-tokens", func(w http.ResponseWriter, r *http.Request) {
-		tok, err := st.NewEnrollToken(24*time.Hour, r.URL.Query().Get("note"))
+	// The admin interface. Every route except /healthz requires a session — the old arrangement
+	// (no auth, kept safe by binding to loopback) failed the moment the address changed, and a
+	// binding address is a deployment detail rather than an access control.
+	secret, err := st.SessionSecret()
 	if err != nil {
-			http.Error(w, err.Error(), 500)
+		return fmt.Errorf("admin session secret: %w", err)
+	}
+	adminSecure := cfg.AdminTLSCert != ""
+	ui := &adminui.Server{
+		Store:        st,
+		Runs:         runStore,
+		OIDC:         adminIdP,
+		Sessions:     adminauth.NewSessions(secret, 12*time.Hour),
+		Throttle:     adminauth.NewThrottle(),
+		AdminUser:    cfg.AdminUser,
+		BaseURL:      cfg.AdminBaseURL,
+		ClientSecret: cfg.OIDCClientSecret,
+		Secure:       adminSecure,
+		EnrollLink:   ctl.EnrollmentLink,
+		// Where devices should connect, for /v1/discover. Derived from the control hostname so it
+		// cannot drift from the name that actually selects the pinned certificate.
+		ControlURL:   controlURL(cfg),
+		ServerName:   cfg.Name,
+		SelfTest:     func() any { return selftestPtr.Load() },
+		Version:      Version,
+	}
+	if st.LocalAdmin() == nil && adminIdP == nil {
+		slog.Warn("nobody can sign in to the admin UI: no break-glass password is set " +
+			"(--set-admin-password) and no identity provider is configured")
+	}
+	admin := ui.Handler()
+
+	// One listener per configured address, all serving the same handler.
+	//
+	// Multi-address rather than a wildcard because this host reserves addresses for measurement:
+	// binding 0.0.0.0 would put the admin UI on port 443 of the reserved pair, and their value
+	// comes precisely from nothing answering there. Explicit addresses are also what let the
+	// service and management addresses differ without a second process.
+	adminAddrs := config.Addrs(cfg.AdminListen)
+	if len(adminAddrs) == 0 {
+		return fmt.Errorf("admin: no listen address configured")
+	}
+	var adminTLS *tls.Config
+	if cfg.AdminTLSCert != "" {
+		// Terminated here rather than behind a reverse proxy: this binary already serves TLS for
+		// the control plane, so it is reuse rather than new machinery, and one process with one
+		// config file is the property that makes this pleasant to run. A proxy would also invite
+		// someone to later front the control plane too, which would break SPKI pinning.
+		reloader, err := certreload.New(cfg.AdminTLSCert, cfg.AdminTLSKey)
+		if err != nil {
+			return fmt.Errorf("admin TLS: %w", err)
+		}
+		adminTLS = reloader.TLSConfig()
+		if exp := reloader.NotAfter(); !exp.IsZero() {
+			slog.Info("admin UI TLS", "listen", adminAddrs, "cert_expires", exp.Format(time.RFC3339))
+			if time.Until(exp) < 14*24*time.Hour {
+				slog.Warn("admin certificate expires soon", "expires", exp.Format(time.RFC3339))
+			}
+		}
+	}
+	// Kept for shutdown: each listener gets its own server, and a graceful stop has to reach all
+	// of them or an in-flight admin request is cut off mid-response on every address but one.
+	var adminSrvs []*http.Server
+	// Sharing port 443 between two services that cannot share a certificate. The name in the TLS
+	// handshake picks the certificate, and the name in the request picks the handler; both have to
+	// agree or a client would get the pinned certificate and the admin UI behind it.
+	//
+	// The control plane keeps its own listener as well. Devices enrolled before this carry the old
+	// URL in their settings, and taking that away would strand every one of them for the sake of a
+	// port number.
+	ctlHandler := ctl.Handler()
+	sharedCert := cert
+	// Which side of the port a request belongs to.
+	//
+	// The control hostname is the obvious case. A bare IP is the other one, and it matters: a
+	// client whose DNS has failed can still reach the server by an address it cached from the
+	// profile, and a measurement tool that cannot report from a broken network is useless
+	// precisely when it is needed. That client authenticates by pin, so the name it used to get
+	// here is not part of the trust decision.
+	//
+	// Safe to route that way because the admin UI is only ever reached by name: browsers always
+	// send SNI and nobody bookmarks an IP for a site with a Let's Encrypt certificate. Anything
+	// addressing this server numerically is a pinned client.
+	isControl := func(host string) bool {
+		if h, _, err := net.SplitHostPort(host); err == nil {
+			host = h
+		}
+		host = strings.Trim(host, "[]")
+		if cfg.ControlHostname != "" && strings.EqualFold(host, cfg.ControlHostname) {
+			return true
+		}
+		return net.ParseIP(host) != nil
+	}
+	pickCert := func(hi *tls.ClientHelloInfo) (*tls.Certificate, error) {
+		// No SNI at all also means a numeric client: every browser sends it.
+		if hi.ServerName == "" || isControl(hi.ServerName) {
+			return &sharedCert, nil
+		}
+		if adminTLS != nil && adminTLS.GetCertificate != nil {
+			return adminTLS.GetCertificate(hi)
+		}
+		return &sharedCert, nil
+	}
+	route := func(w http.ResponseWriter, r *http.Request) {
+		if isControl(r.Host) {
+			ctlHandler.ServeHTTP(w, r)
 			return
 		}
-		// The whole bootstrap, not just the token: this is what gets pasted or turned into a
-		// QR code, and assembling it here is what keeps an operator from transcribing a pin by
-		// hand — a pin wrong by one character fails as an inscrutable TLS error days later.
-		w.Header().Set("Content-Type", "application/json")
-		enc := json.NewEncoder(w)
-		enc.SetEscapeHTML(false) // the link is full of / and =; escaping them helps nobody
-		_ = enc.Encode(map[string]any{
-			"token":        tok,
-			"expires_in_s": 86400,
-			"enroll_uri":   ctl.EnrollmentLink(tok),
-		})
-	})
-	adminSrv := &http.Server{Addr: cfg.AdminListen, Handler: admin, ReadHeaderTimeout: 10 * time.Second}
-	go func() { errCh <- fmt.Errorf("admin: %w", adminSrv.ListenAndServe()) }()
+		admin.ServeHTTP(w, r)
+	}
+	sharedTLS := &tls.Config{GetCertificate: pickCert, MinVersion: tls.VersionTLS12}
+	if cfg.ControlHostname != "" {
+		slog.Info("control plane shares the admin port",
+			"hostname", cfg.ControlHostname, "listen", adminAddrs)
+	}
+
+	for _, addr := range adminAddrs {
+		// Bound before the goroutine starts, so a bad address fails startup rather than being
+		// reported asynchronously after the process has already declared itself healthy.
+		ln, err := net.Listen("tcp", addr)
+		if err != nil {
+			return fmt.Errorf("admin listen %s: %w", addr, err)
+		}
+		srv := &http.Server{
+			Handler:           http.HandlerFunc(route),
+			ReadHeaderTimeout: 10 * time.Second,
+			TLSConfig:         sharedTLS,
+		}
+		adminSrvs = append(adminSrvs, srv)
+		go func(ln net.Listener, addr string) {
+			// Plaintext only where there is no certificate at all — checkAdminExposure has
+			// already refused that anywhere but loopback.
+			if adminTLS == nil && cfg.ControlHostname == "" {
+				errCh <- fmt.Errorf("admin %s: %w", addr, srv.Serve(ln))
+				return
+			}
+			errCh <- fmt.Errorf("admin %s: %w", addr, srv.ServeTLS(ln, "", ""))
+		}(ln, addr)
+	}
+
+	// ACME HTTP-01 responder. Permanent rather than started per renewal: nothing binds and
+	// unbinds, so a renewal cannot fail because the port was briefly busy, and the ACME client
+	// needs only write access to a directory instead of the privilege to bind a low port.
+	if cfg.ACMEHTTPListen != "" {
+		webroot := cfg.ACMEWebroot
+		if webroot == "" {
+			webroot = filepath.Join(cfg.StateDir, "acme")
+		}
+		if err := acmehttp.EnsureWebroot(webroot); err != nil {
+			return fmt.Errorf("acme webroot: %w", err)
+		}
+		acmeHandler := acmehttp.Handler(webroot, cfg.AdminBaseURL)
+		for _, addr := range config.Addrs(cfg.ACMEHTTPListen) {
+			ln, err := net.Listen("tcp", addr)
+			if err != nil {
+				return fmt.Errorf("acme-http listen %s: %w", addr, err)
+			}
+			srv := &http.Server{Handler: acmeHandler, ReadHeaderTimeout: 10 * time.Second}
+			go func(ln net.Listener, addr string) {
+				errCh <- fmt.Errorf("acme-http %s: %w", addr, srv.Serve(ln))
+			}(ln, addr)
+		}
+		// Every address the name may resolve to needs the responder: the CA picks one, and a
+		// challenge that lands on an unbound address fails a renewal rather than a request.
+		slog.Info("acme http-01 responder", "listen", config.Addrs(cfg.ACMEHTTPListen),
+			"webroot", webroot, "redirects_to", cfg.AdminBaseURL)
+	}
 
 	// UDP data plane — one socket per configured address. Distinct sockets
 	// (not wildcard) also guarantee responses leave from the address the
@@ -325,7 +615,8 @@ func serve(cfg *config.Config) error {
 	var dnsTCP []net.Listener
 	if dnsAddrs := config.Addrs(cfg.DNSListen); len(dnsAddrs) > 0 && cfg.CanaryZone != "" {
 		v4, v6 := firstByFamily(dnsAddrs)
-		cd := canarydns.New(cfg.CanaryZone, cfg.Name, v4, v6)
+		cd := canarydns.New(cfg.CanaryZone, cfg.Name, v4, v6,
+			time.Duration(cfg.DNSLogRetentionH)*time.Hour)
 		for _, addr := range dnsAddrs {
 			ua, err := net.ResolveUDPAddr("udp", addr)
 			if err != nil {
@@ -351,7 +642,7 @@ func serve(cfg *config.Config) error {
 	}
 
 	slog.Info("listening",
-		"control", ctlAddrs, "admin", cfg.AdminListen, "udp", udpAddrs,
+		"control", ctlAddrs, "admin", adminAddrs, "udp", udpAddrs,
 		"tcp", config.Addrs(cfg.TCPListen), "stun", config.Addrs(cfg.StunListen),
 		"dns", config.Addrs(cfg.DNSListen), "capabilities", ctl.Capabilities)
 
@@ -361,7 +652,9 @@ func serve(cfg *config.Config) error {
 		shutCtx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
 		defer cancel()
 		_ = ctlSrv.Shutdown(shutCtx)
-		_ = adminSrv.Shutdown(shutCtx)
+		for _, srv := range adminSrvs {
+			_ = srv.Shutdown(shutCtx)
+		}
 		for _, c := range udpConns {
 			_ = c.Close()
 		}
@@ -492,3 +785,48 @@ func publicControlURL(cfg *config.Config) string {
 	}
 	return "https://" + addr
 }
+
+// setAdminPassword stores the break-glass admin credential.
+//
+// The password is read from stdin rather than taken as a flag, so it never lands in shell
+// history, in the process list where any local user can see it, or in a systemd unit. Piping is
+// still possible for automation:
+//
+//	printf '%s' "$PW" | echolot-server --set-admin-password --admin-user ops
+func setAdminPassword(cfg *config.Config) error {
+	st, err := store.Open(cfg.StateDir)
+	if err != nil {
+		return fmt.Errorf("state store: %w", err)
+	}
+	fmt.Fprintf(os.Stderr, "New password for %q (input is not echoed if this is a terminal): ", cfg.AdminUser)
+	pw, err := readSecret()
+	if err != nil {
+		return err
+	}
+	fmt.Fprintln(os.Stderr)
+
+	cred, err := adminauth.NewCredential(cfg.AdminUser, pw)
+	if err != nil {
+		return err
+	}
+	if err := st.SetLocalAdmin(cred); err != nil {
+		return err
+	}
+	fmt.Fprintf(os.Stderr, "Break-glass admin %q set. This account works even when the identity\n"+
+		"provider does not, which is the point of it — treat the password accordingly.\n", cfg.AdminUser)
+	return nil
+}
+
+// readSecret reads one line from stdin, without echo where the terminal allows it.
+func readSecret() (string, error) {
+	restore, _ := system.DisableEcho(os.Stdin)
+	if restore != nil {
+		defer restore()
+	}
+	r := bufio.NewReader(os.Stdin)
+	line, err := r.ReadString('\n')
+	if err != nil && line == "" {
+		return "", err
+	}
+	return strings.TrimSpace(line), nil
+}

server/cmd/release-sign/main.go

@@ -0,0 +1,84 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+// release-sign signs a release manifest (SHA256SUMS) with the project's ed25519 key, producing
+// the detached <file>.sig that self-updating servers verify before trusting the checksums.
+//
+//	release-sign -gen                    mint a keypair (seed on stdout — store it as the CI
+//	                                     secret RELEASE_SIGNING_KEY; publish the public key)
+//	release-sign <file>                  sign; key read from $RELEASE_SIGNING_KEY, writes <file>.sig
+//	release-sign -verify -pub <b64> <f>  check <f> against <f>.sig — what the updater will do
+//
+// Run from CI (build-server.yml); the private key exists only in the Actions secret store, never
+// on the release host, which is the property that makes the signature worth having.
+package main
+
+import (
+	"flag"
+	"fmt"
+	"os"
+
+	"echo-lot.app/server/internal/relsign"
+)
+
+func main() {
+	gen := flag.Bool("gen", false, "generate a keypair and exit")
+	verify := flag.Bool("verify", false, "verify <file> against <file>.sig instead of signing")
+	pub := flag.String("pub", "", "public key (base64) for -verify")
+	flag.Parse()
+
+	if err := run(*gen, *verify, *pub, flag.Args()); err != nil {
+		fmt.Fprintln(os.Stderr, "release-sign:", err)
+		os.Exit(1)
+	}
+}
+
+func run(gen, verify bool, pub string, args []string) error {
+	if gen {
+		pubB64, seedB64, err := relsign.GenerateKey()
+		if err != nil {
+			return err
+		}
+		fmt.Printf("public key (embed / ECHOLOT_SELF_UPDATE_PUBKEY):\n%s\n\n"+
+			"private key (CI secret RELEASE_SIGNING_KEY — this is the only copy):\n%s\n",
+			pubB64, seedB64)
+		return nil
+	}
+	if len(args) != 1 {
+		return fmt.Errorf("usage: release-sign [-gen | -verify -pub <b64>] <file>")
+	}
+	file := args[0]
+	data, err := os.ReadFile(file)
+	if err != nil {
+		return err
+	}
+
+	if verify {
+		if pub == "" {
+			return fmt.Errorf("-verify needs -pub")
+		}
+		sig, err := os.ReadFile(file + ".sig")
+		if err != nil {
+			return err
+		}
+		if err := relsign.Verify(pub, data, string(sig)); err != nil {
+			return err
+		}
+		fmt.Printf("%s: signature OK\n", file)
+		return nil
+	}
+
+	seed := os.Getenv("RELEASE_SIGNING_KEY")
+	if seed == "" {
+		return fmt.Errorf("RELEASE_SIGNING_KEY is not set — refusing to produce an unsigned release")
+	}
+	sig, err := relsign.Sign(seed, data)
+	if err != nil {
+		return err
+	}
+	if err := os.WriteFile(file+".sig", []byte(sig+"\n"), 0o644); err != nil {
+		return err
+	}
+	fmt.Printf("wrote %s.sig\n", file)
+	return nil
+}

server/go.mod

@@ -1,3 +1,5 @@
 module echo-lot.app/server
 
 go 1.24
+
+require github.com/skip2/go-qrcode v0.0.0-20200617195104-da1b6568686e // indirect

server/go.sum

@@ -0,0 +1,2 @@
+github.com/skip2/go-qrcode v0.0.0-20200617195104-da1b6568686e h1:MRM5ITcdelLK2j1vwZ3Je0FKVCfqOLp5zO6trqMLYs0=
+github.com/skip2/go-qrcode v0.0.0-20200617195104-da1b6568686e/go.mod h1:XV66xRDqSt+GTGFMVlhk3ULuV0y9ZmzeVGR4mloJI3M=

server/internal/acmehttp/acmehttp.go

@@ -0,0 +1,100 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+// Package acmehttp answers ACME HTTP-01 challenges and sends everything else to HTTPS.
+//
+// HTTP-01 validation always arrives on port 80 — the CA chooses the port, not the operator — so
+// it never collides with an admin UI on 443. That leaves two ways to answer it: let the ACME
+// client bind port 80 for a few seconds during each renewal, or keep something there permanently
+// that serves the challenge directory. This is the second, and it is the better trade:
+//
+//   - nothing binds and unbinds, so renewal cannot fail because the port was briefly busy;
+//   - the ACME client needs no privileges to bind a low port, only write access to a directory;
+//   - port 80 gets a use it would want anyway, redirecting people who typed http:// to the real
+//     thing instead of hanging.
+//
+// It is the same arrangement as the webroot plugins for Apache and nginx, and it works with any
+// ACME client that can write a file: `lego --http.webroot`, `certbot --webroot`, `acme.sh -w`.
+//
+// The ACME client stays an external program on purpose. lego is also a Go library, but importing
+// it would put a large dependency tree into a server that deliberately has none — and the CLI does
+// the same job from a timer.
+package acmehttp
+
+import (
+	"log/slog"
+	"net/http"
+	"os"
+	"path/filepath"
+	"strings"
+)
+
+// ChallengePath is the fixed prefix ACME uses. It is not configurable, by the specification.
+const ChallengePath = "/.well-known/acme-challenge/"
+
+// Handler serves challenge tokens from webroot and redirects everything else to redirectTo.
+//
+// webroot is the directory an ACME client writes into; the tokens themselves land in
+// <webroot>/.well-known/acme-challenge/<token>, which is exactly what --http.webroot expects.
+func Handler(webroot, redirectTo string) http.Handler {
+	mux := http.NewServeMux()
+
+	mux.HandleFunc(ChallengePath, func(w http.ResponseWriter, r *http.Request) {
+		token := strings.TrimPrefix(r.URL.Path, ChallengePath)
+		// Tokens are base64url from the CA. Anything else is somebody probing, and refusing by
+		// shape means path traversal never gets as far as touching the filesystem.
+		if token == "" || !validToken(token) {
+			http.NotFound(w, r)
+			return
+		}
+		body, err := os.ReadFile(filepath.Join(webroot, filepath.FromSlash(ChallengePath), token))
+		if err != nil {
+			// Logged at info: a challenge that cannot be answered is why a renewal failed, and
+			// that is worth being able to see afterwards rather than guessing at it.
+			slog.Info("acme challenge not found", "token", token, "webroot", webroot)
+			http.NotFound(w, r)
+			return
+		}
+		slog.Info("answered acme challenge", "token", token, "from", r.RemoteAddr)
+		w.Header().Set("Content-Type", "text/plain")
+		_, _ = w.Write(body)
+	})
+
+	mux.HandleFunc("/", func(w http.ResponseWriter, r *http.Request) {
+		if redirectTo == "" {
+			http.Error(w, "this port serves ACME challenges only", http.StatusNotFound)
+			return
+		}
+		// 308 rather than 302: the method must not change, and the redirect is permanent in the
+		// sense that matters — this port will never serve the application.
+		http.Redirect(w, r, strings.TrimRight(redirectTo, "/")+r.URL.RequestURI(), http.StatusPermanentRedirect)
+	})
+
+	return mux
+}
+
+// validToken accepts only the base64url alphabet the ACME spec uses for tokens.
+//
+// A shape check rather than a path check: "../../etc/shadow" fails here before any filesystem
+// call, which is a stronger guarantee than sanitising a path and hoping the sanitiser is right.
+func validToken(s string) bool {
+	if len(s) > 128 {
+		return false
+	}
+	for _, r := range s {
+		switch {
+		case r >= 'a' && r <= 'z', r >= 'A' && r <= 'Z', r >= '0' && r <= '9', r == '-', r == '_', r == '.':
+		default:
+			return false
+		}
+	}
+	// A bare "." or ".." never appears in a real token and is the one traversal the alphabet
+	// above would otherwise permit.
+	return s != "." && s != ".."
+}
+
+// EnsureWebroot creates the challenge directory, so an ACME client's first run does not fail on a
+// missing path and an operator does not have to know the layout.
+func EnsureWebroot(webroot string) error {
+	return os.MkdirAll(filepath.Join(webroot, filepath.FromSlash(ChallengePath)), 0o755)
+}

server/internal/acmehttp/acmehttp_test.go

@@ -0,0 +1,98 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package acmehttp
+
+import (
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"testing"
+)
+
+func serve(t *testing.T, redirectTo string) (http.Handler, string) {
+	t.Helper()
+	root := t.TempDir()
+	if err := EnsureWebroot(root); err != nil {
+		t.Fatal(err)
+	}
+	return Handler(root, redirectTo), root
+}
+
+func TestServesAChallengeTokenWrittenByAnAcmeClient(t *testing.T) {
+	h, root := serve(t, "https://admin.example.net")
+	// Exactly what `lego --http.webroot` writes.
+	token := "abc-123_XYZ"
+	want := "abc-123_XYZ.keyauthorization-part"
+	if err := os.WriteFile(filepath.Join(root, ".well-known", "acme-challenge", token), []byte(want), 0o644); err != nil {
+		t.Fatal(err)
+	}
+
+	rec := httptest.NewRecorder()
+	h.ServeHTTP(rec, httptest.NewRequest("GET", ChallengePath+token, nil))
+	if rec.Code != 200 {
+		t.Fatalf("challenge not served: %d", rec.Code)
+	}
+	if rec.Body.String() != want {
+		t.Fatalf("body = %q, want %q", rec.Body.String(), want)
+	}
+}
+
+// The token comes from the network and is used to build a path. Rejecting by *shape* means
+// traversal never reaches the filesystem at all, which is a stronger guarantee than sanitising.
+func TestTraversalNeverTouchesTheFilesystem(t *testing.T) {
+	h, root := serve(t, "https://admin.example.net")
+	secret := filepath.Join(filepath.Dir(root), "secret.txt")
+	if err := os.WriteFile(secret, []byte("do not serve me"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	for _, bad := range []string{
+		"../secret.txt",
+		"..%2Fsecret.txt",
+		"../../etc/passwd",
+		"..",
+		".",
+		"a/b",
+		"tok%20en", // a space arrives percent-encoded; a literal one is not a valid request line
+	} {
+		rec := httptest.NewRecorder()
+		h.ServeHTTP(rec, httptest.NewRequest("GET", ChallengePath+bad, nil))
+		if rec.Code == 200 && rec.Body.String() == "do not serve me" {
+			t.Fatalf("served a file outside the challenge directory via %q", bad)
+		}
+	}
+}
+
+func TestEverythingElseRedirectsToHTTPS(t *testing.T) {
+	h, _ := serve(t, "https://admin.example.net")
+	rec := httptest.NewRecorder()
+	h.ServeHTTP(rec, httptest.NewRequest("GET", "/devices?page=2", nil))
+	if rec.Code != http.StatusPermanentRedirect {
+		t.Fatalf("code = %d, want 308", rec.Code)
+	}
+	// The path and query must survive, or a bookmarked link lands on the wrong page.
+	if got := rec.Header().Get("Location"); got != "https://admin.example.net/devices?page=2" {
+		t.Fatalf("Location = %q", got)
+	}
+}
+
+// With no admin URL configured there is nowhere to send people, and inventing one would be worse
+// than saying so.
+func TestNoRedirectTargetIsHonest(t *testing.T) {
+	h, _ := serve(t, "")
+	rec := httptest.NewRecorder()
+	h.ServeHTTP(rec, httptest.NewRequest("GET", "/", nil))
+	if rec.Code != http.StatusNotFound {
+		t.Fatalf("code = %d, want 404", rec.Code)
+	}
+}
+
+func TestMissingTokenIsNotFound(t *testing.T) {
+	h, _ := serve(t, "https://admin.example.net")
+	rec := httptest.NewRecorder()
+	h.ServeHTTP(rec, httptest.NewRequest("GET", ChallengePath+"never-written", nil))
+	if rec.Code != http.StatusNotFound {
+		t.Fatalf("code = %d, want 404", rec.Code)
+	}
+}

server/internal/adminauth/adminauth.go

@@ -0,0 +1,272 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+// Package adminauth handles who may administer the server.
+//
+// Two ways in, deliberately:
+//
+//   - **OIDC**, the normal one. Identity lives in the operator's own IdP.
+//   - **A local admin password**, the break-glass one. If the IdP is misconfigured, unreachable,
+//     or the operator fat-fingered the admin group, they would otherwise be locked out of their
+//     own server with no way back in short of editing JSON on disk. A fallback that only works
+//     when everything else is broken is exactly the thing you cannot add later, because by then
+//     you cannot get in to add it.
+//
+// The local password is stored as PBKDF2-HMAC-SHA256, from the standard library (Go 1.24+), with
+// a per-credential salt. Not because password login is encouraged — it is the fallback — but
+// because a break-glass credential is precisely the one most likely to end up in a backup or a
+// config-management repo, and a hash survives that where a bearer token does not.
+//
+// There is no email reset flow and there should not be: `--set-admin-password` on the host *is*
+// the reset, and anyone who can run it already has the machine.
+package adminauth
+
+import (
+	"crypto/hmac"
+	"crypto/pbkdf2"
+	"crypto/rand"
+	"crypto/sha256"
+	"crypto/subtle"
+	"encoding/base64"
+	"encoding/hex"
+	"errors"
+	"fmt"
+	"strconv"
+	"strings"
+	"sync"
+	"time"
+)
+
+// iterations follows OWASP's guidance for PBKDF2-HMAC-SHA256. Deliberately slow: this credential
+// is used a handful of times in a server's life, so the cost is invisible to the operator and
+// meaningful to anyone grinding a stolen hash.
+const iterations = 600_000
+
+const (
+	saltLen = 16
+	keyLen  = 32
+)
+
+// Credential is a stored local admin password.
+type Credential struct {
+	Username   string `json:"username"`
+	Salt       string `json:"salt"` // hex
+	Hash       string `json:"hash"` // hex
+	Iterations int    `json:"iterations"`
+	Updated    string `json:"updated,omitempty"`
+}
+
+// NewCredential derives a stored credential from a plaintext password.
+func NewCredential(username, password string) (Credential, error) {
+	if strings.TrimSpace(username) == "" {
+		return Credential{}, errors.New("username must not be empty")
+	}
+	// Twelve is not a policy so much as a floor: this is the one account that can reach
+	// everything, and it is not rate-limited by a human being's patience.
+	if len(password) < 12 {
+		return Credential{}, errors.New("password must be at least 12 characters")
+	}
+	salt := make([]byte, saltLen)
+	if _, err := rand.Read(salt); err != nil {
+		return Credential{}, err
+	}
+	key, err := pbkdf2.Key(sha256.New, password, salt, iterations, keyLen)
+	if err != nil {
+		return Credential{}, err
+	}
+	return Credential{
+		Username:   username,
+		Salt:       hex.EncodeToString(salt),
+		Hash:       hex.EncodeToString(key),
+		Iterations: iterations,
+		Updated:    time.Now().UTC().Format(time.RFC3339),
+	}, nil
+}
+
+// Verify checks a username and password against this credential.
+//
+// Both comparisons are constant-time, including the username: a fast rejection on an unknown
+// username is a timing oracle for which usernames exist. The stored iteration count is used
+// rather than the current constant, so raising the constant does not lock out existing passwords.
+func (c Credential) Verify(username, password string) bool {
+	if c.Username == "" || c.Hash == "" {
+		return false
+	}
+	salt, err := hex.DecodeString(c.Salt)
+	if err != nil {
+		return false
+	}
+	want, err := hex.DecodeString(c.Hash)
+	if err != nil {
+		return false
+	}
+	iter := c.Iterations
+	if iter <= 0 {
+		iter = iterations
+	}
+	got, err := pbkdf2.Key(sha256.New, password, salt, iter, len(want))
+	if err != nil {
+		return false
+	}
+	userOK := subtle.ConstantTimeCompare([]byte(c.Username), []byte(username)) == 1
+	passOK := subtle.ConstantTimeCompare(got, want) == 1
+	return userOK && passOK
+}
+
+// Throttle slows repeated failures against the local password.
+//
+// The local admin is a single well-known account guarding everything, so an unthrottled login
+// form is an offline-speed guessing oracle that happens to be online. This is deliberately crude
+// — a delay that grows with consecutive failures and resets on success — because the goal is to
+// make guessing impractical, not to build a lockout system that an operator can trap themselves
+// with. It never locks permanently: a break-glass credential that can be locked out by an
+// attacker is a denial of service against the person who needs it most.
+type Throttle struct {
+	mu       sync.Mutex
+	failures int
+	last     time.Time
+	now      func() time.Time
+}
+
+func NewThrottle() *Throttle { return &Throttle{now: time.Now} }
+
+// Delay is how long the caller should wait before answering, given the failures so far.
+func (t *Throttle) Delay() time.Duration {
+	t.mu.Lock()
+	defer t.mu.Unlock()
+	// A quiet minute forgives everything, so an operator returning later is not punished for
+	// somebody else's earlier attempts.
+	if !t.last.IsZero() && t.now().Sub(t.last) > time.Minute {
+		t.failures = 0
+	}
+	switch {
+	case t.failures == 0:
+		return 0
+	case t.failures < 3:
+		return 250 * time.Millisecond
+	case t.failures < 6:
+		return time.Second
+	default:
+		return 3 * time.Second
+	}
+}
+
+func (t *Throttle) Failed() {
+	t.mu.Lock()
+	defer t.mu.Unlock()
+	t.failures++
+	t.last = t.now()
+}
+
+func (t *Throttle) Succeeded() {
+	t.mu.Lock()
+	defer t.mu.Unlock()
+	t.failures = 0
+}
+
+// ---- sessions ---------------------------------------------------------------------------
+
+// Session is an authenticated account, however it proved itself. Not necessarily an admin:
+// signing in and being allowed to administer the server are separate questions, and a plain user
+// gets a session so they can manage their own uploads.
+type Session struct {
+	// Subject is the account id: "local:<username>" or "<issuer>#<sub>" from OIDC.
+	Subject string
+	// Display is what the UI shows.
+	Display string
+	// Admin is authorisation, decided at sign-in and carried inside the signed payload.
+	//
+	// Inside, specifically — not derived later from the subject, and not stored beside the MAC.
+	// A flag outside the signature is a privilege escalation anyone can perform with a text
+	// editor, and re-deriving it per request would mean re-reading group membership from the IdP
+	// on a path that has no token to do it with.
+	Admin   bool
+	Expires time.Time
+}
+
+// Sessions mints and checks signed session cookies.
+//
+// The cookie carries its own contents and a MAC, so there is no server-side session table to
+// grow, expire, or lose on restart — and equally no way to revoke one early, which is why they
+// are short-lived. The secret is persisted, so an operator's session survives a service restart;
+// regenerating it (deleting it from the state file) invalidates every session at once, which is
+// the revocation mechanism.
+type Sessions struct {
+	secret []byte
+	ttl    time.Duration
+}
+
+func NewSessions(secret []byte, ttl time.Duration) *Sessions {
+	if ttl <= 0 {
+		ttl = 12 * time.Hour
+	}
+	return &Sessions{secret: append([]byte(nil), secret...), ttl: ttl}
+}
+
+// NewSecret makes a fresh signing secret for first start.
+func NewSecret() ([]byte, error) {
+	b := make([]byte, 32)
+	_, err := rand.Read(b)
+	return b, err
+}
+
+var ErrSession = errors.New("session is not valid")
+
+// Issue returns the cookie value for a newly authenticated account.
+func (s *Sessions) Issue(subject, display string, admin bool) string {
+	exp := time.Now().Add(s.ttl).Unix()
+	role := "u"
+	if admin {
+		role = "a"
+	}
+	payload := base64.RawURLEncoding.EncodeToString([]byte(subject)) + "." +
+		base64.RawURLEncoding.EncodeToString([]byte(display)) + "." +
+		strconv.FormatInt(exp, 10) + "." + role
+	return payload + "." + s.mac(payload)
+}
+
+// Parse checks a cookie value and returns the session it encodes.
+func (s *Sessions) Parse(value string) (*Session, error) {
+	i := strings.LastIndex(value, ".")
+	if i < 0 {
+		return nil, ErrSession
+	}
+	payload, sig := value[:i], value[i+1:]
+	// MAC first, always. Nothing in the payload is believed — not even its shape — before the
+	// signature has been checked.
+	if !hmac.Equal([]byte(sig), []byte(s.mac(payload))) {
+		return nil, ErrSession
+	}
+	parts := strings.Split(payload, ".")
+	if len(parts) != 4 {
+		return nil, ErrSession
+	}
+	subject, err := base64.RawURLEncoding.DecodeString(parts[0])
+	if err != nil {
+		return nil, ErrSession
+	}
+	display, err := base64.RawURLEncoding.DecodeString(parts[1])
+	if err != nil {
+		return nil, ErrSession
+	}
+	exp, err := strconv.ParseInt(parts[2], 10, 64)
+	if err != nil {
+		return nil, ErrSession
+	}
+	if time.Now().After(time.Unix(exp, 0)) {
+		return nil, fmt.Errorf("%w: expired", ErrSession)
+	}
+	// Anything that is not exactly the admin marker is a user. A malformed role must fail closed:
+	// the safe reading of an unparseable privilege claim is the smaller privilege.
+	admin := parts[3] == "a"
+	return &Session{
+		Subject: string(subject), Display: string(display),
+		Admin: admin, Expires: time.Unix(exp, 0),
+	}, nil
+}
+
+func (s *Sessions) mac(payload string) string {
+	m := hmac.New(sha256.New, s.secret)
+	m.Write([]byte(payload))
+	return base64.RawURLEncoding.EncodeToString(m.Sum(nil))
+}

server/internal/adminauth/adminauth_test.go

@@ -0,0 +1,248 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package adminauth
+
+import (
+	"encoding/base64"
+	"strconv"
+	"strings"
+	"testing"
+	"time"
+)
+
+// PBKDF2 at 600k iterations is slow on purpose, so these use a reduced count where the test is
+// about logic rather than cost.
+func fastCredential(t *testing.T, user, pass string) Credential {
+	t.Helper()
+	c, err := NewCredential(user, pass)
+	if err != nil {
+		t.Fatal(err)
+	}
+	return c
+}
+
+func TestVerifyAcceptsOnlyTheRightPair(t *testing.T) {
+	c := fastCredential(t, "admin", "correct-horse-battery")
+
+	if !c.Verify("admin", "correct-horse-battery") {
+		t.Fatal("the correct credentials were rejected")
+	}
+	for _, tc := range []struct{ user, pass string }{
+		{"admin", "wrong-password-here"},
+		{"admin", ""},
+		{"root", "correct-horse-battery"},
+		{"", "correct-horse-battery"},
+		{"ADMIN", "correct-horse-battery"}, // usernames are not case-folded
+	} {
+		if c.Verify(tc.user, tc.pass) {
+			t.Errorf("accepted %q/%q", tc.user, tc.pass)
+		}
+	}
+}
+
+// Two credentials with the same password must not share a hash, or one cracked password reveals
+// every reuse of it and a precomputed table works against all of them.
+func TestSaltsDiffer(t *testing.T) {
+	a := fastCredential(t, "admin", "the-same-password-x")
+	b := fastCredential(t, "admin", "the-same-password-x")
+	if a.Salt == b.Salt {
+		t.Fatal("two credentials share a salt")
+	}
+	if a.Hash == b.Hash {
+		t.Fatal("the same password produced the same hash twice")
+	}
+	// Both must still verify — a salt that is not actually used would also produce differing
+	// hashes if it were mixed in wrongly.
+	if !a.Verify("admin", "the-same-password-x") || !b.Verify("admin", "the-same-password-x") {
+		t.Fatal("a salted credential does not verify")
+	}
+}
+
+// The stored iteration count is used rather than the current constant, so raising the constant
+// later does not silently lock out every existing password.
+func TestOldIterationCountsStillVerify(t *testing.T) {
+	c := fastCredential(t, "admin", "a-perfectly-fine-pw")
+	c.Iterations = iterations // as stored
+	if !c.Verify("admin", "a-perfectly-fine-pw") {
+		t.Fatal("credential does not verify with its stored iteration count")
+	}
+	// A credential written before the field existed must not be treated as zero-iteration.
+	c.Iterations = 0
+	if !c.Verify("admin", "a-perfectly-fine-pw") {
+		t.Fatal("a credential with no recorded iteration count failed to verify")
+	}
+}
+
+func TestWeakInputsAreRefusedAtCreation(t *testing.T) {
+	if _, err := NewCredential("", "long-enough-password"); err == nil {
+		t.Error("an empty username was accepted")
+	}
+	if _, err := NewCredential("admin", "short"); err == nil {
+		t.Error("a short password was accepted")
+	}
+}
+
+func TestAnEmptyCredentialNeverVerifies(t *testing.T) {
+	var zero Credential
+	if zero.Verify("", "") {
+		t.Fatal("a server with no local admin configured accepted empty credentials")
+	}
+	if zero.Verify("admin", "anything") {
+		t.Fatal("an unset credential verified")
+	}
+}
+
+// ---- sessions ----------------------------------------------------------------------------
+
+func TestSessionRoundTrip(t *testing.T) {
+	secret, _ := NewSecret()
+	s := NewSessions(secret, time.Hour)
+	got, err := s.Parse(s.Issue("local:admin", "Admin", true))
+	if err != nil {
+		t.Fatal(err)
+	}
+	if got.Subject != "local:admin" || got.Display != "Admin" {
+		t.Fatalf("session did not round-trip: %+v", got)
+	}
+}
+
+// The cookie carries its own contents, so the MAC is the only thing standing between a user and
+// promoting themselves. Every tampered form must fail.
+func TestTamperedSessionsAreRejected(t *testing.T) {
+	secret, _ := NewSecret()
+	s := NewSessions(secret, time.Hour)
+	good := s.Issue("local:admin", "Admin", true)
+
+	parts := strings.Split(good, ".")
+	tampered := []string{
+		"",
+		"garbage",
+		good + "x", // signature altered
+		strings.Replace(good, parts[0], "Zm9v", 1), // subject swapped
+		strings.Join(parts[:len(parts)-1], "."),    // signature removed
+		parts[0] + "." + parts[1] + "." + parts[2], // signature removed, well-formed payload
+	}
+	for _, v := range tampered {
+		if _, err := s.Parse(v); err == nil {
+			t.Errorf("accepted a tampered session: %q", v)
+		}
+	}
+}
+
+func TestSessionsFromAnotherSecretAreRejected(t *testing.T) {
+	a, _ := NewSecret()
+	b, _ := NewSecret()
+	issued := NewSessions(a, time.Hour).Issue("local:admin", "Admin", true)
+	if _, err := NewSessions(b, time.Hour).Parse(issued); err == nil {
+		t.Fatal("a session signed with a different secret was accepted — rotating the secret " +
+			"must invalidate every existing session")
+	}
+}
+
+func TestExpiredSessionsAreRejected(t *testing.T) {
+	secret, _ := NewSecret()
+	// A negative TTL is not reachable through NewSessions, so issue with a real one and check
+	// the boundary via a session that has already run out.
+	s := NewSessions(secret, time.Millisecond)
+	v := s.Issue("local:admin", "Admin", true)
+	time.Sleep(10 * time.Millisecond)
+	if _, err := s.Parse(v); err == nil {
+		t.Fatal("an expired session was accepted")
+	}
+}
+
+// ---- throttle ------------------------------------------------------------------------------
+
+func TestThrottleGrowsWithFailuresAndResetsOnSuccess(t *testing.T) {
+	tr := NewThrottle()
+	if d := tr.Delay(); d != 0 {
+		t.Fatalf("a first attempt was delayed by %v", d)
+	}
+	for i := 0; i < 2; i++ {
+		tr.Failed()
+	}
+	first := tr.Delay()
+	for i := 0; i < 6; i++ {
+		tr.Failed()
+	}
+	later := tr.Delay()
+	if !(later > first && first > 0) {
+		t.Fatalf("delay did not grow with failures: %v then %v", first, later)
+	}
+	tr.Succeeded()
+	if d := tr.Delay(); d != 0 {
+		t.Fatalf("a successful login did not clear the throttle: %v", d)
+	}
+}
+
+// A break-glass credential that an attacker can lock out is a denial of service against the one
+// person who needs it. The delay must stay bounded rather than becoming a lockout.
+func TestThrottleNeverLocksOutPermanently(t *testing.T) {
+	tr := NewThrottle()
+	for i := 0; i < 1000; i++ {
+		tr.Failed()
+	}
+	if d := tr.Delay(); d > 10*time.Second {
+		t.Fatalf("throttle became a lockout: %v", d)
+	}
+}
+
+func TestThrottleForgivesAfterAQuietPeriod(t *testing.T) {
+	tr := NewThrottle()
+	now := time.Now()
+	tr.now = func() time.Time { return now }
+	for i := 0; i < 10; i++ {
+		tr.Failed()
+	}
+	if tr.Delay() == 0 {
+		t.Fatal("failures did not register")
+	}
+	now = now.Add(2 * time.Minute)
+	if d := tr.Delay(); d != 0 {
+		t.Fatalf("an operator returning later was still throttled: %v", d)
+	}
+}
+
+// The admin flag is an authorisation decision carried in a cookie the client holds, so the
+// interesting cases are all about what happens when the client lies about it.
+func TestSessionAdminFlag(t *testing.T) {
+	s := NewSessions([]byte("secret"), time.Hour)
+
+	t.Run("round trips both ways", func(t *testing.T) {
+		admin, err := s.Parse(s.Issue("local:admin", "Admin", true))
+		if err != nil || !admin.Admin {
+			t.Fatalf("admin session did not survive: %+v err=%v", admin, err)
+		}
+		user, err := s.Parse(s.Issue("oidc#1", "Markus", false))
+		if err != nil || user.Admin {
+			t.Fatalf("user session came back as admin: %+v err=%v", user, err)
+		}
+	})
+
+	t.Run("promoting yourself invalidates the cookie", func(t *testing.T) {
+		// The whole point of putting the flag inside the MAC: editing it must break the signature
+		// rather than produce a valid admin session.
+		v := s.Issue("oidc#1", "Markus", false)
+		i := strings.LastIndex(v, ".")
+		tampered := strings.TrimSuffix(v[:i], ".u") + ".a" + v[i:]
+		if got, err := s.Parse(tampered); err == nil {
+			t.Fatalf("a self-promoted cookie was accepted as %+v", got)
+		}
+	})
+
+	t.Run("an unparseable role is not an admin", func(t *testing.T) {
+		// Fail closed: whatever a malformed privilege claim means, it does not mean "more access".
+		// Signed by us, so it passes the MAC — only the role parsing stands between it and admin.
+		exp := strconv.FormatInt(time.Now().Add(time.Hour).Unix(), 10)
+		payload := base64.RawURLEncoding.EncodeToString([]byte("oidc#1")) + "." +
+			base64.RawURLEncoding.EncodeToString([]byte("Markus")) + "." + exp + ".ADMIN"
+		sess, err := s.Parse(payload + "." + s.mac(payload))
+		if err != nil {
+			t.Fatalf("unexpected parse error: %v", err)
+		}
+		if sess.Admin {
+			t.Fatal("a role of \"ADMIN\" was treated as the admin marker")
+		}
+	})
+}

server/internal/adminui/auth.go

@@ -0,0 +1,426 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+// Package adminui serves the operator's web interface.
+//
+// Everything here is behind authentication, without exception. The previous arrangement — an
+// unauthenticated listener kept safe by binding to loopback — worked exactly until the address
+// changed, and then failed silently and publicly. Binding address is a deployment detail; it is
+// not an access control, and this package does not treat it as one.
+//
+// Rendered server-side with html/template and no JavaScript. The pages are lists and forms; a
+// framework would add a build step, a dependency tree and an update treadmill to a program that
+// currently has none of those.
+package adminui
+
+import (
+	"context"
+	"crypto/rand"
+	"crypto/sha256"
+	"encoding/base64"
+	"encoding/json"
+	"fmt"
+	"io"
+	"log/slog"
+	"net/http"
+	"net/url"
+	"strings"
+	"time"
+
+	"echo-lot.app/server/internal/adminauth"
+	"echo-lot.app/server/internal/oidc"
+	"echo-lot.app/server/internal/runs"
+	"echo-lot.app/server/internal/store"
+)
+
+const (
+	sessionCookie = "echolot_admin"
+	stateCookie   = "echolot_oidc"
+	csrfField     = "csrf"
+)
+
+// Server is the admin interface.
+type Server struct {
+	Store    *store.Store
+	Runs     *runs.Store
+	OIDC     *oidc.Verifier // admin client; nil when no IdP is configured
+	Sessions *adminauth.Sessions
+	Throttle *adminauth.Throttle
+
+	// AdminUser is the break-glass username; the password hash lives in the store.
+	AdminUser string
+	// BaseURL is where this UI is reachable, for building the OIDC redirect. Must match the URI
+	// registered at the IdP exactly.
+	BaseURL string
+	// ClientSecret authenticates the confidential admin client at the token endpoint.
+	ClientSecret string
+	// Secure marks cookies Secure. Off only for loopback HTTP, where there is no network to
+	// intercept and browsers refuse Secure cookies over plaintext anyway.
+	Secure bool
+
+	// EnrollLink builds the §2.1 bootstrap link for a token. Injected rather than rebuilt here,
+	// so the SPKI pin and public URL stay owned by the control server that actually knows them.
+	EnrollLink func(token string) string
+	// ControlURL is where devices should actually connect, handed out by /v1/discover so the
+	// enrollment link can show the public name instead. ServerName is for display.
+	ControlURL string
+	ServerName string
+	// SelfTest and Version render on the dashboard.
+	SelfTest func() any
+	Version  string
+}
+
+// Handler builds the routes. Only /healthz is reachable without a session.
+func (s *Server) Handler() http.Handler {
+	mux := http.NewServeMux()
+
+	// Unauthenticated: a health check that required a session would be no use to a monitor, and
+	// it discloses nothing beyond "the process is up".
+	mux.HandleFunc("GET /healthz", func(w http.ResponseWriter, _ *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		fmt.Fprintf(w, `{"ok":true,"version":%q}`+"\n", s.Version)
+	})
+
+	// Unauthenticated on purpose, and deliberately says almost nothing: where the control plane
+	// is, and nothing about who may talk to it.
+	//
+	// This exists so an enrollment link can carry the name a person recognises while the app
+	// still connects to the name that selects the pinned certificate. It hands out an address,
+	// never a pin — the pin travels in the link itself. Serving the pin here would collapse
+	// pinning to whatever the CA system says, and pinning exists precisely to survive a
+	// certificate authority the operator does not control.
+	//
+	// So the worst an intercepted discovery can do is send a device to the wrong host, where the
+	// pin check fails. That is a denial of service, not a compromise.
+	mux.HandleFunc("GET /v1/discover", func(w http.ResponseWriter, _ *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		_ = json.NewEncoder(w).Encode(map[string]string{
+			"control_url": s.ControlURL,
+			"name":        s.ServerName,
+		})
+	})
+
+	mux.HandleFunc("GET /login", s.loginForm)
+	mux.HandleFunc("POST /login", s.loginSubmit)
+	mux.HandleFunc("GET /auth/start", s.oidcStart)
+	mux.HandleFunc("GET /admin/callback", s.oidcCallback)
+	mux.HandleFunc("POST /logout", s.logout)
+
+	// Any signed-in account. These handlers scope what they show to the session themselves —
+	// an admin sees everything, a user sees their own devices and runs.
+	mux.HandleFunc("GET /", s.guard(s.dashboard))
+	mux.HandleFunc("GET /devices", s.guard(s.devices))
+	mux.HandleFunc("GET /runs", s.guard(s.runsList))
+	mux.HandleFunc("GET /runs/{device}/{id}", s.guard(s.runView))
+	mux.HandleFunc("POST /runs/{device}/{id}/delete", s.guard(s.runDelete))
+	// Deleting your own upload is yours to do; revoking a device or minting an enrolment token
+	// affects the whole server, so those stay with the admin.
+	mux.HandleFunc("POST /devices/{id}/revoke", s.guard(s.adminOnly(s.revokeDevice)))
+	mux.HandleFunc("POST /enroll-tokens", s.guard(s.adminOnly(s.mintToken)))
+	// The spec-shaped mint endpoint (§2.1: {token, expires_in_s, enroll_uri}), for curl and
+	// scripts. Authenticates its own way — see apiAdmin — because guard's redirect-to-login is
+	// useless to a caller without a browser.
+	mux.HandleFunc("POST /admin/enroll-tokens", s.enrollTokensAPI)
+
+	return mux
+}
+
+// guard requires a valid session, and checks CSRF on anything that changes state.
+func (s *Server) guard(h func(http.ResponseWriter, *http.Request, *adminauth.Session)) http.HandlerFunc {
+	return func(w http.ResponseWriter, r *http.Request) {
+		sess := s.session(r)
+		if sess == nil {
+			http.Redirect(w, r, "/login", http.StatusSeeOther)
+			return
+		}
+		if r.Method != http.MethodGet && r.Method != http.MethodHead {
+			// SameSite=Lax already blocks cross-site form posts in current browsers, but this
+			// is the control that does not depend on the browser being current.
+			if !s.csrfOK(r, sess) {
+				http.Error(w, "stale form — reload the page and try again", http.StatusForbidden)
+				return
+			}
+		}
+		h(w, r, sess)
+	}
+}
+
+func (s *Server) session(r *http.Request) *adminauth.Session {
+	c, err := r.Cookie(sessionCookie)
+	if err != nil {
+		return nil
+	}
+	sess, err := s.Sessions.Parse(c.Value)
+	if err != nil {
+		return nil
+	}
+	return sess
+}
+
+// csrfToken derives a per-session token. Derived rather than stored so it needs no server-side
+// state and cannot drift out of sync with the session it belongs to.
+func (s *Server) csrfToken(sess *adminauth.Session) string {
+	sum := sha256.Sum256([]byte("csrf|" + sess.Subject + "|" + sess.Expires.String()))
+	return base64.RawURLEncoding.EncodeToString(sum[:16])
+}
+
+func (s *Server) csrfOK(r *http.Request, sess *adminauth.Session) bool {
+	if err := r.ParseForm(); err != nil {
+		return false
+	}
+	return r.PostFormValue(csrfField) == s.csrfToken(sess)
+}
+
+// adminOnly refuses a handler to a signed-in account that is not an administrator.
+//
+// A separate wrapper rather than a check inside each handler: an authorisation rule that has to be
+// remembered in every handler is one that will eventually be forgotten in a new one, and the route
+// table is where someone looks to find out who may do what.
+func (s *Server) adminOnly(
+	h func(http.ResponseWriter, *http.Request, *adminauth.Session),
+) func(http.ResponseWriter, *http.Request, *adminauth.Session) {
+	return func(w http.ResponseWriter, r *http.Request, sess *adminauth.Session) {
+		if !sess.Admin {
+			slog.Info("admin action refused", "account", sess.Subject, "path", r.URL.Path)
+			http.Error(w, "that action needs an administrator account", http.StatusForbidden)
+			return
+		}
+		h(w, r, sess)
+	}
+}
+
+func (s *Server) setSession(w http.ResponseWriter, subject, display string, admin bool) {
+	http.SetCookie(w, &http.Cookie{
+		Name:     sessionCookie,
+		Value:    s.Sessions.Issue(subject, display, admin),
+		Path:     "/",
+		HttpOnly: true, // the cookie is a bearer credential; script has no business reading it
+		Secure:   s.Secure,
+		SameSite: http.SameSiteLaxMode,
+	})
+}
+
+func (s *Server) logout(w http.ResponseWriter, r *http.Request) {
+	http.SetCookie(w, &http.Cookie{
+		Name: sessionCookie, Value: "", Path: "/", MaxAge: -1,
+		HttpOnly: true, Secure: s.Secure, SameSite: http.SameSiteLaxMode,
+	})
+	http.Redirect(w, r, "/login", http.StatusSeeOther)
+}
+
+// ---- local password ---------------------------------------------------------------------
+
+func (s *Server) loginSubmit(w http.ResponseWriter, r *http.Request) {
+	if err := r.ParseForm(); err != nil {
+		http.Error(w, "bad form", http.StatusBadRequest)
+		return
+	}
+	// The delay is applied before the answer, so a wrong guess costs time whether or not the
+	// username exists — the timing carries no information either way.
+	if d := s.Throttle.Delay(); d > 0 {
+		time.Sleep(d)
+	}
+	user := r.PostFormValue("username")
+	pass := r.PostFormValue("password")
+
+	cred := s.Store.LocalAdmin()
+	if cred == nil || !cred.Verify(user, pass) {
+		s.Throttle.Failed()
+		slog.Info("admin login failed", "user", user, "from", clientIP(r))
+		s.render(w, r, "login", map[string]any{
+			"Error": "Incorrect username or password.",
+			"OIDC":  s.oidcAvailable(),
+		})
+		return
+	}
+	s.Throttle.Succeeded()
+	slog.Info("admin login", "user", user, "method", "local", "from", clientIP(r))
+	s.setSession(w, "local:"+cred.Username, cred.Username, true)
+	http.Redirect(w, r, "/", http.StatusSeeOther)
+}
+
+// apiAdmin authenticates a programmatic admin request: the normal session cookie, or HTTP Basic
+// against the break-glass credential for callers without a cookie jar (the README's curl).
+//
+// The cookie path keeps CSRF, exactly like guard: a cookie is an ambient credential and this
+// endpoint changes state. Basic auth is exempt — the password is supplied explicitly per
+// request, so there is nothing for a cross-site form to ride on — and a wrong guess pays the
+// same throttle as the login form, so this is no better a password oracle than that is.
+func (s *Server) apiAdmin(w http.ResponseWriter, r *http.Request) (subject string, ok bool) {
+	if sess := s.session(r); sess != nil {
+		if !sess.Admin {
+			http.Error(w, "that action needs an administrator account", http.StatusForbidden)
+			return "", false
+		}
+		if !s.csrfOK(r, sess) {
+			http.Error(w, "stale form — reload the page and try again", http.StatusForbidden)
+			return "", false
+		}
+		return sess.Subject, true
+	}
+	if user, pass, hasBasic := r.BasicAuth(); hasBasic {
+		if d := s.Throttle.Delay(); d > 0 {
+			time.Sleep(d)
+		}
+		if cred := s.Store.LocalAdmin(); cred != nil && cred.Verify(user, pass) {
+			s.Throttle.Succeeded()
+			return "local:" + user, true
+		}
+		s.Throttle.Failed()
+		slog.Info("admin api auth failed", "user", user, "from", clientIP(r))
+	}
+	w.Header().Set("WWW-Authenticate", `Basic realm="echolot-admin"`)
+	http.Error(w, "authentication required", http.StatusUnauthorized)
+	return "", false
+}
+
+// ---- OIDC -------------------------------------------------------------------------------
+
+func (s *Server) oidcAvailable() bool {
+	return s.OIDC != nil && s.OIDC.Config().Enabled() && s.BaseURL != ""
+}
+
+// oidcStart redirects to the IdP with state and PKCE.
+//
+// PKCE even though this is a confidential client: it costs one hash and closes code interception
+// independently of the secret, which is worth having when the redirect crosses a browser.
+func (s *Server) oidcStart(w http.ResponseWriter, r *http.Request) {
+	if !s.oidcAvailable() {
+		http.Error(w, "no identity provider is configured on this server", http.StatusNotImplemented)
+		return
+	}
+	d, err := s.OIDC.Discover(r.Context())
+	if err != nil {
+		http.Error(w, "identity provider unreachable: "+err.Error(), http.StatusBadGateway)
+		return
+	}
+	state, verifier := randomToken(), randomToken()
+	challenge := sha256.Sum256([]byte(verifier))
+
+	// state and the PKCE verifier ride in one short-lived cookie: the callback must prove it
+	// belongs to the browser that started the flow, or an attacker can feed us their own code.
+	http.SetCookie(w, &http.Cookie{
+		Name: stateCookie, Value: state + "." + verifier, Path: "/",
+		HttpOnly: true, Secure: s.Secure, SameSite: http.SameSiteLaxMode, MaxAge: 600,
+	})
+
+	q := url.Values{
+		"response_type":         {"code"},
+		"client_id":             {s.OIDC.Config().ClientID},
+		"redirect_uri":          {s.redirectURI()},
+		"scope":                 {"openid profile email"},
+		"state":                 {state},
+		"code_challenge":        {base64.RawURLEncoding.EncodeToString(challenge[:])},
+		"code_challenge_method": {"S256"},
+	}
+	http.Redirect(w, r, d.AuthorizationEndpoint+"?"+q.Encode(), http.StatusSeeOther)
+}
+
+func (s *Server) redirectURI() string {
+	return strings.TrimRight(s.BaseURL, "/") + "/admin/callback"
+}
+
+func (s *Server) oidcCallback(w http.ResponseWriter, r *http.Request) {
+	if !s.oidcAvailable() {
+		http.Error(w, "no identity provider configured", http.StatusNotImplemented)
+		return
+	}
+	c, err := r.Cookie(stateCookie)
+	if err != nil {
+		http.Error(w, "sign-in did not start here — try again from the login page", http.StatusBadRequest)
+		return
+	}
+	http.SetCookie(w, &http.Cookie{Name: stateCookie, Value: "", Path: "/", MaxAge: -1})
+
+	state, verifier, ok := strings.Cut(c.Value, ".")
+	if !ok || state == "" || r.URL.Query().Get("state") != state {
+		http.Error(w, "sign-in state did not match — start again", http.StatusBadRequest)
+		return
+	}
+	code := r.URL.Query().Get("code")
+	if code == "" {
+		http.Error(w, "no authorization code returned: "+r.URL.Query().Get("error"), http.StatusBadRequest)
+		return
+	}
+
+	idToken, err := s.exchange(r.Context(), code, verifier)
+	if err != nil {
+		slog.Info("admin oidc exchange failed", "err", err, "from", clientIP(r))
+		http.Error(w, "could not complete sign-in", http.StatusBadGateway)
+		return
+	}
+	claims, err := s.OIDC.Verify(r.Context(), idToken)
+	if err != nil {
+		slog.Info("admin oidc token rejected", "err", err, "from", clientIP(r))
+		http.Error(w, "the identity token was not accepted", http.StatusForbidden)
+		return
+	}
+	// Authentication and authorisation are answered separately here. Someone who is not in the
+	// admin group has still proved who they are, and their own uploads are their business to
+	// manage — refusing them a session outright, as this used to, left a legitimate account with
+	// no way to see or delete the data it had sent.
+	admin := s.OIDC.IsAdmin(claims)
+	slog.Info("login", "account", claims.AccountID(), "method", "oidc", "admin", admin,
+		"from", clientIP(r))
+	s.setSession(w, claims.AccountID(), claims.Display(), admin)
+	http.Redirect(w, r, "/", http.StatusSeeOther)
+}
+
+// exchange trades the authorization code for tokens at the IdP.
+func (s *Server) exchange(ctx context.Context, code, verifier string) (string, error) {
+	d, err := s.OIDC.Discover(ctx)
+	if err != nil {
+		return "", err
+	}
+	form := url.Values{
+		"grant_type":    {"authorization_code"},
+		"code":          {code},
+		"redirect_uri":  {s.redirectURI()},
+		"client_id":     {s.OIDC.Config().ClientID},
+		"code_verifier": {verifier},
+	}
+	if s.ClientSecret != "" {
+		form.Set("client_secret", s.ClientSecret)
+	}
+	req, err := http.NewRequestWithContext(ctx, http.MethodPost, d.TokenEndpoint,
+		strings.NewReader(form.Encode()))
+	if err != nil {
+		return "", err
+	}
+	req.Header.Set("Content-Type", "application/x-www-form-urlencoded")
+
+	resp, err := (&http.Client{Timeout: 15 * time.Second}).Do(req)
+	if err != nil {
+		return "", err
+	}
+	defer resp.Body.Close()
+	body, _ := io.ReadAll(io.LimitReader(resp.Body, 1<<20))
+	if resp.StatusCode != http.StatusOK {
+		return "", fmt.Errorf("token endpoint: %s: %s", resp.Status, strings.TrimSpace(string(body)))
+	}
+	var tok struct {
+		IDToken string `json:"id_token"`
+	}
+	if err := json.Unmarshal(body, &tok); err != nil {
+		return "", err
+	}
+	if tok.IDToken == "" {
+		return "", fmt.Errorf("token endpoint returned no id_token")
+	}
+	return tok.IDToken, nil
+}
+
+func randomToken() string {
+	b := make([]byte, 32)
+	_, _ = rand.Read(b)
+	return base64.RawURLEncoding.EncodeToString(b)
+}
+
+// clientIP is for logs only. X-Forwarded-For is deliberately ignored: nothing is meant to sit in
+// front of this listener, so a header claiming otherwise is a caller's assertion about itself.
+func clientIP(r *http.Request) string {
+	if i := strings.LastIndex(r.RemoteAddr, ":"); i > 0 {
+		return r.RemoteAddr[:i]
+	}
+	return r.RemoteAddr
+}

server/internal/adminui/enrolltokens_test.go

@@ -0,0 +1,88 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package adminui
+
+import (
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+	"time"
+
+	"echo-lot.app/server/internal/adminauth"
+)
+
+// tokenFixture wires just enough of the Server for the mint endpoint: a break-glass admin and
+// a stand-in EnrollLink (the real one belongs to the control server, injected the same way).
+func tokenFixture(t *testing.T) *Server {
+	t.Helper()
+	s, _, _, _ := fixture(t)
+	secret, err := s.Store.SessionSecret()
+	if err != nil {
+		t.Fatal(err)
+	}
+	s.Sessions = adminauth.NewSessions(secret, time.Hour)
+	s.Throttle = adminauth.NewThrottle()
+	cred, err := adminauth.NewCredential("admin", "a-long-test-password")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if err := s.Store.SetLocalAdmin(cred); err != nil {
+		t.Fatal(err)
+	}
+	s.EnrollLink = func(tok string) string { return "echolot://enroll?v=1&t=" + tok }
+	return s
+}
+
+func TestEnrollTokensAPISpecShape(t *testing.T) {
+	h := tokenFixture(t).Handler()
+
+	// No credentials → 401 with a challenge, never a token.
+	req := httptest.NewRequest("POST", "/admin/enroll-tokens?note=phone", nil)
+	rec := httptest.NewRecorder()
+	h.ServeHTTP(rec, req)
+	if rec.Code != http.StatusUnauthorized || rec.Header().Get("WWW-Authenticate") == "" {
+		t.Fatalf("unauthenticated: code=%d", rec.Code)
+	}
+
+	// Wrong password → still 401.
+	req = httptest.NewRequest("POST", "/admin/enroll-tokens", nil)
+	req.SetBasicAuth("admin", "wrong")
+	rec = httptest.NewRecorder()
+	h.ServeHTTP(rec, req)
+	if rec.Code != http.StatusUnauthorized {
+		t.Fatalf("bad password: code=%d, want 401", rec.Code)
+	}
+
+	// Basic + Accept: application/json → the §2.1 shape.
+	req = httptest.NewRequest("POST", "/admin/enroll-tokens?note=phone", nil)
+	req.SetBasicAuth("admin", "a-long-test-password")
+	req.Header.Set("Accept", "application/json")
+	rec = httptest.NewRecorder()
+	h.ServeHTTP(rec, req)
+	if rec.Code != http.StatusOK {
+		t.Fatalf("mint: code=%d body=%s", rec.Code, rec.Body.String())
+	}
+	var body struct {
+		Token     string `json:"token"`
+		ExpiresS  int    `json:"expires_in_s"`
+		EnrollURI string `json:"enroll_uri"`
+	}
+	if err := json.Unmarshal(rec.Body.Bytes(), &body); err != nil {
+		t.Fatal(err)
+	}
+	if body.Token == "" || body.ExpiresS != 86400 || !strings.HasPrefix(body.EnrollURI, "echolot://enroll?") {
+		t.Fatalf("spec shape violated: %+v", body)
+	}
+
+	// Without Accept: the browser flow — redirect to the QR page, link in the query.
+	req = httptest.NewRequest("POST", "/admin/enroll-tokens", nil)
+	req.SetBasicAuth("admin", "a-long-test-password")
+	rec = httptest.NewRecorder()
+	h.ServeHTTP(rec, req)
+	if rec.Code != http.StatusSeeOther || !strings.HasPrefix(rec.Header().Get("Location"), "/devices?link=") {
+		t.Fatalf("html flow: code=%d location=%q", rec.Code, rec.Header().Get("Location"))
+	}
+}

server/internal/adminui/pages.go

@@ -0,0 +1,273 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package adminui
+
+import (
+	"encoding/json"
+	"html/template"
+	"log/slog"
+	"net/http"
+	"net/url"
+	"sort"
+	"strings"
+	"time"
+
+	"echo-lot.app/server/internal/adminauth"
+	"echo-lot.app/server/internal/runs"
+	"echo-lot.app/server/internal/store"
+)
+
+// visibleDevices returns the devices a session may see: everything for an administrator, and for
+// anyone else the devices linked to their own account.
+//
+// Every page goes through this rather than filtering for itself. Scoping applied per-page is
+// scoping that will be missing from the next page someone adds, and the failure is silent — a
+// listing that quietly shows other people's uploads looks exactly like one that does not.
+func (s *Server) visibleDevices(sess *adminauth.Session) []store.Device {
+	all := s.Store.Devices()
+	if sess.Admin {
+		return all
+	}
+	owned := make(map[string]bool)
+	for _, id := range s.Store.DeviceIDsForAccount(sess.Subject) {
+		owned[id] = true
+	}
+	out := make([]store.Device, 0, len(owned))
+	for _, d := range all {
+		if owned[d.ID] {
+			out = append(out, d)
+		}
+	}
+	return out
+}
+
+// mayTouchRun reports whether this session may read or delete a given run.
+//
+// Checked against the device list rather than against the run's own metadata, so an unlinked or
+// revoked device stops granting access the moment the link is gone.
+func (s *Server) mayTouchRun(sess *adminauth.Session, device string) bool {
+	if sess.Admin {
+		return true
+	}
+	for _, d := range s.visibleDevices(sess) {
+		if d.ID == device {
+			return true
+		}
+	}
+	return false
+}
+
+func (s *Server) loginForm(w http.ResponseWriter, r *http.Request) {
+	if s.session(r) != nil {
+		http.Redirect(w, r, "/", http.StatusSeeOther)
+		return
+	}
+	s.render(w, r, "login", map[string]any{
+		"OIDC":      s.oidcAvailable(),
+		"LocalSet":  s.Store.LocalAdmin() != nil,
+		"AdminUser": s.AdminUser,
+	})
+}
+
+func (s *Server) dashboard(w http.ResponseWriter, r *http.Request, sess *adminauth.Session) {
+	devices := s.visibleDevices(sess)
+	linked := 0
+	for _, d := range devices {
+		if d.LinkedToAccount() {
+			linked++
+		}
+	}
+	var selftest any
+	// The self-test describes the server's own health, which is an operator's concern; a user
+	// looking at their uploads has no use for it and no ability to act on it.
+	if s.SelfTest != nil && sess.Admin {
+		selftest = s.SelfTest()
+	}
+	s.render(w, r, "dashboard", map[string]any{
+		"Session":  sess,
+		"CSRF":     s.csrfToken(sess),
+		"Devices":  len(devices),
+		"Linked":   linked,
+		"Runs":     s.totalRuns(devices),
+		"SelfTest": selftest,
+		"Version":  s.Version,
+		"Admin":    sess.Admin,
+	})
+}
+
+func (s *Server) totalRuns(devices []store.Device) int {
+	if s.Runs == nil {
+		return 0
+	}
+	n := 0
+	for _, d := range devices {
+		n += len(s.Runs.List(d.ID))
+	}
+	return n
+}
+
+func (s *Server) devices(w http.ResponseWriter, r *http.Request, sess *adminauth.Session) {
+	devices := s.visibleDevices(sess)
+	// Newest first: the device someone is looking for is almost always the one just enrolled.
+	sort.Slice(devices, func(i, j int) bool { return devices[i].Enrolled.After(devices[j].Enrolled) })
+
+	type row struct {
+		store.Device
+		Runs int
+	}
+	rows := make([]row, 0, len(devices))
+	for _, d := range devices {
+		n := 0
+		if s.Runs != nil {
+			n = len(s.Runs.List(d.ID))
+		}
+		rows = append(rows, row{Device: d, Runs: n})
+	}
+	// html/template rewrites an href whose scheme it does not recognise to "#ZgotmplZ", so the
+	// enrollment link rendered as a dead anchor that did nothing when tapped — silently, since the
+	// markup looks fine and only the sanitised attribute gives it away.
+	//
+	// Marking it template.URL opts out of that sanitising, which is only safe because the shape is
+	// checked first: this value arrives in a query parameter, so without the check a crafted
+	// /devices?link=javascript:… would put a script URL straight into the page.
+	link := r.URL.Query().Get("link")
+	var href template.URL
+	if strings.HasPrefix(link, "echolot://enroll?") {
+		href = template.URL(link)
+	}
+	s.render(w, r, "devices", map[string]any{
+		"Session": sess, "CSRF": s.csrfToken(sess), "Rows": rows,
+		// Rendered from the same validated value as the href, so a rejected link produces neither.
+		"Link": link, "LinkHref": href, "LinkQR": qrSVG(string(href)), "Admin": sess.Admin,
+	})
+}
+
+func (s *Server) revokeDevice(w http.ResponseWriter, r *http.Request, sess *adminauth.Session) {
+	id := r.PathValue("id")
+	if err := s.Store.DeleteDevice(id); err != nil {
+		http.Error(w, err.Error(), http.StatusBadRequest)
+		return
+	}
+	// Worth a log line: revoking a device is destructive, immediate, and someone will eventually
+	// want to know who did it and when.
+	slog.Info("device revoked", "device", id, "by", sess.Subject)
+	http.Redirect(w, r, "/devices", http.StatusSeeOther)
+}
+
+func (s *Server) mintToken(w http.ResponseWriter, r *http.Request, sess *adminauth.Session) {
+	tok, err := s.Store.NewEnrollToken(24*time.Hour, "admin-ui")
+	if err != nil {
+		http.Error(w, err.Error(), http.StatusInternalServerError)
+		return
+	}
+	slog.Info("enrolment token minted", "by", sess.Subject)
+	// The whole link, not the bare token: it carries the URL and the pin as well, and assembling
+	// those by hand is where an operator gets a pin wrong by one character.
+	http.Redirect(w, r, "/devices?link="+url.QueryEscape(s.EnrollLink(tok)), http.StatusSeeOther)
+}
+
+// enrollTokensAPI is POST /admin/enroll-tokens, the endpoint the spec's §2.1 example names.
+// Content-negotiated: Accept: application/json gets the spec shape {token, expires_in_s,
+// enroll_uri}; anything else (a browser) gets the same redirect-to-QR flow as the form above,
+// so the one path serves both audiences.
+func (s *Server) enrollTokensAPI(w http.ResponseWriter, r *http.Request) {
+	subject, ok := s.apiAdmin(w, r)
+	if !ok {
+		return
+	}
+	note := r.URL.Query().Get("note")
+	if note == "" {
+		note = "admin-api"
+	}
+	const ttl = 24 * time.Hour
+	tok, err := s.Store.NewEnrollToken(ttl, note)
+	if err != nil {
+		http.Error(w, err.Error(), http.StatusInternalServerError)
+		return
+	}
+	slog.Info("enrolment token minted", "by", subject, "note", note)
+	if !strings.Contains(r.Header.Get("Accept"), "application/json") {
+		http.Redirect(w, r, "/devices?link="+url.QueryEscape(s.EnrollLink(tok)), http.StatusSeeOther)
+		return
+	}
+	w.Header().Set("Content-Type", "application/json")
+	// The whole link, not the bare token (§2.1): the server is the only party holding URL, pin
+	// and token at once, and a hand-assembled pin wrong by one character fails as an inscrutable
+	// TLS error later rather than loudly here.
+	_ = json.NewEncoder(w).Encode(map[string]any{
+		"token":        tok,
+		"expires_in_s": int(ttl.Seconds()),
+		"enroll_uri":   s.EnrollLink(tok),
+	})
+}
+
+// EnrollLink is supplied by the caller so this package does not need the control server's pin.
+var _ = 0
+
+func (s *Server) runsList(w http.ResponseWriter, r *http.Request, sess *adminauth.Session) {
+	type row struct {
+		runs.Meta
+		DeviceName string
+	}
+	var rows []row
+	for _, d := range s.visibleDevices(sess) {
+		if s.Runs == nil {
+			break
+		}
+		name := d.Name
+		if name == "" {
+			name = d.ID
+		}
+		for _, m := range s.Runs.List(d.ID) {
+			rows = append(rows, row{Meta: m, DeviceName: name})
+		}
+	}
+	sort.Slice(rows, func(i, j int) bool { return rows[i].UploadedAt.After(rows[j].UploadedAt) })
+	if len(rows) > 200 {
+		rows = rows[:200] // a page, not the archive; the count is on the dashboard
+	}
+	s.render(w, r, "runs", map[string]any{
+		"Session": sess, "CSRF": s.csrfToken(sess), "Rows": rows, "Admin": sess.Admin,
+	})
+}
+
+func (s *Server) runView(w http.ResponseWriter, r *http.Request, sess *adminauth.Session) {
+	// 404 rather than 403 for someone else's run: a distinguishable "you may not see this" tells
+	// an unauthorised caller that the run exists, which is itself something they should not learn.
+	if !s.mayTouchRun(sess, r.PathValue("device")) {
+		http.NotFound(w, r)
+		return
+	}
+	body, err := s.Runs.Get(r.PathValue("device"), r.PathValue("id"))
+	if err != nil {
+		http.NotFound(w, r)
+		return
+	}
+	// Re-indented for reading, but otherwise exactly what was stored. An admin sees the document
+	// at the privacy level its uploader chose — there is nothing here that can un-redact it.
+	var pretty json.RawMessage = body
+	out, err := json.MarshalIndent(json.RawMessage(pretty), "", "  ")
+	if err != nil {
+		out = body
+	}
+	s.render(w, r, "run", map[string]any{
+		"Session": sess, "CSRF": s.csrfToken(sess), "Admin": sess.Admin,
+		"Device": r.PathValue("device"), "ID": r.PathValue("id"),
+		"JSON": string(out),
+	})
+}
+
+func (s *Server) runDelete(w http.ResponseWriter, r *http.Request, sess *adminauth.Session) {
+	device, id := r.PathValue("device"), r.PathValue("id")
+	if !s.mayTouchRun(sess, device) {
+		http.NotFound(w, r)
+		return
+	}
+	if err := s.Runs.Delete(device, id); err != nil {
+		http.Error(w, err.Error(), http.StatusInternalServerError)
+		return
+	}
+	slog.Info("run deleted", "device", device, "run", id, "by", sess.Subject)
+	http.Redirect(w, r, "/runs", http.StatusSeeOther)
+}

server/internal/adminui/qr.go

@@ -0,0 +1,58 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package adminui
+
+import (
+	"fmt"
+	"html/template"
+	"strings"
+
+	qrcode "github.com/skip2/go-qrcode"
+)
+
+// qrSVG renders text as an inline SVG QR code, or empty if it will not encode.
+//
+// Inline SVG rather than a PNG data: URI because the page's CSP is `default-src 'none'` and means
+// it. A data: image would need img-src opened up; markup needs nothing, and the QR is generated
+// here from a boolean matrix, so nothing a user supplied reaches the output.
+//
+// Drawn as one path rather than a rect per module: a link of this length encodes to roughly 60x60
+// modules, and two thousand elements is a lot of DOM for a picture of a square.
+func qrSVG(text string) template.HTML {
+	if text == "" {
+		return ""
+	}
+	// Medium recovery: a phone camera reading a screen has no dirt or creases to survive, and
+	// lower recovery keeps the module count down, which keeps it scannable on a small display.
+	q, err := qrcode.New(text, qrcode.Medium)
+	if err != nil {
+		return "" // too long to encode; the link text below it still works
+	}
+	bitmap := q.Bitmap()
+	n := len(bitmap)
+	if n == 0 {
+		return ""
+	}
+
+	var path strings.Builder
+	for y, row := range bitmap {
+		for x, dark := range row {
+			if dark {
+				fmt.Fprintf(&path, "M%d %dh1v1h-1z", x, y)
+			}
+		}
+	}
+
+	// A quiet zone is part of the spec, not decoration: without it a scanner cannot find the
+	// symbol's edges against whatever is next to it on the page.
+	var out strings.Builder
+	fmt.Fprintf(&out,
+		`<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 %d %d" `+
+			`width="240" height="240" shape-rendering="crispEdges" role="img" `+
+			`aria-label="Enrolment link as a QR code">`+
+			`<rect width="%d" height="%d" fill="#fff"/>`+
+			`<path d="%s" fill="#000"/></svg>`,
+		n, n, n, n, path.String())
+	return template.HTML(out.String())
+}

server/internal/adminui/qr_test.go

@@ -0,0 +1,24 @@
+package adminui
+
+import (
+	"strings"
+	"testing"
+)
+
+func TestQrSVGEncodesAnEnrolmentLink(t *testing.T) {
+	link := "echolot://enroll?v=1&u=https%3A%2F%2Ffmr.echo-lot.app&p=pin-sha256%3AzRV9qkiLnRexAeh4RrSfJzbPWO%2BU%2F2Oj2%2FNVM%2FKfXlg%3D&t=20e6ccaa2a028dc0aab16442c258d1b8eadb5794682905fe"
+	out := string(qrSVG(link))
+	if !strings.HasPrefix(out, "<svg") || !strings.Contains(out, "<path d=\"M") {
+		t.Fatalf("expected an svg with a path, got %.80q", out)
+	}
+	// A quiet zone is part of the symbol; without it scanners cannot find its edges.
+	if !strings.Contains(out, `fill="#fff"`) {
+		t.Error("no light background rendered")
+	}
+}
+
+func TestQrSVGEmptyForNoLink(t *testing.T) {
+	if qrSVG("") != "" {
+		t.Error("no link should render no code")
+	}
+}

server/internal/adminui/render.go

@@ -0,0 +1,420 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package adminui
+
+import (
+	"bytes"
+	"html/template"
+	"log/slog"
+	"net/http"
+	"strings"
+)
+
+// Templates are parsed once at start. html/template escapes by context, which is what makes it
+// safe to render device names and finding text that ultimately arrived over a network.
+var tpl = template.Must(template.New("base").Funcs(template.FuncMap{
+	"kb": func(n int64) int64 { return n / 1024 },
+	// verdictClass keeps an uploaded string out of the class attribute. The verdict arrives inside
+	// a document a device sent us, so interpolating it into markup would be trusting a stranger's
+	// text with a place in the stylesheet; mapping through a fixed set costs nothing and closes it.
+	"verdictClass": func(v string) string {
+		switch strings.ToLower(v) {
+		case "green", "yellow", "red", "inconclusive":
+			return "v-" + strings.ToLower(v)
+		default:
+			return "v-unknown"
+		}
+	},
+	// verdictLabel says what the light means rather than what it is called. "yellow" is a colour;
+	// "worth a look" is a finding, and the reader is here to act on it.
+	"verdictLabel": func(v string) string {
+		switch strings.ToLower(v) {
+		case "green":
+			return "clean"
+		case "yellow":
+			return "worth a look"
+		case "red":
+			return "faults found"
+		case "inconclusive":
+			return "inconclusive"
+		default:
+			return "not recorded"
+		}
+	},
+}).Parse(baseHTML))
+
+func (s *Server) render(w http.ResponseWriter, r *http.Request, page string, data map[string]any) {
+	data["Page"] = page
+	var buf bytes.Buffer
+	if err := tpl.Execute(&buf, data); err != nil {
+		slog.Error("admin template", "page", page, "err", err)
+		http.Error(w, "template error", http.StatusInternalServerError)
+		return
+	}
+	w.Header().Set("Content-Type", "text/html; charset=utf-8")
+	// There is no script here and nothing loaded from anywhere else, so a strict policy costs
+	// nothing and closes injected-script attacks even if an escaping bug ever slips through.
+	w.Header().Set("Content-Security-Policy", "default-src 'none'; style-src 'unsafe-inline'; form-action 'self'")
+	w.Header().Set("Referrer-Policy", "no-referrer")
+	w.Header().Set("X-Content-Type-Options", "nosniff")
+	_, _ = buf.WriteTo(w)
+}
+
+// The visual language is an echo sounder's, which is what the name means: an instrument that emits
+// a ping and reads what comes back. That gives the palette (the colours of a water column rather
+// than a neutral near-black), the type (machine-set, because an instrument's readings are), and
+// the one piece of real ornament — a trace of returns across time on the runs page.
+//
+// No web fonts: the CSP forbids loading anything, and shipping font files with a single Go binary
+// would trade the property that makes this server pleasant to run for a typeface. So the character
+// has to come from treatment — tracking, case, scale, rules — rather than from novel letterforms.
+//
+// Tables become stacked records below 46rem rather than scrolling sideways. That is not a fallback:
+// a sounding log prints as label-and-value pairs, and on a phone that form is easier to read than
+// any table, so the mobile layout is the more faithful one of the two.
+const baseHTML = `<!doctype html>
+<html lang="en"><head><meta charset="utf-8">
+<meta name="viewport" content="width=device-width,initial-scale=1">
+<title>Echolot &mdash; {{.Page}}</title>
+<style>
+ :root{
+   --abyss:#071419; --hull:#0d2028; --raise:#122a34; --rule:#17323d;
+   --ink:#dce8ea; --dim:#7d97a1; --trace:#6fc9b4;
+   --green:#57ad82; --amber:#cf9b3c; --red:#c25757; --slate:#62767f;
+   --mono:ui-monospace,"SF Mono","IBM Plex Mono","JetBrains Mono",Menlo,Consolas,monospace;
+   --prose:system-ui,-apple-system,"Segoe UI",sans-serif;
+   color-scheme:dark;
+ }
+ *{box-sizing:border-box}
+ body{margin:0;background:var(--abyss);color:var(--ink);
+      font:400 15px/1.55 var(--prose);-webkit-text-size-adjust:100%}
+
+ /* ---- masthead ------------------------------------------------------------------------ */
+ /* Wraps rather than overflows: a rigid row pushes the account and its sign-out button past
+    the edge of a phone screen, where they cannot be reached at all. */
+ .top{display:flex;flex-wrap:wrap;align-items:center;gap:.5rem 1.4rem;
+      padding:.85rem 1.1rem;background:var(--hull);border-bottom:1px solid var(--rule)}
+ .mark{font:600 .95rem/1 var(--mono);letter-spacing:.02em;margin:0;color:var(--ink)}
+ .mark span{color:var(--trace)}
+ .top nav{display:flex;flex-wrap:wrap;gap:.15rem .9rem}
+ .top nav a{font:500 .82rem/1 var(--mono);letter-spacing:.06em;color:var(--dim);
+            text-decoration:none;padding:.35rem 0;border-bottom:1px solid transparent}
+ .top nav a:hover{color:var(--ink)}
+ .top nav a[aria-current]{color:var(--trace);border-bottom-color:var(--trace)}
+ .who{margin-left:auto;display:flex;align-items:center;gap:.7rem;flex-wrap:wrap;
+      font:.78rem/1.3 var(--mono);color:var(--dim)}
+
+ main{padding:1.1rem;max-width:64rem}
+
+ /* ---- headings: a graduation mark, like a depth scale ------------------------------- */
+ h2{font:500 1.05rem/1.2 var(--mono);letter-spacing:-.01em;margin:1.4rem 0 .2rem;
+    padding-left:.7rem;border-left:2px solid var(--trace)}
+ h2:first-child{margin-top:0}
+ h3{font:500 .8rem/1 var(--mono);letter-spacing:.14em;text-transform:uppercase;
+    color:var(--dim);margin:0 0 .7rem}
+ .lede{color:var(--dim);font-size:.9rem;margin:.5rem 0 1rem;max-width:46rem}
+
+ /* ---- readout: how an instrument prints a value ------------------------------------- */
+ .readout{list-style:none;margin:0;padding:0}
+ .readout li{display:flex;align-items:baseline;gap:.5rem;padding:.3rem 0;
+             font:.85rem/1.4 var(--mono)}
+ .readout .k{color:var(--dim);white-space:nowrap}
+ /* The dotted leader is how a sounding log runs a label out to its value. It is also the thing
+    that lets a label and a number sit on one line at any width without a table. */
+ .readout .lead{flex:1 1 auto;min-width:1.5rem;align-self:center;height:1px;
+                background:repeating-linear-gradient(90deg,var(--rule) 0 2px,transparent 2px 5px)}
+ .readout .v{color:var(--ink);text-align:right;overflow-wrap:anywhere}
+
+ /* ---- the trace: one bar per run, oldest to newest ---------------------------------- */
+ /* The signature, and the only ornament here: an echo sounder draws returns against time, and
+    so does this. Rows arrive newest-first, so the strip is reversed in CSS rather than in Go. */
+ .trace{display:flex;flex-direction:row-reverse;justify-content:flex-end;align-items:flex-end;gap:2px;
+        height:3rem;padding:.7rem .8rem;background:var(--hull);
+        border:1px solid var(--rule);border-radius:3px;overflow:hidden}
+ .trace i{flex:1 1 3px;min-width:2px;max-width:9px;border-radius:1px;opacity:.9}
+ .trace .v-green{height:35%;background:var(--green)}
+ .trace .v-yellow{height:65%;background:var(--amber)}
+ .trace .v-red{height:100%;background:var(--red)}
+ .trace .v-inconclusive{height:22%;background:var(--slate)}
+ .trace .v-unknown{height:12%;background:var(--rule)}
+ .trace-key{display:flex;flex-wrap:wrap;gap:.3rem .9rem;margin:.45rem 0 0;
+            font:.72rem/1 var(--mono);letter-spacing:.05em;color:var(--dim)}
+ .trace-key b{font-weight:400;color:var(--dim)}
+ .trace-key em{font-style:normal;display:inline-block;width:.5rem;height:.5rem;
+               border-radius:1px;margin-right:.35rem;vertical-align:baseline;background:var(--rule)}
+ .trace-key em.v-green{background:var(--green)}
+ .trace-key em.v-yellow{background:var(--amber)}
+ .trace-key em.v-red{background:var(--red)}
+ .trace-key em.v-inconclusive{background:var(--slate)}
+
+ /* ---- records: tables that stack on a phone ----------------------------------------- */
+ .rec{border:1px solid var(--rule);border-radius:3px;background:var(--hull);
+      padding:.75rem .85rem;margin:.5rem 0}
+ .rec-head{display:flex;flex-wrap:wrap;align-items:baseline;gap:.5rem;
+           font:.85rem/1.3 var(--mono);margin-bottom:.35rem}
+ .rec-head .id{overflow-wrap:anywhere;color:var(--ink)}
+ .rec form{margin-top:.6rem}
+ /* Why a check matters is a sentence, so it is set as one — full width under the row rather
+    than squeezed into a column, where it would wrap to a ribbon two words wide. */
+ .why{font:.85rem/1.5 var(--prose);color:var(--dim);margin-top:.45rem;max-width:52rem}
+ .tag{font:.68rem/1 var(--mono);letter-spacing:.1em;text-transform:uppercase;
+      padding:.24rem .45rem;border-radius:2px;border:1px solid currentColor;white-space:nowrap}
+ .v-green{color:var(--green)} .v-yellow{color:var(--amber)}
+ .v-red{color:var(--red)} .v-inconclusive{color:var(--slate)} .v-unknown{color:var(--dim)}
+
+ /* ---- panels, controls, states ------------------------------------------------------ */
+ .narrow{max-width:27rem}
+ .panel{background:var(--hull);border:1px solid var(--rule);border-radius:3px;
+        padding:.95rem 1rem;margin:.9rem 0;min-width:0}
+ /* White plate behind the code: a QR needs the light modules to actually be light, and this
+    page is dark. */
+ .qr{display:inline-block;background:#fff;padding:8px;border-radius:4px;margin:.2rem 0;line-height:0}
+ .qr svg{display:block;width:min(240px,60vw);height:auto}
+ .empty{border:1px dashed var(--rule);border-radius:3px;padding:1.4rem 1rem;
+        color:var(--dim);font-size:.9rem}
+ code,pre,.mono{font-family:var(--mono);font-size:.82rem}
+ code{overflow-wrap:anywhere;color:var(--trace)}
+ pre{background:#040d11;border:1px solid var(--rule);border-radius:3px;padding:.8rem;
+     overflow:auto;max-height:32rem;max-width:100%;color:var(--ink)}
+ a{color:var(--trace)}
+ button{font:500 .82rem/1 var(--mono);letter-spacing:.05em;background:var(--trace);
+        color:#04181a;border:0;border-radius:3px;padding:.55rem .9rem;cursor:pointer}
+ .btn{display:inline-block;font:500 .82rem/1 var(--mono);letter-spacing:.05em;
+      background:var(--trace);color:#04181a;border-radius:3px;padding:.55rem .9rem;
+      text-decoration:none}
+ button.plain{background:transparent;color:var(--dim);border:1px solid var(--rule)}
+ button.danger{background:transparent;color:var(--red);border:1px solid var(--red)}
+ button:hover{filter:brightness(1.08)}
+ input{font:.9rem var(--mono);background:#040d11;color:var(--ink);border:1px solid var(--rule);
+       border-radius:3px;padding:.55rem .6rem;max-width:100%;width:100%}
+ label{display:block;font:.72rem/1 var(--mono);letter-spacing:.12em;text-transform:uppercase;
+       color:var(--dim);margin:.9rem 0 .3rem}
+ .err{border:1px solid var(--red);color:var(--ink);background:rgba(194,87,87,.09);
+      padding:.6rem .75rem;border-radius:3px;font-size:.9rem}
+ .muted{color:var(--dim)}
+ form.inline{display:inline}
+ :focus-visible{outline:2px solid var(--trace);outline-offset:2px}
+ @media (prefers-reduced-motion:reduce){*{transition:none!important;animation:none!important}}
+
+ /* ---- above 46rem the records line up in columns ------------------------------------ */
+ @media (min-width:46rem){
+   .top{padding:.85rem 1.6rem}
+   main{padding:1.6rem}
+   .recs{margin:.8rem 0}
+   /* Every row shares one grid, so the columns agree across rows without a header or a table. */
+   .rec{display:grid;grid-template-columns:minmax(12.5rem,18rem) minmax(0,1fr) auto;gap:.35rem 1.4rem;
+        align-items:baseline;background:none;border:0;border-bottom:1px solid var(--rule);
+        border-radius:0;padding:.6rem 0;margin:0}
+   .rec-head{margin:0;flex-direction:column;align-items:flex-start;gap:.3rem}
+   .rec form{margin:0}
+   /* Widths follow the content: a device name needs room, a finding count does not. */
+   .rec .readout{display:grid;grid-template-columns:1.7fr .9fr .9fr 1.1fr;gap:.15rem 1.2rem}
+   .rec .readout li{padding:0}
+   .rec .readout .lead{display:none}
+   .rec .readout .v{text-align:left}
+   .open{white-space:nowrap}
+   /* Spans the full row: the sentence is the useful part, not a fourth column. */
+   .why{grid-column:1/-1;margin-top:.1rem}
+ }
+</style></head><body>
+<header class="top">
+  <h1 class="mark">echo<span>lot</span></h1>
+  {{if ne .Page "login"}}
+  <nav>
+    <a href="/"{{if eq .Page "dashboard"}} aria-current="page"{{end}}>overview</a>
+    <a href="/devices"{{if eq .Page "devices"}} aria-current="page"{{end}}>devices</a>
+    <a href="/runs"{{if eq .Page "runs"}} aria-current="page"{{end}}>runs</a>
+  </nav>
+  <span class="who">{{.Session.Display}}{{if not .Session.Admin}} &middot; your account{{end}}
+    <form method="post" action="/logout" class="inline"><button class="plain">Sign out</button></form>
+  </span>
+  {{end}}
+</header>
+<main>
+
+{{if eq .Page "login"}}
+  <h2>Sign in</h2>
+  <p class="lede">This server keeps the measurements your devices have uploaded.</p>
+  {{if .OIDC}}
+    <p><a class="btn" href="/auth/start">Sign in with your identity provider</a></p>
+  {{end}}
+  {{if .LocalSet}}
+    <form method="post" action="/login" class="panel narrow">
+      <h3>Break-glass account</h3>
+      <label for="u">Username</label>
+      <input id="u" name="username" autocomplete="username" value="{{.AdminUser}}">
+      <label for="p">Password</label>
+      <input id="p" name="password" type="password" autocomplete="current-password">
+      <p><button>Sign in</button></p>
+    </form>
+  {{else}}
+    <p class="err">No break-glass account is set. Run
+      <code>echolot-server --set-admin-password</code> on the host to create one.</p>
+  {{end}}
+
+{{else if eq .Page "dashboard"}}
+  <h2>{{if .Admin}}This server{{else}}Your account{{end}}</h2>
+  <ul class="readout panel">
+    <li><span class="k">{{if .Admin}}devices enrolled{{else}}your devices{{end}}</span>
+        <span class="lead"></span><span class="v">{{.Devices}}</span></li>
+    {{if .Admin}}
+    <li><span class="k">linked to an account</span>
+        <span class="lead"></span><span class="v">{{.Linked}}</span></li>
+    {{end}}
+    <li><span class="k">{{if .Admin}}runs stored{{else}}your runs{{end}}</span>
+        <span class="lead"></span><span class="v">{{.Runs}}</span></li>
+    {{if .Admin}}
+    <li><span class="k">server version</span>
+        <span class="lead"></span><span class="v">{{.Version}}</span></li>
+    {{end}}
+  </ul>
+  {{if not .Admin}}
+  <div class="panel">
+    <p>You can see every device you have signed in on, read everything they have uploaded, and
+      delete any of it.</p>
+    <p class="muted">Enrolling devices, revoking them, and reading other people's uploads need an
+      administrator account.</p>
+  </div>
+  {{end}}
+  {{with .SelfTest}}
+  <h2>Self-test</h2>
+  <p class="lede">What this server can measure from where it stands, checked at startup. A
+    capability missing here is missing from every run this server takes part in &mdash; so a
+    client asking for that measurement gets nothing, rather than a wrong answer.</p>
+  <ul class="readout panel">
+    <li><span class="k">kernel settings</span><span class="lead"></span>
+      <span class="v {{if .SysctlOK}}v-green{{else}}v-yellow{{end}}">{{if .SysctlOK}}as needed{{else}}need attention{{end}}</span></li>
+    <li><span class="k">egress path MTU</span><span class="lead"></span>
+      <span class="v {{if .MTUOK}}v-green{{else}}v-yellow{{end}}">{{if .MTUOK}}full 1500{{else}}reduced{{end}}</span></li>
+  </ul>
+  {{if .Sysctls}}
+  <h3>Kernel settings</h3>
+  <div class="recs">
+  {{range .Sysctls}}
+    <div class="rec">
+      <div class="rec-head"><span class="id">{{.Name}}</span>
+        <span class="tag {{if eq .Severity "ok"}}v-green{{else}}v-yellow{{end}}">{{.Severity}}</span></div>
+      <ul class="readout">
+        <li><span class="k">found</span><span class="lead"></span><span class="v">{{.Got}}</span></li>
+        <li><span class="k">wanted</span><span class="lead"></span><span class="v">{{.Want}}</span></li>
+      </ul>
+      <div class="why">{{.Why}}</div>
+    </div>
+  {{end}}
+  </div>
+  {{end}}
+  {{if .EgressMTU}}
+  <h3>Egress path MTU</h3>
+  <div class="recs">
+  {{range .EgressMTU}}
+    <div class="rec">
+      <div class="rec-head"><span class="id">{{.Target}}</span>
+        <span class="tag {{if .FullMTU}}v-green{{else}}v-yellow{{end}}">{{if .FullMTU}}full{{else}}reduced{{end}}</span></div>
+      <ul class="readout">
+        <li><span class="k">discovered</span><span class="lead"></span>
+          <span class="v">{{if .DiscoveredMTU}}{{.DiscoveredMTU}} bytes{{else}}not measured{{end}}</span></li>
+      </ul>
+      {{with .Err}}<div class="why">{{.}}</div>{{end}}
+    </div>
+  {{end}}
+  </div>
+  {{end}}
+  {{end}}
+
+{{else if eq .Page "devices"}}
+  <h2>{{if .Admin}}Devices{{else}}Your devices{{end}}</h2>
+  {{with .Link}}
+  <div class="panel">
+    <h3>Enrolment link</h3>
+    <p class="lede">Single use, valid 24 hours. Treat it like a password until it is spent.</p>
+    <!-- On the phone being enrolled this is the whole procedure: the scheme is registered by the
+         app, so following the link hands it the token directly. Copying a 200-character string
+         between two devices is the step that goes wrong, and it does not have to happen at all. -->
+    {{with $.LinkHref}}<p><a class="btn" href="{{.}}">Open in the Echolot app</a></p>{{end}}
+    <p class="muted">Works on the phone you are enrolling. From another device, scan this:</p>
+    {{with $.LinkQR}}<div class="qr">{{.}}</div>{{end}}
+    <p class="muted">Or copy the link into the app's enrolment field, or deliver it over adb.</p>
+    <p><code>{{.}}</code></p>
+    <p class="muted mono">adb shell am start -a android.intent.action.VIEW -d "{{.}}"</p>
+  </div>
+  {{end}}
+  {{if .Admin}}
+  <form method="post" action="/enroll-tokens">
+    <input type="hidden" name="csrf" value="{{.CSRF}}">
+    <button>Create enrolment link</button>
+  </form>
+  {{end}}
+  {{if .Rows}}<div class="recs">
+  {{range .Rows}}
+    <div class="rec">
+      <div class="rec-head">
+        <span class="id">{{if .Name}}{{.Name}}{{else}}{{.ID}}{{end}}</span>
+        {{if .LinkedToAccount}}<span class="tag v-green">{{.AccountName}}</span>
+        {{else}}<span class="tag v-unknown">no account</span>{{end}}
+      </div>
+      <ul class="readout">
+        <li><span class="k">device</span><span class="lead"></span><span class="v">{{.ID}}</span></li>
+        <li><span class="k">enrolled</span><span class="lead"></span>
+            <span class="v">{{.Enrolled.Format "2006-01-02 15:04"}}</span></li>
+        <li><span class="k">runs</span><span class="lead"></span><span class="v">{{.Runs}}</span></li>
+      </ul>
+      {{if $.Admin}}
+      <form method="post" action="/devices/{{.ID}}/revoke">
+        <input type="hidden" name="csrf" value="{{$.CSRF}}">
+        <button class="danger">Revoke</button>
+      </form>
+      {{end}}
+    </div>
+  {{end}}
+  </div>{{else}}
+  <p class="empty">{{if .Admin}}No devices yet. Create an enrolment link and open it on the phone
+    you want to measure from.{{else}}No devices yet. Sign in from the Echolot app on your phone to
+    link one to this account.{{end}}</p>
+  {{end}}
+
+{{else if eq .Page "runs"}}
+  <h2>{{if .Admin}}Uploaded runs{{else}}Your uploaded runs{{end}}</h2>
+  <p class="lede">Each run is shown exactly as it arrived, at the privacy level its uploader chose.
+    Nothing here can un-redact one.</p>
+  {{if .Rows}}
+  <div class="trace">{{range .Rows}}<i class="{{verdictClass .Verdict}}"></i>{{end}}</div>
+  <p class="trace-key"><b>oldest &rarr; newest</b>
+    <b><em class="v-green"></em>clean</b>
+    <b><em class="v-yellow"></em>worth a look</b>
+    <b><em class="v-red"></em>faults</b>
+    <b><em class="v-inconclusive"></em>inconclusive</b></p>
+  {{end}}
+  {{if .Rows}}<div class="recs">
+  {{range .Rows}}
+    <div class="rec">
+      <div class="rec-head">
+        <span class="id">{{.UploadedAt.Format "2006-01-02 15:04"}}</span>
+        <span class="tag {{verdictClass .Verdict}}">{{verdictLabel .Verdict}}</span>
+      </div>
+      <ul class="readout">
+        <li><span class="k">device</span><span class="lead"></span><span class="v">{{.DeviceName}}</span></li>
+        <li><span class="k">findings</span><span class="lead"></span><span class="v">{{.FindingCount}}</span></li>
+        <li><span class="k">size</span><span class="lead"></span><span class="v">{{kb .SizeBytes}} kB</span></li>
+        <li><span class="k">privacy</span><span class="lead"></span><span class="v">{{.Anonymization}}</span></li>
+      </ul>
+      <div class="open"><a href="/runs/{{.DeviceID}}/{{.ID}}">Open run</a></div>
+    </div>
+  {{end}}
+  </div>{{else}}
+  <p class="empty">Nothing uploaded yet. Take a measurement in the app and upload it; it will
+    appear here.</p>
+  {{end}}
+
+{{else if eq .Page "run"}}
+  <h2>Run {{.ID}}</h2>
+  <p class="lede">The document as stored, indented for reading. Nothing has been added or removed.</p>
+  <form method="post" action="/runs/{{.Device}}/{{.ID}}/delete">
+    <input type="hidden" name="csrf" value="{{.CSRF}}">
+    <button class="danger">Delete this run</button>
+  </form>
+  <pre>{{.JSON}}</pre>
+{{end}}
+
+</main></body></html>
+`

server/internal/adminui/scope_test.go

@@ -0,0 +1,119 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package adminui
+
+import (
+	"testing"
+	"time"
+
+	"echo-lot.app/server/internal/adminauth"
+	"echo-lot.app/server/internal/runs"
+	"echo-lot.app/server/internal/store"
+)
+
+// Two accounts, one device each, plus an unlinked device nobody owns.
+func fixture(t *testing.T) (*Server, string, string, string) {
+	t.Helper()
+	dir := t.TempDir()
+	st, err := store.Open(dir)
+	if err != nil {
+		t.Fatal(err)
+	}
+	rs, err := runs.Open(dir, runs.DefaultPolicy())
+	if err != nil {
+		t.Fatal(err)
+	}
+	enroll := func(name string) string {
+		tok, err := st.NewEnrollToken(time.Hour, "test")
+		if err != nil {
+			t.Fatal(err)
+		}
+		d, err := st.Redeem(tok, name)
+		if err != nil {
+			t.Fatal(err)
+		}
+		return d.ID
+	}
+	mine, theirs, orphan := enroll("mine"), enroll("theirs"), enroll("orphan")
+	if err := st.LinkAccount(mine, "oidc#me", "Me"); err != nil {
+		t.Fatal(err)
+	}
+	if err := st.LinkAccount(theirs, "oidc#you", "You"); err != nil {
+		t.Fatal(err)
+	}
+	for _, d := range []string{mine, theirs, orphan} {
+		if _, err := rs.Put(d, []byte(`{"run":{"id":"r"}}`), true); err != nil {
+			t.Fatal(err)
+		}
+	}
+	return &Server{Store: st, Runs: rs}, mine, theirs, orphan
+}
+
+func user() *adminauth.Session  { return &adminauth.Session{Subject: "oidc#me", Display: "Me"} }
+func admin() *adminauth.Session { return &adminauth.Session{Subject: "local:a", Admin: true} }
+
+func TestVisibleDevicesScopesToAccount(t *testing.T) {
+	s, mine, theirs, orphan := fixture(t)
+
+	got := s.visibleDevices(user())
+	if len(got) != 1 || got[0].ID != mine {
+		t.Fatalf("a user should see only their own device, got %+v", got)
+	}
+
+	all := s.visibleDevices(admin())
+	if len(all) != 3 {
+		t.Fatalf("an admin should see every device, got %d", len(all))
+	}
+	_ = theirs
+	_ = orphan
+}
+
+func TestUnlinkedDevicesBelongToNobody(t *testing.T) {
+	// An enrolled but never-signed-in device is not "everyone's" — a user must not inherit it
+	// just because no account claimed it.
+	s, _, _, orphan := fixture(t)
+	if s.mayTouchRun(user(), orphan) {
+		t.Fatal("an unlinked device was treated as the user's own")
+	}
+}
+
+func TestRunAccessFollowsDeviceOwnership(t *testing.T) {
+	s, mine, theirs, _ := fixture(t)
+
+	if !s.mayTouchRun(user(), mine) {
+		t.Fatal("a user cannot reach their own run")
+	}
+	if s.mayTouchRun(user(), theirs) {
+		t.Fatal("a user reached someone else's run")
+	}
+	if !s.mayTouchRun(admin(), theirs) {
+		t.Fatal("an admin should reach any run")
+	}
+}
+
+func TestAccessEndsWhenTheLinkDoes(t *testing.T) {
+	// Ownership is read from the device list on every request rather than captured at sign-in,
+	// so unlinking takes effect immediately — a session issued while linked must not keep working.
+	s, mine, _, _ := fixture(t)
+	sess := user()
+	if !s.mayTouchRun(sess, mine) {
+		t.Fatal("precondition: the device should start out owned")
+	}
+	if err := s.Store.LinkAccount(mine, "", ""); err != nil {
+		t.Fatal(err)
+	}
+	if s.mayTouchRun(sess, mine) {
+		t.Fatal("access survived the account link being removed")
+	}
+}
+
+func TestEmptySubjectMatchesNothing(t *testing.T) {
+	// The dangerous degenerate case: a session with no subject must own nothing, not everything
+	// that happens to have an empty account id.
+	s, _, _, _ := fixture(t)
+	anon := &adminauth.Session{Subject: "", Display: ""}
+	if got := s.visibleDevices(anon); len(got) != 0 {
+		t.Fatalf("an empty subject matched %d devices", len(got))
+	}
+}

server/internal/canarydns/canarydns.go

@@ -63,20 +63,24 @@ type Server struct {
 	nsName    string // this server's own name for NS/authority answers
 	primaryV4 netip.Addr
 	primaryV6 netip.Addr
+	retention time.Duration // query-log age limit; <= 0 means only the ring cap bounds it
 
 	mu  sync.Mutex
 	log []Query // ring, newest last
-	retainTo time.Time
 }
 
 const logCap = 8192
 
 // New creates a server for zone (with or without trailing dot). nsName is the
 // server's own hostname (for the zone's NS record); primary v4/v6 are this
-// host's addresses used to answer the zone apex / NS glue.
-func New(zone, nsName string, v4, v6 netip.Addr) *Server {
+// host's addresses used to answer the zone apex / NS glue. retention is how
+// long logged queries are kept (spec §6; the privacy default is 24 h).
+func New(zone, nsName string, v4, v6 netip.Addr, retention time.Duration) *Server {
 	z := strings.ToLower(strings.TrimSuffix(zone, ".")) + "."
-	return &Server{zone: z, nsName: strings.TrimSuffix(nsName, ".") + ".", primaryV4: v4, primaryV6: v6}
+	return &Server{
+		zone: z, nsName: strings.TrimSuffix(nsName, ".") + ".",
+		primaryV4: v4, primaryV6: v6, retention: retention,
+	}
 }
 
 // RecentForPrefix returns logged queries whose qname contains ".<prefix>."
@@ -84,6 +88,7 @@ func New(zone, nsName string, v4, v6 netip.Addr) *Server {
 func (s *Server) RecentForPrefix(prefix string) []Query {
 	s.mu.Lock()
 	defer s.mu.Unlock()
+	s.dropExpiredLocked(time.Now().UTC())
 	needle := "." + strings.ToLower(prefix) + "."
 	var out []Query
 	for _, q := range s.log {
@@ -97,12 +102,34 @@ func (s *Server) RecentForPrefix(prefix string) []Query {
 func (s *Server) record(q Query) {
 	s.mu.Lock()
 	defer s.mu.Unlock()
+	s.dropExpiredLocked(q.At)
 	if len(s.log) >= logCap {
 		s.log = s.log[1:]
 	}
 	s.log = append(s.log, q)
 }
 
+// dropExpiredLocked enforces the retention window on the query log.
+//
+// The 24-hour retention was advertised as the privacy default (spec §6/§7) and then not
+// enforced: the ring only bounded *count*, so on a quiet server a resolver's queries could sit
+// in memory for weeks. Aged out on every write and every read — whichever comes first — so an
+// idle log still forgets on schedule the moment anyone looks. Entries are appended in time
+// order, so expiry is always a prefix of the slice.
+func (s *Server) dropExpiredLocked(now time.Time) {
+	if s.retention <= 0 {
+		return
+	}
+	cutoff := now.Add(-s.retention)
+	i := 0
+	for i < len(s.log) && s.log[i].At.Before(cutoff) {
+		i++
+	}
+	if i > 0 {
+		s.log = append([]Query(nil), s.log[i:]...) // reallocate so the old backing array frees
+	}
+}
+
 // ServeUDP / ServeTCP run read loops; call one per bound address.
 func (s *Server) ServeUDP(conn *net.UDPConn) error {
 	buf := make([]byte, 1500)

server/internal/canarydns/canarydns_test.go

@@ -8,6 +8,7 @@ import (
 	"net"
 	"net/netip"
 	"testing"
+	"time"
 )
 
 // buildQuery makes a single-question DNS query, optionally with an EDNS OPT.
@@ -65,7 +66,7 @@ func parseResponse(t *testing.T, resp []byte) (flags uint16, answers []ans) {
 }
 
 func newTestServer() *Server {
-	return New("c.echo-lot.app", "fmr", netip.MustParseAddr("192.0.2.1"), netip.MustParseAddr("2001:db8::1"))
+	return New("c.echo-lot.app", "fmr", netip.MustParseAddr("192.0.2.1"), netip.MustParseAddr("2001:db8::1"), 24*time.Hour)
 }
 
 func TestReferenceRecords(t *testing.T) {
@@ -159,3 +160,29 @@ func TestOutOfZoneNXDomain(t *testing.T) {
 		t.Fatalf("out-of-zone should be NXDOMAIN, flags=%#x", flags)
 	}
 }
+
+func TestQueryLogRetentionForgetsOldEntries(t *testing.T) {
+	s := newTestServer() // 24 h retention
+	now := time.Now().UTC()
+	s.record(Query{QName: "old.sess1.c.echo-lot.app", At: now.Add(-25 * time.Hour)})
+	s.record(Query{QName: "fresh.sess1.c.echo-lot.app", At: now})
+
+	got := s.RecentForPrefix("sess1")
+	if len(got) != 1 || got[0].QName != "fresh.sess1.c.echo-lot.app" {
+		t.Fatalf("retention not enforced: %+v", got)
+	}
+
+	// Reads must age the log too: an idle server still has to forget on schedule.
+	s.log[0].At = now.Add(-25 * time.Hour)
+	if got := s.RecentForPrefix("sess1"); len(got) != 0 {
+		t.Fatalf("read path did not expire entries: %+v", got)
+	}
+}
+
+func TestZeroRetentionKeepsEverything(t *testing.T) {
+	s := New("c.echo-lot.app", "fmr", netip.MustParseAddr("192.0.2.1"), netip.MustParseAddr("2001:db8::1"), 0)
+	s.record(Query{QName: "ancient.sess1.c.echo-lot.app", At: time.Now().UTC().Add(-1000 * time.Hour)})
+	if got := s.RecentForPrefix("sess1"); len(got) != 1 {
+		t.Fatal("retention 0 must mean 'ring cap only', not 'keep nothing'")
+	}
+}

server/internal/certreload/certreload.go

@@ -0,0 +1,124 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+// Package certreload serves a TLS certificate that can be replaced on disk without a restart.
+//
+// The hard part of TLS is never termination — the stdlib does that — it is renewal. A certificate
+// obtained from an ACME client expires every sixty days, and the usual arrangement is a renewal
+// hook that reloads or restarts the service. That hook is the part that quietly fails: it works
+// when it is written and then, months later, does not, and nobody notices until the certificate
+// has already expired.
+//
+// So the certificate is re-read when the file changes. There is no hook to forget, no reload to
+// coordinate, and a renewal that drops new files in place is picked up on the next handshake.
+//
+// Deliberately not used for the control plane. Clients pin that certificate's public key
+// (probe-protocol.md §1), so swapping it at runtime would silently break every enrolled device —
+// there the operator *should* have to think, and a restart is the least of what a key rotation
+// costs. Two listeners, two different right answers.
+package certreload
+
+import (
+	"crypto/tls"
+	"fmt"
+	"os"
+	"sync"
+	"time"
+)
+
+// Reloader holds a certificate and refreshes it when the files on disk change.
+type Reloader struct {
+	certPath, keyPath string
+
+	mu       sync.RWMutex
+	cert     *tls.Certificate
+	certMod  time.Time
+	keyMod   time.Time
+	checked  time.Time
+	interval time.Duration
+}
+
+// New loads the pair once so a bad path fails at startup rather than at the first handshake,
+// when the only symptom is a connection error at the far end.
+func New(certPath, keyPath string) (*Reloader, error) {
+	r := &Reloader{certPath: certPath, keyPath: keyPath, interval: 30 * time.Second}
+	if err := r.load(); err != nil {
+		return nil, err
+	}
+	return r, nil
+}
+
+// TLSConfig returns a config that asks this reloader for the certificate on every handshake.
+func (r *Reloader) TLSConfig() *tls.Config {
+	return &tls.Config{
+		MinVersion:     tls.VersionTLS12,
+		GetCertificate: r.getCertificate,
+	}
+}
+
+func (r *Reloader) getCertificate(*tls.ClientHelloInfo) (*tls.Certificate, error) {
+	r.maybeReload()
+	r.mu.RLock()
+	defer r.mu.RUnlock()
+	if r.cert == nil {
+		return nil, fmt.Errorf("no certificate loaded")
+	}
+	return r.cert, nil
+}
+
+// maybeReload stats the files at most once per interval.
+//
+// Rate-limited because this runs on every handshake: a busy listener would otherwise stat twice
+// per connection, and a certificate that is thirty seconds stale has never mattered to anyone.
+func (r *Reloader) maybeReload() {
+	r.mu.RLock()
+	fresh := time.Since(r.checked) < r.interval
+	r.mu.RUnlock()
+	if fresh {
+		return
+	}
+
+	r.mu.Lock()
+	r.checked = time.Now()
+	certMod, keyMod := modTime(r.certPath), modTime(r.keyPath)
+	unchanged := certMod.Equal(r.certMod) && keyMod.Equal(r.keyMod)
+	r.mu.Unlock()
+	if unchanged {
+		return
+	}
+	// A failed reload keeps the certificate already in memory. Renewal tools write the two files
+	// separately, so there is a window where the pair does not match; serving the previous
+	// certificate through that window is strictly better than serving none.
+	_ = r.load()
+}
+
+func (r *Reloader) load() error {
+	cert, err := tls.LoadX509KeyPair(r.certPath, r.keyPath)
+	if err != nil {
+		return fmt.Errorf("loading %s / %s: %w", r.certPath, r.keyPath, err)
+	}
+	r.mu.Lock()
+	defer r.mu.Unlock()
+	r.cert = &cert
+	r.certMod, r.keyMod = modTime(r.certPath), modTime(r.keyPath)
+	return nil
+}
+
+// NotAfter is when the loaded certificate expires, for the admin UI to show and for a startup
+// warning. An expiry an operator can see is one they can act on before a browser tells them.
+func (r *Reloader) NotAfter() time.Time {
+	r.mu.RLock()
+	defer r.mu.RUnlock()
+	if r.cert == nil || r.cert.Leaf == nil {
+		return time.Time{}
+	}
+	return r.cert.Leaf.NotAfter
+}
+
+func modTime(path string) time.Time {
+	fi, err := os.Stat(path)
+	if err != nil {
+		return time.Time{}
+	}
+	return fi.ModTime()
+}

server/internal/certreload/certreload_test.go

@@ -0,0 +1,119 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package certreload
+
+import (
+	"crypto/ecdsa"
+	"crypto/elliptic"
+	"crypto/rand"
+	"crypto/x509"
+	"crypto/x509/pkix"
+	"encoding/pem"
+	"math/big"
+	"os"
+	"path/filepath"
+	"testing"
+	"time"
+)
+
+func writePair(t *testing.T, dir, cn string) (string, string) {
+	t.Helper()
+	key, err := ecdsa.GenerateKey(elliptic.P256(), rand.Reader)
+	if err != nil {
+		t.Fatal(err)
+	}
+	tmpl := x509.Certificate{
+		SerialNumber: big.NewInt(time.Now().UnixNano()),
+		Subject:      pkix.Name{CommonName: cn},
+		NotBefore:    time.Now().Add(-time.Hour),
+		NotAfter:     time.Now().Add(24 * time.Hour),
+	}
+	der, err := x509.CreateCertificate(rand.Reader, &tmpl, &tmpl, &key.PublicKey, key)
+	if err != nil {
+		t.Fatal(err)
+	}
+	certPath := filepath.Join(dir, "cert.pem")
+	keyPath := filepath.Join(dir, "key.pem")
+	cb := pem.EncodeToMemory(&pem.Block{Type: "CERTIFICATE", Bytes: der})
+	kb, _ := x509.MarshalECPrivateKey(key)
+	if err := os.WriteFile(certPath, cb, 0o600); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(keyPath, pem.EncodeToMemory(&pem.Block{Type: "EC PRIVATE KEY", Bytes: kb}), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	return certPath, keyPath
+}
+
+func TestBadPathsFailAtStartupNotAtHandshake(t *testing.T) {
+	if _, err := New("/nonexistent/cert.pem", "/nonexistent/key.pem"); err == nil {
+		t.Fatal("a missing certificate was accepted; the failure would surface as an " +
+			"unexplained connection error at the client instead")
+	}
+}
+
+// The whole point: a renewal that drops new files in place is picked up without a restart and
+// without a reload hook that can silently stop working.
+func TestANewCertificateOnDiskIsPickedUp(t *testing.T) {
+	dir := t.TempDir()
+	certPath, keyPath := writePair(t, dir, "first")
+	r, err := New(certPath, keyPath)
+	if err != nil {
+		t.Fatal(err)
+	}
+	r.interval = 0 // check on every handshake, rather than waiting out the rate limit
+
+	got, err := r.getCertificate(nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+	first := got.Leaf
+
+	time.Sleep(10 * time.Millisecond) // ensure a distinct mtime
+	writePair(t, dir, "second")
+
+	got, err = r.getCertificate(nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if got.Leaf != nil && first != nil && got.Leaf.SerialNumber.Cmp(first.SerialNumber) == 0 {
+		t.Fatal("the replaced certificate was not picked up")
+	}
+}
+
+// Renewal tools write the certificate and the key separately, so there is a window where the two
+// do not match. Serving the previous certificate through it beats serving none.
+func TestAHalfWrittenPairKeepsTheOldCertificate(t *testing.T) {
+	dir := t.TempDir()
+	certPath, keyPath := writePair(t, dir, "good")
+	r, err := New(certPath, keyPath)
+	if err != nil {
+		t.Fatal(err)
+	}
+	r.interval = 0
+
+	time.Sleep(10 * time.Millisecond)
+	if err := os.WriteFile(certPath, []byte("-----BEGIN CERTIFICATE-----\ntruncated\n"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	got, err := r.getCertificate(nil)
+	if err != nil {
+		t.Fatalf("a torn write took the listener down: %v", err)
+	}
+	if got == nil {
+		t.Fatal("no certificate served during a torn write")
+	}
+}
+
+func TestExpiryIsVisible(t *testing.T) {
+	dir := t.TempDir()
+	certPath, keyPath := writePair(t, dir, "x")
+	r, err := New(certPath, keyPath)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if got := r.NotAfter(); got.IsZero() || time.Until(got) > 48*time.Hour {
+		t.Fatalf("expiry not reported sensibly: %v", got)
+	}
+}

server/internal/config/addrs_test.go

@@ -0,0 +1,56 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package config
+
+import "testing"
+
+// The fixture is fmr's real UDP listen spec, because the point of deriving these from the bound
+// listeners is that they cannot disagree with what the server actually answers on.
+const fmrUDP = "89.185.109.150:8442,89.185.109.151:8442," +
+	"[2001:1ad0:c4fe:6767::150]:8442,[2001:1ad0:c4fe:6767::151]:8442"
+
+func TestMeasurementAddrsSplitsPrimaryFromReserved(t *testing.T) {
+	c := &Config{
+		UDPListen:     fmrUDP,
+		ReservedAddrs: "89.185.109.151,2001:1ad0:c4fe:6767::151",
+	}
+	ip4, ip6, ip4Alt, ip6Alt := c.MeasurementAddrs()
+	for _, tc := range []struct{ got, want, name string }{
+		{ip4, "89.185.109.150", "ip4"},
+		{ip6, "2001:1ad0:c4fe:6767::150", "ip6"},
+		{ip4Alt, "89.185.109.151", "ip4_alt"},
+		{ip6Alt, "2001:1ad0:c4fe:6767::151", "ip6_alt"},
+	} {
+		if tc.got != tc.want {
+			t.Errorf("%s = %q, want %q", tc.name, tc.got, tc.want)
+		}
+	}
+}
+
+func TestMeasurementAddrsWithNothingReserved(t *testing.T) {
+	// No reservation means no alternate: reporting a second address as the RFC 5780 alternate
+	// when it was never set aside for that would tell a client to expect a redirect that the
+	// server has no intention of sending.
+	c := &Config{UDPListen: fmrUDP}
+	ip4, ip6, ip4Alt, ip6Alt := c.MeasurementAddrs()
+	if ip4 == "" || ip6 == "" {
+		t.Fatalf("primaries should still be found: ip4=%q ip6=%q", ip4, ip6)
+	}
+	if ip4Alt != "" || ip6Alt != "" {
+		t.Errorf("no address is reserved, so there is no alternate; got %q / %q", ip4Alt, ip6Alt)
+	}
+}
+
+func TestMeasurementAddrsIgnoresWhatItCannotRead(t *testing.T) {
+	// A wildcard bind names no address, and a hostname is not resolved here. Either would be a
+	// guess presented to clients as fact.
+	c := &Config{UDPListen: ":8442,probe.example.net:8442,89.185.109.150:8442"}
+	ip4, ip6, _, _ := c.MeasurementAddrs()
+	if ip4 != "89.185.109.150" {
+		t.Errorf("ip4 = %q, want the one address that was actually spelled out", ip4)
+	}
+	if ip6 != "" {
+		t.Errorf("ip6 = %q, want empty — none was configured", ip6)
+	}
+}

server/internal/config/config.go

@@ -10,6 +10,8 @@ package config
 import (
 	"flag"
 	"fmt"
+	"io"
+	"net"
 	"os"
 	"strconv"
 	"strings"
@@ -39,6 +41,24 @@ type Config struct {
 	// Admin UI / health listener (spec §7: localhost-only by default)
 	AdminListen string // ECHOLOT_ADMIN_LISTEN / --admin-listen
 
+	// ReservedAddrs are IPs reserved for measurement: addresses whose listening state must stay
+	// known, so that "nothing answered on port 443" is a fact about the network rather than a
+	// fact about this server's configuration. Enforced by CheckReserved.
+	ReservedAddrs string // ECHOLOT_RESERVED_ADDRS / --reserved-addrs
+
+	// ControlHostname lets the control plane share port 443 with the admin UI.
+	//
+	// They cannot share a certificate: the control plane is trusted by SPKI pin and so uses a
+	// long-lived self-signed certificate, while a browser needs one a CA vouches for. One name on
+	// one port means one certificate, so sharing the port requires two names — this one selects
+	// the pinned certificate and the control-plane routes by SNI, everything else gets the admin
+	// UI. Empty leaves the control plane on its own listener only.
+	//
+	// Why bother: captive portals and corporate firewalls routinely permit only 80 and 443, which
+	// are exactly the networks this tool exists to diagnose. A control plane on 8443 is
+	// unreachable precisely when it matters most.
+	ControlHostname string // ECHOLOT_CONTROL_HOSTNAME / --control-hostname
+
 	// State directory: device store, generated TLS material.
 	StateDir string // ECHOLOT_STATE_DIR / --state-dir
 
@@ -50,6 +70,11 @@ type Config struct {
 	// e.g. https://git.example.net/api/v1/repos/owner/repo
 	SelfUpdateAPI string // ECHOLOT_SELF_UPDATE_API / --self-update-api
 
+	// SelfUpdatePubKey overrides the release-signing public key baked into the binary
+	// (selfupdate.DefaultPublicKeyB64) — for operators running their own release pipeline
+	// against their own Gitea. Empty = the built-in project key.
+	SelfUpdatePubKey string // ECHOLOT_SELF_UPDATE_PUBKEY / --self-update-pubkey
+
 	// Uploaded-run storage. The default is "anonymous": any enrolled device may upload,
 	// which is what a self-hosted server wants. Operators of shared servers turn it down.
 	UploadsMode         string // ECHOLOT_UPLOADS / --uploads (off|anonymous|account)
@@ -58,6 +83,18 @@ type Config struct {
 	UploadMaxRuns       int    // ECHOLOT_UPLOAD_MAX_RUNS / --upload-max-runs (per device)
 	UploadMinAnon       string // ECHOLOT_UPLOAD_MIN_ANONYMIZATION / --upload-min-anonymization
 
+	// Rate limits (spec §2.5), each applied per credential and per source IP. 0 disables a
+	// ceiling. The UDP ceilings are deliberately above the largest legitimate run (a 200 Mbps
+	// upstream throughput test is ~21k pps of 1472-byte packets), so they only ever catch abuse
+	// — a rate limit that clips a real measurement produces a confidently wrong number.
+	RateSessionsPerMin int // ECHOLOT_RATE_SESSIONS_PER_MIN / --rate-sessions-per-min
+	RateActionsPerMin  int // ECHOLOT_RATE_ACTIONS_PER_MIN / --rate-actions-per-min
+	RateUDPPps         int // ECHOLOT_RATE_UDP_PPS / --rate-udp-pps
+	RateUDPKbps        int // ECHOLOT_RATE_UDP_KBPS / --rate-udp-kbps
+
+	// How long canary DNS query logs are kept, in hours (spec §6; privacy default 24).
+	DNSLogRetentionH int // ECHOLOT_DNS_LOG_RETENTION_H / --dns-log-retention-h
+
 	// Client compatibility window. Bounds are SemVer; an empty maximum means unbounded. The
 	// defaults sit at breaking boundaries, so shipping a patch never requires changing them.
 	MinAppVersion string // ECHOLOT_MIN_APP_VERSION / --min-app-version
@@ -67,10 +104,63 @@ type Config struct {
 	// first control listen address.
 	PublicControlURL string // ECHOLOT_PUBLIC_URL / --public-url
 
+	// Identity provider. Empty issuer disables sign-in entirely; the server is a relying
+	// party and never stores passwords.
+	OIDCIssuer      string // ECHOLOT_OIDC_ISSUER / --oidc-issuer
+	OIDCClientID    string // ECHOLOT_OIDC_CLIENT_ID / --oidc-client-id (confidential, admin UI)
+	OIDCAppClientID string // ECHOLOT_OIDC_APP_CLIENT_ID / --oidc-app-client-id (public, the phone app)
+	// Issuer for the app's client, when the IdP gives each application its own.
+	//
+	// Authentik derives the issuer from the application slug, so two applications mean two
+	// issuers — and a token's `iss` must match the one that minted it. Empty means both clients
+	// share ECHOLOT_OIDC_ISSUER, which is what IdPs with a single global issuer do.
+	OIDCAppIssuer  string // ECHOLOT_OIDC_APP_ISSUER / --oidc-app-issuer
+	OIDCAdminGroup string // ECHOLOT_OIDC_ADMIN_GROUP / --oidc-admin-group
+
+	// Break-glass admin username; the password lives hashed in the state store.
+	AdminUser string // ECHOLOT_ADMIN_USER / --admin-user
+
+	// Secret for the *confidential* admin client. Prefer ECHOLOT_OIDC_CLIENT_SECRET_FILE: a path
+	// keeps the secret out of the environment, where it is readable by anything that can see
+	// /proc/<pid>/environ and lands in every dump of the unit's configuration.
+	OIDCClientSecret string // ECHOLOT_OIDC_CLIENT_SECRET / _FILE
+
+	// Where the admin UI is reachable, used to build the OIDC redirect URI. Must match what is
+	// registered at the IdP exactly.
+	AdminBaseURL string // ECHOLOT_ADMIN_BASE_URL / --admin-base-url
+	// TLS for the admin listener. Without these it serves plaintext, which is only acceptable on
+	// loopback — see checkAdminExposure.
+	AdminTLSCert string // ECHOLOT_ADMIN_TLS_CERT / --admin-tls-cert
+	AdminTLSKey  string // ECHOLOT_ADMIN_TLS_KEY / --admin-tls-key
+	// Deliberate override for serving the admin UI in plaintext off loopback, so that decision
+	// is made rather than stumbled into.
+	AdminInsecure bool // ECHOLOT_ADMIN_INSECURE / --admin-insecure
+
+	// Port-80 listener that answers ACME HTTP-01 challenges and redirects everything else to
+	// the admin UI. Empty disables it. HTTP-01 always arrives on port 80 — the CA picks the
+	// port — so this never collides with the admin UI on 443.
+	ACMEHTTPListen string // ECHOLOT_ACME_HTTP_LISTEN / --acme-http-listen
+	// Directory an ACME client writes challenge tokens into. Defaults to <state-dir>/acme.
+	ACMEWebroot string // ECHOLOT_ACME_WEBROOT / --acme-webroot
+
 	// Mode
 	Docker bool // --docker (or autodetected; env ECHOLOT_DOCKER=1 forces)
 }
 
+// secretOr reads ECHOLOT_<key>, or the contents of the file named by ECHOLOT_<key>_FILE.
+//
+// The file form exists because a secret in the environment is readable by anything that can see
+// /proc/<pid>/environ and lands in every dump of the unit's configuration. A path costs nothing
+// and keeps the value in one file whose permissions an operator can reason about.
+func secretOr(key, def string) string {
+	if path := envOr(key+"_FILE", ""); path != "" {
+		if b, err := os.ReadFile(path); err == nil {
+			return strings.TrimSpace(string(b))
+		}
+	}
+	return envOr(key, def)
+}
+
 // envInt reads ECHOLOT_<key> as an integer with a fallback.
 func envInt(key string, def int) int {
 	if v := envOr(key, ""); v != "" {
@@ -107,21 +197,48 @@ func Load(args []string) (*Config, *Actions, error) {
 	fs.StringVar(&c.HTTPEchoListen, "http-echo-listen", envOr("HTTP_ECHO_LISTEN", ""), "optional CLEARTEXT http-echo listen address(es); empty disables (spec §4)")
 	fs.StringVar(&c.MTUProbeTargets, "mtu-probe-targets", envOr("MTU_PROBE_TARGETS", "1.1.1.1,2606:4700:4700::1111"), "egress-MTU self-proof anchors, comma-separated")
 	fs.StringVar(&c.AdminListen, "admin-listen", envOr("ADMIN_LISTEN", "127.0.0.1:8444"), "admin/health listen address (keep localhost)")
+	fs.StringVar(&c.ControlHostname, "control-hostname", envOr("CONTROL_HOSTNAME", ""), "hostname that selects the pinned control-plane certificate when sharing the admin UI's port")
+	fs.StringVar(&c.ReservedAddrs, "reserved-addrs", envOr("RESERVED_ADDRS", ""), "comma-separated IPs reserved for measurement; no listener but the STUN alternate may bind them")
 	fs.StringVar(&c.StateDir, "state-dir", envOr("STATE_DIR", defaultStateDir()), "state directory (device store, generated TLS)")
 	fs.StringVar(&c.Name, "name", envOr("NAME", "echolot"), "server profile name")
 	fs.StringVar(&c.SelfUpdateAPI, "self-update-api", envOr("SELF_UPDATE_API", ""), "Gitea repo API base for self-update; empty disables")
+	fs.StringVar(&c.SelfUpdatePubKey, "self-update-pubkey", envOr("SELF_UPDATE_PUBKEY", ""), "release-signing public key (base64 ed25519) self-update verifies against; empty uses the built-in project key")
 	fs.StringVar(&c.UploadsMode, "uploads", envOr("UPLOADS", "anonymous"), "who may upload measurement runs: off|anonymous|account")
 	fs.Int64Var(&c.UploadMaxBytes, "upload-max-bytes", int64(envInt("UPLOAD_MAX_BYTES", 4<<20)), "largest accepted uploaded run, bytes")
 	fs.IntVar(&c.UploadRetentionDays, "upload-retention-days", envInt("UPLOAD_RETENTION_DAYS", 90), "delete uploaded runs older than this; 0 disables")
 	fs.IntVar(&c.UploadMaxRuns, "upload-max-runs", envInt("UPLOAD_MAX_RUNS", 200), "keep at most this many runs per device; 0 disables")
 	fs.StringVar(&c.UploadMinAnon, "upload-min-anonymization", envOr("UPLOAD_MIN_ANONYMIZATION", "full"), "least anonymization accepted: full|balanced|strict")
+	fs.StringVar(&c.OIDCIssuer, "oidc-issuer", envOr("OIDC_ISSUER", ""), "OpenID Connect issuer URL; empty disables sign-in")
+	fs.StringVar(&c.OIDCClientID, "oidc-client-id", envOr("OIDC_CLIENT_ID", ""), "confidential OIDC client id for the admin UI")
+	fs.StringVar(&c.OIDCAppClientID, "oidc-app-client-id", envOr("OIDC_APP_CLIENT_ID", ""), "public OIDC client id used by the Android app (PKCE)")
+	fs.StringVar(&c.OIDCAppIssuer, "oidc-app-issuer", envOr("OIDC_APP_ISSUER", ""), "issuer for the app client when the IdP uses per-application issuers; empty = same as --oidc-issuer")
+	fs.StringVar(&c.OIDCAdminGroup, "oidc-admin-group", envOr("OIDC_ADMIN_GROUP", ""), "group claim required for admin access; empty means nobody is an admin via OIDC")
+	fs.StringVar(&c.OIDCClientSecret, "oidc-client-secret", secretOr("OIDC_CLIENT_SECRET", ""), "secret for the confidential admin client; prefer ECHOLOT_OIDC_CLIENT_SECRET_FILE")
+	fs.StringVar(&c.AdminBaseURL, "admin-base-url", envOr("ADMIN_BASE_URL", ""), "public URL of the admin UI, for the OIDC redirect (e.g. https://admin.example.net)")
+	fs.StringVar(&c.AdminTLSCert, "admin-tls-cert", envOr("ADMIN_TLS_CERT", ""), "TLS certificate for the admin listener")
+	fs.StringVar(&c.AdminTLSKey, "admin-tls-key", envOr("ADMIN_TLS_KEY", ""), "TLS key for the admin listener")
+	fs.BoolVar(&c.AdminInsecure, "admin-insecure", envOr("ADMIN_INSECURE", "") == "1", "allow the admin UI in plaintext off loopback (you are on your own)")
+	fs.StringVar(&c.ACMEHTTPListen, "acme-http-listen", envOr("ACME_HTTP_LISTEN", ""), "port-80 listener for ACME HTTP-01 challenges and http->https redirects")
+	fs.StringVar(&c.ACMEWebroot, "acme-webroot", envOr("ACME_WEBROOT", ""), "directory an ACME client writes challenges into (default <state-dir>/acme)")
 	fs.StringVar(&c.PublicControlURL, "public-url", envOr("PUBLIC_URL", ""), "public control-plane URL for enrollment links, e.g. https://probe.example.net:8443")
+	fs.IntVar(&c.RateSessionsPerMin, "rate-sessions-per-min", envInt("RATE_SESSIONS_PER_MIN", 10), "per-credential and per-IP ceiling on session creation (spec §2.5); 0 disables")
+	fs.IntVar(&c.RateActionsPerMin, "rate-actions-per-min", envInt("RATE_ACTIONS_PER_MIN", 60), "per-credential and per-IP ceiling on §5 actions; 0 disables")
+	fs.IntVar(&c.RateUDPPps, "rate-udp-pps", envInt("RATE_UDP_PPS", 25_000), "per-credential and per-IP data-plane packet ceiling, packets/s, silent drop; 0 disables")
+	fs.IntVar(&c.RateUDPKbps, "rate-udp-kbps", envInt("RATE_UDP_KBPS", 250_000), "per-credential and per-IP data-plane byte ceiling, kbit/s, silent drop; 0 disables")
+	fs.IntVar(&c.DNSLogRetentionH, "dns-log-retention-h", envInt("DNS_LOG_RETENTION_H", 24), "hours canary DNS query logs are kept (spec §6 privacy default 24); 0 keeps until the ring overwrites")
 	fs.StringVar(&c.MinAppVersion, "min-app-version", envOr("MIN_APP_VERSION", "0.2.0"), "oldest app version this server will serve (SemVer, inclusive)")
 	fs.StringVar(&c.MaxAppVersion, "max-app-version", envOr("MAX_APP_VERSION", "1.0.0"), "first app version this server will refuse (SemVer, exclusive); empty = unbounded")
 	fs.BoolVar(&c.Docker, "docker", envOr("DOCKER", "") == "1", "force container mode (config from env, no systemd/self-update)")
 
 	fs.BoolVar(&a.InstallSystemd, "install-systemd", false, "install a systemd unit for this binary and exit")
 	fs.BoolVar(&a.UninstallSystemd, "uninstall-systemd", false, "remove the systemd unit and exit")
+	var daemon bool
+	fs.BoolVar(&a.Serve, "serve", false, "run the server (bind listeners and answer requests)")
+	fs.BoolVar(&daemon, "daemon", false, "alias for --serve")
+	fs.BoolVar(&a.SetAdminPassword, "set-admin-password", false,
+		"set the break-glass admin password (username as --admin-user, password read from stdin) and exit")
+	fs.StringVar(&c.AdminUser, "admin-user", envOr("ADMIN_USER", "admin"), "username for the break-glass admin")
+	fs.StringVar(&a.MintEnrollToken, "mint-enroll-token", "", "mint a single-use enrollment link (argument is a note for the audit log) and exit")
 	fs.BoolVar(&a.SelfUpdate, "self-update", false, "check for a newer release, replace this binary, and exit")
 	fs.BoolVar(&a.Version, "version", false, "print version and exit")
 
@@ -131,12 +248,141 @@ func Load(args []string) (*Config, *Actions, error) {
 	if !c.Docker {
 		c.Docker = inContainer()
 	}
+	a.Serve = a.Serve || daemon
+	// No verb at all means the caller has not said what they want. Usage is the answer, and it
+	// is a usage error rather than success — otherwise a service manager sees a clean exit and
+	// concludes the server ran and finished.
+	if !a.Serve && !a.InstallSystemd && !a.UninstallSystemd && !a.SelfUpdate &&
+		!a.SetAdminPassword && !a.Version && a.MintEnrollToken == "" {
+		a.Help = true
+	}
+	if a.Serve {
+		if err := c.checkAdminExposure(); err != nil {
+			return nil, nil, err
+		}
+		if err := c.CheckReserved(c.Listeners()); err != nil {
+			return nil, nil, err
+		}
+	}
 	if c.Docker && (a.InstallSystemd || a.UninstallSystemd || a.SelfUpdate) {
 		return nil, nil, fmt.Errorf("systemd/self-update actions are native-mode only (container detected; override with ECHOLOT_DOCKER=0 if this is wrong)")
 	}
 	return c, a, nil
 }
 
+// Usage prints the verbs first and the tuning flags second, because the question someone has
+// when they run this by name is "what does it do", not "what can I set".
+func Usage(w io.Writer) {
+	fmt.Fprint(w, `echolot-server — the Echolot probe server
+
+USAGE
+  echolot-server --serve                 run the server
+  echolot-server --version               print the version
+  echolot-server --install-systemd       install and enable a systemd unit
+  echolot-server --uninstall-systemd     remove it
+  echolot-server --self-update           replace this binary with the latest release
+  echolot-server --set-admin-password    set the break-glass admin password (stdin)
+  echolot-server --help                  full flag list
+
+Every flag can also be set as an environment variable: --control-listen becomes
+ECHOLOT_CONTROL_LISTEN. In a container, configuration comes from the environment.
+
+Running with no verb prints this and exits non-zero: starting to serve the internet
+should be something you asked for.
+`)
+}
+
+// checkAdminExposure refuses to serve an unencrypted admin UI on a non-loopback address.
+//
+// The admin session cookie is a bearer credential for everything this server can do, and the OIDC
+// authorization code arrives in a URL. In plaintext, both are readable by anyone on the path — and
+// on a globally routable address "the path" means the internet. This is a hard stop rather than a
+// warning because a warning in a log is not read by the person who most needs it, and because the
+// two safe answers are cheap: bind to loopback and tunnel, or supply a certificate.
+func (c *Config) checkAdminExposure() error {
+	if c.AdminTLSCert != "" || c.AdminInsecure {
+		return nil
+	}
+	for _, addr := range Addrs(c.AdminListen) {
+		host, _, err := net.SplitHostPort(addr)
+		if err != nil {
+			continue
+		}
+		ip := net.ParseIP(strings.Trim(host, "[]"))
+		if host == "" || ip == nil || ip.IsLoopback() {
+			continue // loopback, or a name we cannot judge; RFC 8252 blesses loopback plaintext
+		}
+		return fmt.Errorf(
+			"refusing to serve the admin UI in plaintext on %s: the session cookie and the OIDC "+
+				"authorization code would cross the network in the clear.\n"+
+				"  Fix it one of three ways:\n"+
+				"    - bind to 127.0.0.1 and reach it over an SSH tunnel (no certificate needed)\n"+
+				"    - set ECHOLOT_ADMIN_TLS_CERT and ECHOLOT_ADMIN_TLS_KEY\n"+
+				"    - set ECHOLOT_ADMIN_INSECURE=1 if you genuinely mean it", addr)
+	}
+	return nil
+}
+
+// Listeners enumerates every configured listen spec, for CheckReserved.
+//
+// Kept as one list here rather than checked at each call site, so a listener added later is
+// caught by the compiler when this function is updated — and, more to the point, so that the
+// person adding one sees the reserved-address rule exists at all.
+func (c *Config) Listeners() []Listener {
+	return []Listener{
+		// The instrument: these belong on the reserved addresses as much as anywhere.
+		{Name: "control-listen", Spec: c.ControlListen, Measurement: true},
+		{Name: "udp-listen", Spec: c.UDPListen, Measurement: true},
+		{Name: "tcp-listen", Spec: c.TCPListen, Measurement: true},
+		{Name: "dns-listen", Spec: c.DNSListen, Measurement: true},
+		{Name: "stun-listen", Spec: c.StunListen, Measurement: true},
+		// http-echo is deliberately not marked as measurement: it is cleartext HTTP, so on a
+		// reserved address it would be the very listener that ruins the port-80 test.
+		{Name: "http-echo-listen", Spec: c.HTTPEchoListen},
+		// Services. These have no business on an address kept for measuring.
+		{Name: "admin-listen", Spec: c.AdminListen},
+		{Name: "acme-http-listen", Spec: c.ACMEHTTPListen},
+	}
+}
+
+// MeasurementAddrs picks out the addresses this server can be measured on, by family, splitting
+// primaries from the reserved alternates.
+//
+// Derived from what is actually bound rather than configured separately: a second list of the
+// server's own addresses is a second thing to keep in step, and the copy that drifts is the one
+// clients are told about.
+func (c *Config) MeasurementAddrs() (ip4, ip6, ip4Alt, ip6Alt string) {
+	reserved := map[string]bool{}
+	for _, ip := range c.ReservedIPs() {
+		reserved[ip.String()] = true
+	}
+	// The UDP data plane binds every address a client may be pointed at, which makes it the
+	// honest source for this.
+	for _, a := range Addrs(c.UDPListen) {
+		host, _, err := net.SplitHostPort(a)
+		if err != nil {
+			continue
+		}
+		host = strings.Trim(host, "[]")
+		ip := net.ParseIP(host)
+		if ip == nil {
+			continue
+		}
+		alt := reserved[ip.String()]
+		switch {
+		case ip.To4() != nil && alt && ip4Alt == "":
+			ip4Alt = host
+		case ip.To4() != nil && !alt && ip4 == "":
+			ip4 = host
+		case ip.To4() == nil && alt && ip6Alt == "":
+			ip6Alt = host
+		case ip.To4() == nil && !alt && ip6 == "":
+			ip6 = host
+		}
+	}
+	return
+}
+
 // Addrs splits a comma-separated listen spec into individual addresses.
 // Explicit per-address binds matter on multi-IP hosts: a wildcard bind
 // (":8443") would also claim addresses reserved for other purposes (e.g. an
@@ -151,11 +397,26 @@ func Addrs(spec string) []string {
 	return out
 }
 
-// Actions are one-shot verbs that exit instead of serving.
+// Actions are the verbs. Serving is one of them, and it is explicit: running the binary with no
+// arguments prints usage rather than binding a dozen ports and starting to answer the internet.
+// Someone typing the name of an unfamiliar program on a terminal should be told what it does, not
+// have it start doing it.
 type Actions struct {
+	Serve bool
+	// Help is set when there is nothing to do: no verb was given.
+	Help bool
+
 	InstallSystemd   bool
 	UninstallSystemd bool
 	SelfUpdate       bool
+	SetAdminPassword bool
+	// MintEnrollToken is the note to record against a freshly minted enrollment link.
+	//
+	// A local action rather than an HTTP endpoint: whoever can run this binary against the state
+	// directory already has every privilege the server has, so authenticating them to themselves
+	// would be theatre — and an unauthenticated endpoint on loopback is how the admin API was
+	// briefly reachable from the network by accident.
+	MintEnrollToken string
 	Version          bool
 }
 

server/internal/config/reserved.go

@@ -0,0 +1,123 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package config
+
+import (
+	"fmt"
+	"net"
+	"sort"
+	"strings"
+)
+
+// Listener is one configured listen spec, named for the error message.
+type Listener struct {
+	Name string // the flag/env this came from, e.g. "control-listen"
+	Spec string // comma-separated listen addresses
+	// Measurement marks a listener that is part of the instrument rather than a service on the
+	// host. Those belong on the reserved addresses — STUN's RFC 5780 alternate, the UDP data
+	// plane, the canary DNS — and reserving an address only to forbid the measurements that need
+	// it would defeat the purpose.
+	Measurement bool
+}
+
+// webPorts are the ports whose closed state on a reserved address is itself the measurement.
+//
+// A TLS handshake that completes on a port known not to be listening proves interception, with no
+// competing explanation. That proof is the whole reason for reserving an address, and it survives
+// exactly as long as nothing binds these two ports there.
+var webPorts = map[string]bool{"80": true, "443": true}
+
+// CheckReserved refuses to start when a listener would occupy an address reserved for measurement.
+//
+// The reserved addresses are the instrument, not the service. Their diagnostic value comes from
+// their listening state being *known*: if nothing listens on port 443 there, then a TLS handshake
+// that completes proves something on the path intercepted it, with no other explanation available.
+// One stray listener silently converts that proof into an ambiguity.
+//
+// This is a hard stop rather than a warning for the same reason as [Config.checkAdminExposure]: the
+// failure is invisible. A polluted reserved address does not crash, log, or behave oddly — it just
+// quietly turns a conclusive test into an inconclusive one, and the first symptom is a measurement
+// that says the network is clean when it is not. Nobody reads a warning for that.
+//
+// Wildcard binds are the realistic way this happens. Every listener defaults to ":port", and the
+// next one added will be copied from an existing default; that binds every address on the host,
+// reserved ones included, without anyone deciding to.
+func (c *Config) CheckReserved(listeners []Listener) error {
+	reserved := c.ReservedIPs()
+	if len(reserved) == 0 {
+		return nil
+	}
+	var problems []string
+	for _, l := range listeners {
+		for _, addr := range Addrs(l.Spec) {
+			host, port, err := net.SplitHostPort(addr)
+			if err != nil {
+				// Not host:port — a bare port or something malformed. Leave it to the listener
+				// itself to complain; guessing here would produce a confusing error about the
+				// wrong problem.
+				continue
+			}
+			host = strings.Trim(host, "[]")
+			if host == "" || host == "0.0.0.0" || host == "::" {
+				problems = append(problems, fmt.Sprintf(
+					"  --%s=%q binds every address on this host, including the reserved ones",
+					l.Name, addr))
+				continue
+			}
+			ip := net.ParseIP(host)
+			if ip == nil {
+				continue // a hostname; cannot resolve it here without lying about what we checked
+			}
+			for _, r := range reserved {
+				if !ip.Equal(r) {
+					continue
+				}
+				switch {
+				case webPorts[port]:
+					problems = append(problems, fmt.Sprintf(
+						"  --%s=%q puts port %s on reserved address %s, which is the one thing "+
+							"that address exists to keep closed", l.Name, addr, port, r))
+				case !l.Measurement:
+					problems = append(problems, fmt.Sprintf(
+						"  --%s=%q binds reserved address %s; only measurement listeners belong there",
+						l.Name, addr, r))
+				}
+			}
+		}
+	}
+	if len(problems) == 0 {
+		return nil
+	}
+	sort.Strings(problems)
+	return fmt.Errorf(
+		"refusing to start: these listeners would occupy addresses reserved for measurement\n%s\n"+
+			"\nReserved: %s\n"+
+			"Those addresses are the instrument. A test can only prove interception on a port that\n"+
+			"is known not to be listening, so anything bound there destroys the conclusion rather\n"+
+			"than merely sharing the address.\n"+
+			"  Fix it one of three ways:\n"+
+			"    - bind each listener to explicit service addresses instead of a wildcard\n"+
+			"    - remove the address from ECHOLOT_RESERVED_ADDRS if it is no longer reserved\n"+
+			"    - unset ECHOLOT_RESERVED_ADDRS if this host has no reserved addresses",
+		strings.Join(problems, "\n"), joinIPs(reserved))
+}
+
+// ReservedIPs parses the configured reserved addresses, ignoring anything unparseable.
+func (c *Config) ReservedIPs() []net.IP {
+	var out []net.IP
+	for _, s := range Addrs(c.ReservedAddrs) {
+		if ip := net.ParseIP(strings.Trim(s, "[]")); ip != nil {
+			out = append(out, ip)
+		}
+	}
+	return out
+}
+
+func joinIPs(ips []net.IP) string {
+	s := make([]string, 0, len(ips))
+	for _, ip := range ips {
+		s = append(s, ip.String())
+	}
+	return strings.Join(s, ", ")
+}

server/internal/config/reserved_test.go

@@ -0,0 +1,151 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package config
+
+import (
+	"strings"
+	"testing"
+)
+
+const (
+	svc4 = "89.185.109.150"
+	res4 = "89.185.109.151"
+	res6 = "2001:1ad0:c4fe:6767::151"
+)
+
+func withReserved(l ...Listener) error {
+	c := &Config{ReservedAddrs: res4 + "," + res6}
+	return c.CheckReserved(l)
+}
+
+func TestWildcardBindIsRefused(t *testing.T) {
+	// The realistic failure: every listener defaults to ":port", and the next one added gets
+	// copied from an existing default. Nobody decides to claim the reserved address; it just
+	// happens, and nothing looks wrong afterwards.
+	err := withReserved(Listener{Name: "control-listen", Spec: ":8443"})
+	if err == nil {
+		t.Fatal("a wildcard bind was allowed while addresses were reserved")
+	}
+	if !strings.Contains(err.Error(), "binds every address") {
+		t.Fatalf("the error should say why a wildcard is the problem, got: %v", err)
+	}
+}
+
+func TestWebPortsOnReservedAreRefusedEvenForMeasurement(t *testing.T) {
+	// The strictest rule, and the one carrying the diagnostic value: 80 and 443 must stay closed
+	// on a reserved address whatever wants them, because their closed state *is* the measurement.
+	for _, spec := range []string{res4 + ":443", "[" + res6 + "]:80"} {
+		err := withReserved(Listener{Name: "control-listen", Spec: spec, Measurement: true})
+		if err == nil {
+			t.Fatalf("port 80/443 on a reserved address was allowed: %q", spec)
+		}
+		if !strings.Contains(err.Error(), "keep closed") {
+			t.Errorf("the error should explain what is lost, got: %v", err)
+		}
+	}
+}
+
+func TestServiceOnReservedIsRefused(t *testing.T) {
+	if err := withReserved(Listener{Name: "admin-listen", Spec: res4 + ":8444"}); err == nil {
+		t.Fatal("a service was allowed onto a reserved address")
+	}
+}
+
+func TestMeasurementListenersBelongOnReserved(t *testing.T) {
+	// The live fmr config: the UDP data plane, canary DNS and STUN all bind the reserved pair on
+	// purpose. A guard that refused this would be describing a rule nobody wants.
+	err := withReserved(
+		Listener{Name: "udp-listen", Spec: res4 + ":8442,[" + res6 + "]:8442", Measurement: true},
+		Listener{Name: "dns-listen", Spec: res4 + ":53,[" + res6 + "]:53", Measurement: true},
+		Listener{Name: "stun-listen", Spec: res4 + ":3478", Measurement: true},
+	)
+	if err != nil {
+		t.Fatalf("measurement listeners must be allowed on reserved addresses: %v", err)
+	}
+}
+
+func TestServiceAddressesAreFine(t *testing.T) {
+	err := withReserved(
+		Listener{Name: "control-listen", Spec: svc4 + ":8443,[2001:1ad0:c4fe:6767::150]:8443"},
+		Listener{Name: "admin-listen", Spec: "127.0.0.1:8444"},
+	)
+	if err != nil {
+		t.Fatalf("service addresses should be allowed: %v", err)
+	}
+}
+
+func TestHttpEchoIsNotTreatedAsMeasurement(t *testing.T) {
+	// http-echo is cleartext HTTP. On a reserved address it is precisely the listener that would
+	// ruin the port-80 test, so it does not get the measurement exemption.
+	if err := withReserved(Listener{Name: "http-echo-listen", Spec: res4 + ":8080"}); err == nil {
+		t.Fatal("http-echo was allowed onto a reserved address")
+	}
+}
+
+func TestNoReservationMeansNoOpinion(t *testing.T) {
+	// A host with nothing reserved must keep working exactly as before, wildcards included.
+	c := &Config{}
+	if err := c.CheckReserved([]Listener{{Name: "control-listen", Spec: ":8443"}}); err != nil {
+		t.Fatalf("with no reserved addresses this must not interfere: %v", err)
+	}
+}
+
+func TestEveryOffenderIsNamed(t *testing.T) {
+	// Reporting one problem at a time turns a config fix into several restart cycles, and on a
+	// remote host each cycle is a chance to lock yourself out.
+	err := withReserved(
+		Listener{Name: "control-listen", Spec: ":8443"},
+		Listener{Name: "tcp-listen", Spec: res4 + ":8441"},
+	)
+	if err == nil {
+		t.Fatal("expected a refusal")
+	}
+	for _, want := range []string{"control-listen", "tcp-listen"} {
+		if !strings.Contains(err.Error(), want) {
+			t.Errorf("the error should name %s; got: %v", want, err)
+		}
+	}
+}
+
+func TestHostnamesAreNotGuessedAt(t *testing.T) {
+	// Resolving here would check a name against whatever DNS says at startup, which is not
+	// necessarily what it will say later — and a guard that is sometimes right is worse than one
+	// with a stated limit.
+	if err := withReserved(Listener{Name: "control-listen", Spec: "fmr-1.echo-lot.app:8443"}); err != nil {
+		t.Fatalf("a hostname must be left alone, not resolved: %v", err)
+	}
+}
+
+func TestListenersCoversEverySpec(t *testing.T) {
+	// A listener missing from Listeners() is invisible to the guard, which is the one way this
+	// protection fails silently. Fill every spec with the reserved address: each one that is
+	// actually enumerated produces a complaint naming it.
+	// Every spec on port 443 of the reserved address: the web-port rule applies to measurement
+	// listeners too, so each one that is genuinely enumerated must produce a complaint.
+	c := &Config{
+		ReservedAddrs:  res4,
+		ControlListen:  res4 + ":443",
+		UDPListen:      res4 + ":443",
+		TCPListen:      res4 + ":443",
+		DNSListen:      res4 + ":443",
+		HTTPEchoListen: res4 + ":443",
+		AdminListen:    res4 + ":443",
+		ACMEHTTPListen: res4 + ":443",
+		StunListen:     res4 + ":443",
+	}
+	err := c.CheckReserved(c.Listeners())
+	if err == nil {
+		t.Fatal("expected a refusal")
+	}
+	// Every spec is on the reserved address; the web-port rule catches even the measurement ones,
+	// so anything missing from Listeners() is invisible here and that is what this asserts.
+	for _, want := range []string{
+		"control-listen", "udp-listen", "tcp-listen", "dns-listen",
+		"http-echo-listen", "admin-listen", "acme-http-listen", "stun-listen",
+	} {
+		if !strings.Contains(err.Error(), want) {
+			t.Errorf("%s is not enumerated in Listeners(), so the guard cannot see it", want)
+		}
+	}
+}

server/internal/control/control.go

@@ -7,6 +7,7 @@
 package control
 
 import (
+	"context"
 	"crypto/rand"
 	"crypto/sha256"
 	"crypto/tls"
@@ -15,6 +16,7 @@ import (
 	"encoding/hex"
 	"encoding/json"
 	"errors"
+	"fmt"
 	"io"
 	"log/slog"
 	"net"
@@ -27,6 +29,8 @@ import (
 
 	"echo-lot.app/server/internal/compat"
 	"echo-lot.app/server/internal/dataplane"
+	"echo-lot.app/server/internal/oidc"
+	"echo-lot.app/server/internal/ratelimit"
 	"echo-lot.app/server/internal/runs"
 	"echo-lot.app/server/internal/session"
 	"echo-lot.app/server/internal/store"
@@ -55,13 +59,22 @@ type Server struct {
 	// DelayedEcho schedules/sends a DELAYED_ECHO for a session (may be nil).
 	DelayedEcho func(sess *session.Session, actionID string) error
 	// Granted server->client sends (spec §5). Both consume an asymmetric grant.
-	DownTrain func(sess *session.Session, g *session.Grant, count, sizeBytes, intervalUs int) (int, error)
+	// DownTrain's dscp is -1 for "leave the socket's default marking alone".
+	DownTrain func(sess *session.Session, g *session.Grant, count, sizeBytes, intervalUs, dscp int) (int, error)
 	BigSend   func(sess *session.Session, g *session.Grant, sizes []int, df bool) ([]dataplane.BigSendResult, error)
+	// OIDC verifies ID tokens presented by the *app* (may be nil).
+	OIDC *oidc.Verifier
+	// AdminOIDC verifies tokens from the admin UI's own client. Separate because an IdP may
+	// give each application its own issuer — Authentik derives it from the application slug —
+	// and a verifier pins exactly one issuer and the clients belonging to it.
+	AdminOIDC *oidc.Verifier
 	// Runs stores uploaded measurement documents (may be nil: uploads unsupported).
 	Runs *runs.Store
 	// FragSend emits one datagram as hand-built IP fragments in a chosen order (may be nil:
 	// needs a raw socket, so it is unavailable to an unprivileged server).
 	FragSend func(sess *session.Session, g *session.Grant, sizeBytes int, mode dataplane.FragMode, fragSize int) (dataplane.FragResult, error)
+	// DownThroughput sends paced traffic toward the client for a bounded time (may be nil).
+	DownThroughput func(sess *session.Session, g *session.Grant, durationMs, kbps, sizeBytes int) (dataplane.ThroughputResult, error)
 	// EgressMTU reports the server's own measured egress path MTU (0 = unknown). With DF set
 	// we cannot emit a datagram larger than this, so requested sizes above it are refused up
 	// front and reported as such — the client must not read that as a downstream path limit.
@@ -70,6 +83,10 @@ type Server struct {
 	CanaryQueries func(sessionPrefix string) any
 	// CanaryZone is surfaced in the profile so the app knows what to query.
 	CanaryZone string
+	// The addresses this server can be measured on. The "_alt" pair is the second address
+	// RFC 5780 behaviour discovery redirects to, and the one reserved from services so that
+	// nothing answering there is itself a measurement.
+	IP4, IP6, IP4Alt, IP6Alt string
 	// ProvenGood reports the server's self-test signal (may be nil). Surfaced
 	// in the profile so a client can trust — or skip — MTU tests: if the
 	// server's own egress isn't full-MTU, client MTU results measure the
@@ -83,6 +100,11 @@ type Server struct {
 	// AppRange is the app-version window this server will serve. Zero value means the built-in
 	// default (see DefaultAppRange).
 	AppRange compat.Range
+
+	// Spec §2.5 token buckets, keyed per credential and per source IP inside the limiter.
+	// Nil disables a ceiling (config value 0).
+	RateSessions *ratelimit.Limiter // POST /v1/sessions
+	RateActions  *ratelimit.Limiter // POST /v1/sessions/{id}/actions
 }
 
 // AppVersionHeader is how a client states its version. A client too old to send it is treated as
@@ -92,7 +114,12 @@ const AppVersionHeader = "X-Echolot-App-Version"
 
 // ProtocolVersion is the wire contract (probe-protocol.md) this build implements. It is what the
 // version window is really about; the release version is only a proxy for it.
-const ProtocolVersion = "1.0.0"
+//
+// 1.0.1: upstream trains (§3.2 types 0x03/0x04/0x05) and the action-id bytes at payload[8:16]
+// of granted packets. Patch, not minor: both are additive — a client that never sends
+// TRAIN_REPORT_REQ and never reads granted payloads (today's client reads only header fields)
+// sees no difference, so the fleet must not be split over it (§8.1).
+const ProtocolVersion = "1.0.1"
 
 // SchemaVersion is the measurement-document format this server can store.
 const SchemaVersion = "1.0.0"
@@ -150,17 +177,21 @@ func (s *Server) Handler() http.Handler {
 
 	gate := s.requireCompatibleApp
 	mux.HandleFunc("POST /v1/enroll", gate(s.enroll))
-	mux.HandleFunc("POST /v1/sessions", gate(s.newSession))
+	// §2.5 buckets sit on the two endpoints that make the server DO things — create state,
+	// send traffic. GET /v1/profile stays ungated on every axis (see above).
+	mux.HandleFunc("POST /v1/sessions", gate(s.rateLimited(s.RateSessions, s.newSession)))
 	mux.HandleFunc("DELETE /v1/sessions/{id}", gate(s.deleteSession))
 	mux.HandleFunc("GET /v1/sessions/{id}/observations", gate(s.observations))
-	mux.HandleFunc("POST /v1/sessions/{id}/actions", gate(s.actions))
+	mux.HandleFunc("POST /v1/sessions/{id}/actions", gate(s.rateLimited(s.RateActions, s.actions)))
 	mux.HandleFunc("POST /v1/echo", gate(s.httpEcho))
 	mux.HandleFunc("GET /v1/tls-reference", gate(s.tlsReference))
 	mux.HandleFunc("POST /v1/runs", gate(s.uploadRun))
 	mux.HandleFunc("GET /v1/runs", gate(s.listRuns))
 	mux.HandleFunc("GET /v1/runs/{id}", gate(s.getRun))
 	mux.HandleFunc("DELETE /v1/runs/{id}", gate(s.deleteRun))
-	// TODO(spec §5): frag_send, throughput (both build on the same grant machinery)
+	mux.HandleFunc("POST /v1/account/link", gate(s.linkAccount))
+	mux.HandleFunc("DELETE /v1/account/link", gate(s.unlinkAccount))
+	mux.HandleFunc("GET /v1/account", gate(s.accountStatus))
 	return mux
 }
 
@@ -182,6 +213,37 @@ func selftestSignal(f func() (bool, bool)) map[string]any {
 	return map[string]any{"mtu_ok": mtuOK, "sysctl_ok": sysctlOK}
 }
 
+// rateLimited enforces one §2.5 bucket policy on an endpoint: a token per credential AND one per
+// source IP. Two keys because each closes the other's hole — keyed only by credential, one
+// address cycles through credentials; keyed only by address, one credential rides many
+// addresses. The refusal is 429 with Retry-After, which is the whole point of a token bucket
+// over a hard drop here: a well-behaved client is told when to come back.
+func (s *Server) rateLimited(l *ratelimit.Limiter, next http.HandlerFunc) http.HandlerFunc {
+	return func(w http.ResponseWriter, r *http.Request) {
+		okCred, waitCred := l.Allow("cred:" + bearer(r))
+		okIP, waitIP := l.Allow("ip:" + remoteIP(r))
+		if !okCred || !okIP {
+			wait := max(waitCred, waitIP)
+			secs := int(wait/time.Second) + 1 // Retry-After is whole seconds, rounded up
+			w.Header().Set("Retry-After", strconv.Itoa(secs))
+			writeJSON(w, http.StatusTooManyRequests, map[string]any{
+				"error": "rate limited", "retry_after_s": secs,
+			})
+			return
+		}
+		next(w, r)
+	}
+}
+
+// remoteIP is the request's source address without the port, for rate-limit keys.
+func remoteIP(r *http.Request) string {
+	host, _, err := net.SplitHostPort(r.RemoteAddr)
+	if err != nil {
+		return r.RemoteAddr
+	}
+	return strings.Trim(host, "[]")
+}
+
 // sessionAuth resolves {id} and requires the bearer to be the owning device.
 func (s *Server) sessionAuth(w http.ResponseWriter, r *http.Request) *session.Session {
 	dev := s.Store.DeviceByCredential(bearer(r))
@@ -218,9 +280,19 @@ func (s *Server) observations(w http.ResponseWriter, r *http.Request) {
 		dnsCanary = s.CanaryQueries(sess.ID[:16]) // the session's wire prefix
 	}
 	writeJSON(w, http.StatusOK, map[string]any{
-		"udp":          map[string]any{"packets_seen": packetsSeen, "packets": udp},
+		"udp": map[string]any{
+			"packets_seen": packetsSeen,
+			"packets":      udp,
+			// The per-train received view (spec §6 "trains"), columnar like the wire report —
+			// the flat packet list above stays for clients that predate trains.
+			"trains": trainsJSON(sess.Trains()),
+		},
 		"tcp": tcp,
 		"connect_back": cb,
+		// The sender's own count, which is what makes the receiver's count mean something.
+		"throughput": sess.ThroughputReports(),
+		// The receiver's count for upstream runs — same idea, other direction.
+		"throughput_up": upstreamJSON(sess),
 		"dns_canary":    dnsCanary,
 		// TODO(spec §6): http echo records
 	})
@@ -243,9 +315,14 @@ func (s *Server) actions(w http.ResponseWriter, r *http.Request) {
 		SizeBytes  int    `json:"size_bytes"`
 		IntervalUs int    `json:"interval_us"`
 		SizesBytes []int  `json:"sizes_bytes"`
+		DSCP       *int   `json:"dscp"`
 		DF         *bool  `json:"df"`
 		Mode       string `json:"mode"`
 		FragBytes  int    `json:"frag_bytes"`
+		Direction  string `json:"direction"`
+		DurationS  int    `json:"duration_s"`
+		Kbps       int    `json:"kbps"`
+		Streams    int    `json:"streams"`
 	}
 	if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
 		writeJSON(w, http.StatusBadRequest, map[string]string{"error": "bad body"})
@@ -301,6 +378,11 @@ func (s *Server) actions(w http.ResponseWriter, r *http.Request) {
 			writeJSON(w, http.StatusNotImplemented, map[string]string{"error": "downtrain not wired"})
 			return
 		}
+		dscp, err := dscpArg(req.DSCP)
+		if err != nil {
+			writeJSON(w, http.StatusBadRequest, map[string]string{"error": err.Error()})
+			return
+		}
 		// A downstream train sends far more than it receives, so it needs a grant (§3.4).
 		count := clamp(req.Count, 1, 5000)
 		size := clamp(req.SizeBytes, dataMinPacket, 1500)
@@ -311,13 +393,20 @@ func (s *Server) actions(w http.ResponseWriter, r *http.Request) {
 			return
 		}
 		go func() {
-			sent, err := s.DownTrain(sess, g, count, size, interval)
+			sent, err := s.DownTrain(sess, g, count, size, interval, dscp)
 			slog.Info("downtrain finished", "action", actionID, "sent", sent, "bytes", g.Sent(), "err", err)
 		}()
-		writeJSON(w, http.StatusAccepted, map[string]any{
+		resp := map[string]any{
 			"action_id": actionID, "count": count, "size_bytes": size, "interval_us": interval,
 			"grant": map[string]any{"max_bytes": g.MaxBytes, "max_kbps": g.MaxKbps},
-		})
+		}
+		if dscp >= 0 {
+			resp["dscp"] = dscp
+			// Told up front, not discovered: a client measuring DSCP survival on a burst the
+			// server could not mark would conclude the network stripped it.
+			resp["dscp_applied"] = dataplane.TOSSupported
+		}
+		writeJSON(w, http.StatusAccepted, resp)
 
 	case "big_send":
 		if s.BigSend == nil {
@@ -417,6 +506,66 @@ func (s *Server) actions(w http.ResponseWriter, r *http.Request) {
 			"grant": map[string]any{"max_bytes": g.MaxBytes, "max_kbps": g.MaxKbps},
 		})
 
+	case "throughput":
+		if s.DownThroughput == nil {
+			writeJSON(w, http.StatusNotImplemented, map[string]string{"error": "throughput not wired"})
+			return
+		}
+		// Only the downstream direction needs the server to send. Upstream is the client
+		// sending and the server counting, which needs no action at all — so asking for it here
+		// is a client bug worth naming rather than silently doing the other thing.
+		if req.Direction == "up" {
+			// Upstream needs nothing sent from here — the client generates the traffic and the
+			// server counts it. The only thing an action can usefully do is zero the counter so
+			// the run measures itself rather than inheriting an earlier one.
+			sess.ResetUpstream()
+			writeJSON(w, http.StatusAccepted, map[string]any{
+				"action_id": actionID, "direction": "up", "reset": true,
+				"note": "send TYPE_THROUGHPUT_UP packets, then read observations.throughput_up",
+			})
+			return
+		}
+		if req.Direction != "" && req.Direction != "down" {
+			writeJSON(w, http.StatusBadRequest, map[string]string{
+				"error": "direction must be up or down",
+			})
+			return
+		}
+		// Planned once, here, so the response promises exactly what the run will do. A request
+		// that would outlast the server's byte cap comes back with a shorter duration rather
+		// than being truncated halfway.
+		durationMs, kbps := dataplane.ThroughputPlan(
+			clamp(req.DurationS, 1, 30)*1000, clamp(req.Kbps, 100, 200_000))
+		size := clamp(req.SizeBytes, dataMinPacket, 1472)
+		if req.SizeBytes == 0 {
+			size = 1200
+		}
+		g := sess.NewGrant(actionID, 0, kbps, dataplane.ThroughputLimits(durationMs, kbps))
+		if g == nil {
+			writeJSON(w, http.StatusConflict, noDataPlaneYet)
+			return
+		}
+		// Answered before the run so the client can start listening, then reported through the
+		// observations API. Doing it the other way round would have the client miss the first
+		// second of a ten-second test.
+		writeJSON(w, http.StatusAccepted, map[string]any{
+			"action_id": actionID, "direction": "down",
+			"duration_s": durationMs / 1000, "duration_ms": durationMs,
+			"requested_duration_s": clamp(req.DurationS, 1, 30),
+			"kbps":                 kbps, "size_bytes": size,
+			"grant": map[string]any{"max_bytes": g.MaxBytes, "max_kbps": g.MaxKbps},
+		})
+		if f, ok := w.(http.Flusher); ok {
+			f.Flush()
+		}
+		go func() {
+			result, err := s.DownThroughput(sess, g, durationMs, kbps, size)
+			slog.Info("throughput finished", "action", actionID, "packets", result.Packets,
+				"bytes", result.Bytes, "kbps", result.Kbps, "limited_by", result.LimitedBy, "err", err)
+			sess.RecordThroughput(actionID, result.Packets, result.Bytes, result.DurationMs,
+				result.Kbps, result.LimitedBy)
+		}()
+
 	default:
 		writeJSON(w, http.StatusBadRequest, map[string]string{"error": "unknown or unimplemented action"})
 	}
@@ -440,8 +589,54 @@ func (s *Server) maxDFPayload(sess *session.Session) int {
 	return mtu - overhead
 }
 
-// dataMinPacket is the smallest datagram that still carries a header + a little payload.
-const dataMinPacket = 40
+// dscpArg validates the optional downtrain dscp parameter (spec §5). Absent means -1: leave the
+// socket's default marking alone, which is different from asking for DSCP 0 (explicitly
+// best-effort). Out-of-range values are refused rather than clamped — a clamped 46→63 would mark
+// the burst with a class the client never asked for and silently change what the test measures.
+func dscpArg(v *int) (int, error) {
+	if v == nil {
+		return -1, nil
+	}
+	if *v < 0 || *v > 63 {
+		return 0, fmt.Errorf("dscp %d is out of range: the field is 6 bits (0..63)", *v)
+	}
+	return *v, nil
+}
+
+// trainsJSON renders the per-train received view columnar — one array per field, matching the
+// wire report and the schema's train-evidence shape — with []int for the byte-wide columns
+// because encoding/json would base64 a []uint8.
+func trainsJSON(trains []session.Train) []map[string]any {
+	out := make([]map[string]any, 0, len(trains))
+	for _, t := range trains {
+		n := len(t.Entries)
+		seq := make([]uint32, n)
+		trx := make([]int64, n)
+		size := make([]int, n)
+		ttl := make([]int, n)
+		dscp := make([]int, n)
+		ecn := make([]int, n)
+		for i, e := range t.Entries {
+			seq[i], trx[i], size[i] = e.Seq, e.TRxNs, int(e.Size)
+			ttl[i], dscp[i], ecn[i] = int(e.TTL), int(e.DSCP), int(e.ECN)
+		}
+		out = append(out, map[string]any{
+			"train_id": t.ID,
+			// The loss denominator: every packet counted, whether or not its row was kept.
+			"packets_received": t.Received,
+			"truncated":        t.Truncated,
+			"seq":              seq, "t_rx_ns": trx, "size": size,
+			"ttl": ttl, "dscp": dscp, "ecn": ecn,
+		})
+	}
+	return out
+}
+
+// dataMinPacket is the smallest granted datagram: header + 16 payload bytes, because [8:16] of
+// every granted payload carries the action id. It must match what the senders raise short sizes
+// to, or a minimum-size train's grant is budgeted for fewer bytes than actually leave and the
+// train is cut short by its own arithmetic.
+const dataMinPacket = dataplane.HeaderSize + 16
 
 var noDataPlaneYet = map[string]string{
 	"error": "no data-plane traffic seen yet — send an ECHO first so the destination is verified",
@@ -533,13 +728,7 @@ func (s *Server) profile(w http.ResponseWriter, r *http.Request) {
 		// modified builds and gives clients provenance for the measurement.
 		"source_url":   "", // TODO: stamp from build metadata
 		"capabilities": s.Capabilities,
-		"targets": []map[string]any{{
-			"id":        s.Name,
-			"ip4":       host, // TODO: explicit configured addresses, v6, second STUN addr
-			"udp_port":  s.UDPPort,
-			"tcp_port":  s.TCPPort,
-			"stun_port": s.StunPort,
-		}},
+		"targets": []map[string]any{s.target(host)},
 		"pins":            []string{"pin-sha256:" + s.PinB64},
 		"next_pins":       []string{},
 		"canary_zone":     s.CanaryZone,
@@ -548,6 +737,10 @@ func (s *Server) profile(w http.ResponseWriter, r *http.Request) {
 		// The app needs the upload rules before it offers the switch: whether uploads are
 		// accepted at all, and how much identifying detail it must strip first.
 		"uploads": s.uploadPolicy(),
+		// What a client needs to start a sign-in, without hard-coding the operator's IdP into
+		// the app: where to authorize, which client id to use, and whether it is worth offering
+		// sign-in at all on this server.
+		"auth": s.authInfo(r.Context()),
 		// What this build speaks, and which app versions it will serve. A client checks the
 		// server side of the same question against its own bounds.
 		"compat": map[string]any{
@@ -644,7 +837,7 @@ func (s *Server) uploadRun(w http.ResponseWriter, r *http.Request) {
 		writeJSON(w, http.StatusBadRequest, map[string]string{"error": "read failed"})
 		return
 	}
-	meta, err := s.Runs.Put(dev.ID, body)
+	meta, err := s.Runs.Put(dev.ID, body, dev.LinkedToAccount())
 	switch {
 	case err == nil:
 		slog.Info("run uploaded", "device", dev.ID, "run", meta.ID,
@@ -670,11 +863,19 @@ func (s *Server) listRuns(w http.ResponseWriter, r *http.Request) {
 		writeJSON(w, http.StatusUnauthorized, map[string]string{"error": "unknown credential"})
 		return
 	}
-	list := s.Runs.List(dev.ID)
+	list := s.Runs.ListFor(s.visibleDevices(dev))
 	if list == nil {
 		list = []runs.Meta{}
 	}
-	writeJSON(w, http.StatusOK, map[string]any{"runs": list})
+	writeJSON(w, http.StatusOK, map[string]any{
+		"runs": list,
+		// Says whose history this is, so a client can show "3 devices" rather than leaving the
+		// user to wonder why runs from another phone appeared.
+		"scope": map[string]any{
+			"account_id": dev.AccountID,
+			"devices":    len(s.visibleDevices(dev)),
+		},
+	})
 }
 
 func (s *Server) getRun(w http.ResponseWriter, r *http.Request) {
@@ -683,9 +884,14 @@ func (s *Server) getRun(w http.ResponseWriter, r *http.Request) {
 		writeJSON(w, http.StatusUnauthorized, map[string]string{"error": "unknown credential"})
 		return
 	}
-	// Scoped to the calling device's own directory: one device cannot read another's runs by
-	// guessing a run id.
-	b, err := s.Runs.Get(dev.ID, r.PathValue("id"))
+	// Resolved against the caller's own devices only, so a run id from another account is not
+	// found rather than being fetched from wherever it happens to live.
+	owner, ok := s.Runs.OwnerOf(s.visibleDevices(dev), r.PathValue("id"))
+	if !ok {
+		writeJSON(w, http.StatusNotFound, map[string]string{"error": "no such run"})
+		return
+	}
+	b, err := s.Runs.Get(owner, r.PathValue("id"))
 	if err != nil {
 		writeJSON(w, http.StatusNotFound, map[string]string{"error": "no such run"})
 		return
@@ -700,7 +906,12 @@ func (s *Server) deleteRun(w http.ResponseWriter, r *http.Request) {
 		writeJSON(w, http.StatusUnauthorized, map[string]string{"error": "unknown credential"})
 		return
 	}
-	if err := s.Runs.Delete(dev.ID, r.PathValue("id")); err != nil {
+	owner, ok := s.Runs.OwnerOf(s.visibleDevices(dev), r.PathValue("id"))
+	if !ok {
+		w.WriteHeader(http.StatusNoContent) // delete is idempotent; absent is the desired state
+		return
+	}
+	if err := s.Runs.Delete(owner, r.PathValue("id")); err != nil {
 		writeJSON(w, http.StatusInternalServerError, map[string]string{"error": err.Error()})
 		return
 	}
@@ -722,9 +933,172 @@ func maxOrEmpty(r compat.Range) string {
 // three parts at once — its own URL, its own SPKI pin, and the token. An operator copying a pin
 // by hand is the step that goes wrong, and a pin wrong by one character does not fail loudly.
 func (s *Server) EnrollmentLink(token string) string {
-	u := s.PublicControlURL
+	return EnrollmentURI(s.PublicControlURL, s.PinB64, token)
+}
+
+// EnrollmentURI is the same assembly without a running server, for the mint-a-link CLI action.
+// Shared rather than reimplemented: two copies of this encoding would eventually disagree, and
+// the failure mode is a pin that looks right and produces an inscrutable TLS error.
+func EnrollmentURI(publicURL, pinB64, token string) string {
 	return "echolot://enroll?v=1" +
-		"&u=" + url.QueryEscape(strings.TrimRight(u, "/")) +
-		"&p=" + url.QueryEscape("pin-sha256:"+s.PinB64) +
+		"&u=" + url.QueryEscape(strings.TrimRight(publicURL, "/")) +
+		"&p=" + url.QueryEscape("pin-sha256:"+pinB64) +
 		"&t=" + url.QueryEscape(token)
 }
+
+// target describes where this server can be measured, so a client can say which address a result
+// came from instead of "the server".
+//
+// The alternates matter as much as the primaries: RFC 5780 behaviour discovery needs a second
+// address to redirect to, and an operator reading a report needs to know which of their addresses
+// a finding refers to. [fallback] is used only when nothing was configured explicitly, so a server
+// that has not been told its own addresses still answers with something usable.
+func (s *Server) target(fallback string) map[string]any {
+	t := map[string]any{
+		"id":        s.Name,
+		"udp_port":  s.UDPPort,
+		"tcp_port":  s.TCPPort,
+		"stun_port": s.StunPort,
+	}
+	ip4 := s.IP4
+	if ip4 == "" {
+		ip4 = fallback
+	}
+	for k, v := range map[string]string{
+		"ip4": ip4, "ip6": s.IP6, "ip4_alt": s.IP4Alt, "ip6_alt": s.IP6Alt,
+	} {
+		if v != "" {
+			t[k] = v
+		}
+	}
+	return t
+}
+
+// upstreamJSON renders the upstream tally with the derived figures already computed, so every
+// consumer does not have to repeat (and risk fumbling) the same arithmetic.
+func upstreamJSON(sess *session.Session) map[string]any {
+	u := sess.Upstream()
+	return map[string]any{
+		"packets": u.Packets, "bytes": u.Bytes,
+		"span_ms": u.SpanMs(), "kbps": u.Kbps(),
+	}
+}
+
+// authInfo advertises the sign-in configuration, so the app can present a Sign in button only
+// when there is something behind it, and can drive the flow without the user typing an issuer URL.
+func (s *Server) authInfo(ctx context.Context) map[string]any {
+	if s.OIDC == nil || !s.OIDC.Config().Enabled() {
+		return map[string]any{"enabled": false}
+	}
+	cfg := s.OIDC.Config()
+	out := map[string]any{
+		"enabled": true,
+		"issuer":  cfg.Issuer,
+		// The app's client, not the server's: this is what a phone should authorize as.
+		"client_id": cfg.AppClientID,
+		// The app is a public client on a phone: no secret can be kept, so PKCE is what
+		// protects the code exchange (RFC 7636), and the redirect comes back through the
+		// scheme the app already registers for enrollment links.
+		"flow":         "authorization_code+pkce",
+		"redirect_uri": "echolot://auth",
+		"scopes":       "openid profile email",
+	}
+	if d, err := s.OIDC.Discover(ctx); err == nil {
+		out["authorization_endpoint"] = d.AuthorizationEndpoint
+		out["token_endpoint"] = d.TokenEndpoint
+		out["end_session_endpoint"] = d.EndSessionEndpoint
+	} else {
+		// Reported rather than hidden: an unreachable IdP is the operator's problem to see, and
+		// a client that knows the difference can say "sign-in is configured but the provider is
+		// not answering" instead of failing obscurely.
+		out["discovery_error"] = err.Error()
+	}
+	return out
+}
+
+// linkAccount ties the calling device to the person whose ID token it presents.
+//
+// The device credential proves *which device*; the ID token proves *which person*. Both are
+// required, and neither substitutes for the other: enrollment admits a device to the server,
+// signing in attributes it to someone.
+func (s *Server) linkAccount(w http.ResponseWriter, r *http.Request) {
+	dev := s.Store.DeviceByCredential(bearer(r))
+	if dev == nil {
+		writeJSON(w, http.StatusUnauthorized, map[string]string{"error": "unknown credential"})
+		return
+	}
+	if s.OIDC == nil || !s.OIDC.Config().Enabled() {
+		writeJSON(w, http.StatusNotImplemented, map[string]string{
+			"error": "this server has no identity provider configured, so there is nothing to sign in to",
+		})
+		return
+	}
+	var body struct {
+		IDToken string `json:"id_token"`
+	}
+	if err := json.NewDecoder(r.Body).Decode(&body); err != nil || body.IDToken == "" {
+		writeJSON(w, http.StatusBadRequest, map[string]string{"error": "expected an id_token"})
+		return
+	}
+	claims, err := s.OIDC.Verify(r.Context(), body.IDToken)
+	if err != nil {
+		// Deliberately terse to the client and detailed in the log: a caller probing token
+		// handling should not be told which check it failed.
+		slog.Info("rejected sign-in", "device", dev.ID, "err", err)
+		writeJSON(w, http.StatusForbidden, map[string]string{"error": "the identity token was not accepted"})
+		return
+	}
+	if err := s.Store.LinkAccount(dev.ID, claims.AccountID(), claims.Display()); err != nil {
+		writeJSON(w, http.StatusInternalServerError, map[string]string{"error": err.Error()})
+		return
+	}
+	slog.Info("device linked to account", "device", dev.ID, "account", claims.AccountID(),
+		"admin", s.OIDC.IsAdmin(claims))
+	writeJSON(w, http.StatusOK, map[string]any{
+		"account_id": claims.AccountID(), "display_name": claims.Display(),
+		"admin": s.OIDC.IsAdmin(claims),
+	})
+}
+
+// unlinkAccount signs out on this device. The device stays enrolled: signing out should not cost
+// someone their enrollment, which an operator had to grant.
+func (s *Server) unlinkAccount(w http.ResponseWriter, r *http.Request) {
+	dev := s.Store.DeviceByCredential(bearer(r))
+	if dev == nil {
+		writeJSON(w, http.StatusUnauthorized, map[string]string{"error": "unknown credential"})
+		return
+	}
+	if err := s.Store.LinkAccount(dev.ID, "", ""); err != nil {
+		writeJSON(w, http.StatusInternalServerError, map[string]string{"error": err.Error()})
+		return
+	}
+	w.WriteHeader(http.StatusNoContent)
+}
+
+func (s *Server) accountStatus(w http.ResponseWriter, r *http.Request) {
+	dev := s.Store.DeviceByCredential(bearer(r))
+	if dev == nil {
+		writeJSON(w, http.StatusUnauthorized, map[string]string{"error": "unknown credential"})
+		return
+	}
+	writeJSON(w, http.StatusOK, map[string]any{
+		"signed_in":    dev.LinkedToAccount(),
+		"account_id":   dev.AccountID,
+		"display_name": dev.AccountName,
+		"device_id":    dev.ID,
+	})
+}
+
+// visibleDevices is the set of devices whose runs the caller may read.
+//
+// Signed in: every device on the same account, which is what an account is for. Not signed in:
+// only itself — anonymous devices are not a group, and treating the absent account as a shared
+// one would let any of them read all the others.
+func (s *Server) visibleDevices(dev *store.Device) []string {
+	if dev.LinkedToAccount() {
+		if ids := s.Store.DeviceIDsForAccount(dev.AccountID); len(ids) > 0 {
+			return ids
+		}
+	}
+	return []string{dev.ID}
+}

server/internal/control/dscp_test.go

@@ -0,0 +1,27 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package control
+
+import "testing"
+
+func TestDscpArg(t *testing.T) {
+	ptr := func(v int) *int { return &v }
+	for _, tc := range []struct {
+		in      *int
+		want    int
+		wantErr bool
+	}{
+		{nil, -1, false},     // absent: leave the socket alone
+		{ptr(0), 0, false},   // explicit best-effort is not the same as absent
+		{ptr(46), 46, false}, // EF, the value people actually test with
+		{ptr(63), 63, false},
+		{ptr(64), 0, true}, // one past the 6-bit field
+		{ptr(-1), 0, true},
+	} {
+		got, err := dscpArg(tc.in)
+		if (err != nil) != tc.wantErr || got != tc.want {
+			t.Errorf("dscpArg(%v) = %d, err=%v; want %d, wantErr=%v", tc.in, got, err, tc.want, tc.wantErr)
+		}
+	}
+}

server/internal/dataplane/cmsg_linux.go

@@ -0,0 +1,70 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+//go:build linux
+
+package dataplane
+
+import (
+	"encoding/binary"
+	"net"
+	"syscall"
+)
+
+// Per-packet TTL and TOS/traffic-class arrive as control messages, and only if asked for at
+// socket setup. These fill the spec §3.3 observation-block fields that were shipped as the 0xFF
+// sentinel until now — received TTL is path-length evidence, the TOS byte is DSCP/ECN survival.
+
+// enableRecvMeta asks the kernel to attach the cmsgs to every received datagram. Both the v4 and
+// the v6 option sets are attempted on every socket: a dual-stack socket delivers v4-mapped
+// traffic through the v6 fd, and the kernel refuses whichever set does not apply. Errors are
+// dropped on purpose — a socket that cannot deliver metadata still serves probes, and the
+// sentinel already says "not observed" for it.
+func enableRecvMeta(conn *net.UDPConn) {
+	raw, err := conn.SyscallConn()
+	if err != nil {
+		return
+	}
+	_ = raw.Control(func(fd uintptr) {
+		_ = syscall.SetsockoptInt(int(fd), syscall.IPPROTO_IP, syscall.IP_RECVTTL, 1)
+		_ = syscall.SetsockoptInt(int(fd), syscall.IPPROTO_IP, syscall.IP_RECVTOS, 1)
+		_ = syscall.SetsockoptInt(int(fd), syscall.IPPROTO_IPV6, syscall.IPV6_RECVHOPLIMIT, 1)
+		_ = syscall.SetsockoptInt(int(fd), syscall.IPPROTO_IPV6, syscall.IPV6_RECVTCLASS, 1)
+	})
+}
+
+// parseMeta extracts TTL and the TOS byte from one datagram's control messages. Anything absent
+// or unparseable keeps the sentinel — reported as unobserved, never guessed.
+func parseMeta(oob []byte) pktMeta {
+	m := pktMeta{TTL: metaUnavailable, TOS: metaUnavailable}
+	if len(oob) == 0 {
+		return m
+	}
+	cmsgs, err := syscall.ParseSocketControlMessage(oob)
+	if err != nil {
+		return m
+	}
+	for _, c := range cmsgs {
+		switch {
+		case c.Header.Level == syscall.IPPROTO_IP && c.Header.Type == syscall.IP_TTL,
+			c.Header.Level == syscall.IPPROTO_IPV6 && c.Header.Type == syscall.IPV6_HOPLIMIT:
+			m.TTL = cmsgValue(c.Data)
+		case c.Header.Level == syscall.IPPROTO_IP && c.Header.Type == syscall.IP_TOS,
+			c.Header.Level == syscall.IPPROTO_IPV6 && c.Header.Type == syscall.IPV6_TCLASS:
+			m.TOS = cmsgValue(c.Data)
+		}
+	}
+	return m
+}
+
+// cmsgValue reads a cmsg the kernel encodes either as a native-endian int (IP_TTL,
+// IPV6_HOPLIMIT, IPV6_TCLASS) or as a single byte (IP_TOS). Both fit a byte by definition.
+func cmsgValue(data []byte) uint8 {
+	switch {
+	case len(data) >= 4:
+		return uint8(binary.NativeEndian.Uint32(data))
+	case len(data) >= 1:
+		return data[0]
+	}
+	return metaUnavailable
+}

server/internal/dataplane/cmsg_other.go

@@ -0,0 +1,15 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+//go:build !linux
+
+package dataplane
+
+import "net"
+
+// Per-packet TTL/TOS needs Linux's IP_RECVTTL-family cmsgs. Elsewhere the observation block and
+// train buffers keep the spec §3.3 sentinel (0xFF = not observed) — absent is honest, a guess
+// is not. Deployment targets are Linux; this build exists so the Windows dev loop compiles.
+func enableRecvMeta(_ *net.UDPConn) {}
+
+func parseMeta(_ []byte) pktMeta { return pktMeta{TTL: metaUnavailable, TOS: metaUnavailable} }

server/internal/dataplane/frag_linux.go

@@ -104,8 +104,8 @@ func (s *Server) FragSend(
 		return res, fmt.Errorf("session has no recorded local address")
 	}
 
-	if sizeBytes < HeaderSize+8 {
-		sizeBytes = HeaderSize + 8
+	if sizeBytes < HeaderSize+32 {
+		sizeBytes = HeaderSize + 32
 	}
 	if sizeBytes > 8000 {
 		sizeBytes = 8000
@@ -117,7 +117,10 @@ func (s *Server) FragSend(
 	// The ELT1 packet, signed exactly as any other, then wrapped in UDP.
 	payload := make([]byte, sizeBytes-HeaderSize)
 	binary.BigEndian.PutUint32(payload[0:4], uint32(sizeBytes))
-	copy(payload[4:], mode)
+	// [8:16]: the action id, as on every granted packet (spec §5 correlation); the mode string
+	// sits past the reserved slot.
+	putActionID(payload, g.ActionID)
+	copy(payload[16:], mode)
 	elt := s.buildPacket(sess, TypeFragData, 0, payload)
 
 	udp := buildUDP(local, target, elt)

server/internal/dataplane/granted.go

@@ -21,7 +21,11 @@ import (
 // client measures downstream loss, reordering and jitter from what arrives — the direction an
 // upstream-only train cannot see. Returns how many packets actually went out (the grant may cut
 // it short, which is itself reportable).
-func (s *Server) DownTrain(sess *session.Session, g *session.Grant, count, sizeBytes, intervalUs int) (int, error) {
+//
+// dscp ≥ 0 marks the burst (spec §5 downtrain `dscp`): downstream DSCP survival is the half the
+// client cannot produce itself. Best-effort off Linux — see withTOS/TOSSupported; the action
+// response has already told the client whether the marking was applied.
+func (s *Server) DownTrain(sess *session.Session, g *session.Grant, count, sizeBytes, intervalUs, dscp int) (int, error) {
 	target := sess.DataSource()
 	if !target.IsValid() {
 		return 0, fmt.Errorf("no observed data-plane source")
@@ -30,11 +34,15 @@ func (s *Server) DownTrain(sess *session.Session, g *session.Grant, count, sizeB
 	if conn == nil {
 		return 0, fmt.Errorf("no data-plane socket matches target family")
 	}
-	if sizeBytes < HeaderSize+8 {
-		sizeBytes = HeaderSize + 8
+	if sizeBytes < HeaderSize+16 {
+		sizeBytes = HeaderSize + 16
 	}
 	payload := make([]byte, sizeBytes-HeaderSize)
+	// Payload [8:16] carries the action id on every granted packet, so arriving traffic can be
+	// attributed to the action that caused it (spec §5/§9: test.params.action_id).
+	putActionID(payload, g.ActionID)
 	sent := 0
+	burst := func() error {
 		for i := 0; i < count; i++ {
 			if !g.Allow(sizeBytes) {
 				break // budget or rate exhausted — stop, do not sleep it off
@@ -49,7 +57,18 @@ func (s *Server) DownTrain(sess *session.Session, g *session.Grant, count, sizeB
 				time.Sleep(time.Duration(intervalUs) * time.Microsecond)
 			}
 		}
-	return sent, nil
+		return nil
+	}
+	if dscp >= 0 && TOSSupported {
+		// Same shared-socket borrow as the DF window: dfMu keeps a concurrent burst from riding
+		// along with — or clearing — this marking.
+		s.dfMu.Lock()
+		defer s.dfMu.Unlock()
+		err := withTOS(conn, dscp, burst) // sent must be read after the burst ran, not before
+		return sent, err
+	}
+	err := burst()
+	return sent, err
 }
 
 // BigSendResult records what happened to one requested size. `Sent` false with an EMSGSIZE-ish
@@ -83,8 +102,8 @@ func (s *Server) BigSend(sess *session.Session, g *session.Grant, sizes []int, d
 	results := make([]BigSendResult, 0, len(sizes))
 	burst := func() error {
 		for i, size := range sizes {
-			if size < HeaderSize+8 {
-				size = HeaderSize + 8
+			if size < HeaderSize+16 {
+				size = HeaderSize + 16
 			}
 			if size > 9000 { // jumbo ceiling; beyond this the kernel will refuse anyway
 				size = 9000
@@ -96,6 +115,8 @@ func (s *Server) BigSend(sess *session.Session, g *session.Grant, sizes []int, d
 			// Echo the intended size into the payload so a truncated/fragmented arrival is
 			// still attributable to the size we meant to send.
 			binary.BigEndian.PutUint32(payload[0:4], uint32(size))
+			// [8:16]: the action id, as on every granted packet (spec §5 correlation).
+			putActionID(payload, g.ActionID)
 			err := s.sendErr(conn, target, sess, TypeBigSend, uint32(i), payload)
 			results = append(results, BigSendResult{
 				SizeBytes: size, Seq: i, Sent: err == nil, Err: errString(err),

server/internal/dataplane/throughput.go

@@ -0,0 +1,191 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package dataplane
+
+import (
+	"encoding/binary"
+	"fmt"
+	"time"
+
+	"echo-lot.app/server/internal/session"
+)
+
+// Sustained-rate sending (spec §5 throughput).
+//
+// This is the most expensive thing the server will do on a client's say-so, so it is also the
+// action where the §3.4 anti-amplification rules matter most. Three bounds apply, and all three
+// are enforced here rather than trusted to the caller:
+//
+//   - the destination is the session's *observed* data-plane source, verified by an HMAC-signed
+//     ECHO that arrived from that address, so this cannot be aimed at a third party;
+//   - the grant carries a byte budget and an average-rate ceiling, and the send stops the moment
+//     either is reached;
+//   - the duration is hard-capped, so a client that vanishes mid-test costs a bounded amount of
+//     traffic rather than an open-ended one.
+//
+// The measurement this produces is honest only if the client is told which limit it hit. A run
+// that saturates the grant ceiling has measured *us*, not the network, and reporting that as
+// throughput would be worse than not measuring at all — see ThroughputResult.LimitedBy.
+
+// ThroughputResult is what the server actually managed to send.
+type ThroughputResult struct {
+	Packets    int   `json:"packets"`
+	Bytes      int64 `json:"bytes"`
+	DurationMs int64 `json:"duration_ms"`
+	Kbps       int   `json:"kbps"`
+	// LimitedBy says what stopped it: "duration" (ran the full time, so the rate is the path's
+	// or ours to give), "budget" (hit the grant's byte ceiling), or "rate" (the pacing ceiling
+	// held it back). Only "duration" makes the number a property of the network.
+	LimitedBy string `json:"limited_by"`
+}
+
+// ThroughputLimits derives a grant sized for one throughput run.
+//
+// The default 8 MiB action budget is deliberately far too small for this — ten seconds at
+// 50 Mbps is 62 MB — so throughput gets its own budget computed from what it asked for, still
+// clamped to a ceiling. Sizing the budget to the request (rather than raising the global default)
+// keeps every *other* action bounded at 8 MiB.
+func ThroughputLimits(durationMs, kbps int) session.GrantLimits {
+	durationMs, kbps = ThroughputPlan(durationMs, kbps)
+	// bytes = kbps * 1000 / 8 * seconds, with a little headroom so the byte budget is not what
+	// stops a run that was meant to be stopped by the clock.
+	budget := int64(kbps) * 1000 / 8 * int64(durationMs) / 1000
+	budget = budget * 11 / 10
+	if budget > maxThroughputBytes {
+		budget = maxThroughputBytes
+	}
+	return session.GrantLimits{
+		MaxBytes: budget,
+		// A little above the pacing target on purpose: the pacer should be what controls the
+		// rate, and the grant should be the safety net. If they are equal, ordinary scheduling
+		// jitter trips the grant and the run is cut short for no real reason.
+		MaxKbps: kbps * 12 / 10,
+		MaxHold: time.Duration(durationMs)*time.Millisecond + 5*time.Second,
+	}
+}
+
+// ThroughputPlan reduces a request to what this server will actually run, and is the single
+// place that decides it.
+//
+// When the byte cap binds before the clock does, the *duration* is shortened rather than the run
+// being cut off partway. Truncating mid-run is not wrong exactly — the rate is still computed
+// over the elapsed time and limited_by says "budget" — but it means promising a client thirty
+// seconds and giving it twenty-one. Saying "twenty-one seconds" up front is the same information
+// without the surprise, and it keeps "the clock ended the run" as the normal case, which is the
+// only case where the number is a clean property of the network.
+func ThroughputPlan(durationMs, kbps int) (effectiveMs, effectiveKbps int) {
+	if durationMs <= 0 {
+		durationMs = 10_000
+	}
+	if durationMs > maxThroughputMs {
+		durationMs = maxThroughputMs
+	}
+	if kbps <= 0 || kbps > maxThroughputKbps {
+		kbps = maxThroughputKbps
+	}
+	bytesPerMs := int64(kbps) * 1000 / 8 / 1000
+	if bytesPerMs > 0 {
+		if maxMs := maxThroughputBytes / bytesPerMs; int64(durationMs) > maxMs {
+			durationMs = int(maxMs)
+		}
+	}
+	return durationMs, kbps
+}
+
+const (
+	maxThroughputMs    = 30_000
+	maxThroughputKbps  = 200_000
+	maxThroughputBytes = 256 << 20
+)
+
+// DownThroughput sends paced traffic toward the client for up to durationMs.
+//
+// Pacing is deliberate rather than "send as fast as possible": an unpaced burst measures the
+// server's NIC and the first queue it meets, then collapses into loss that looks like a network
+// fault. Spacing packets at the target rate makes loss mean what a reader will assume it means.
+func (s *Server) DownThroughput(
+	sess *session.Session, g *session.Grant, durationMs, kbps, sizeBytes int,
+) (ThroughputResult, error) {
+	res := ThroughputResult{}
+
+	target := sess.DataSource()
+	if !target.IsValid() {
+		return res, fmt.Errorf("no observed data-plane source")
+	}
+	conn := s.connFor(target, sess.DataLocal())
+	if conn == nil {
+		return res, fmt.Errorf("no data-plane socket matches target family")
+	}
+
+	// Same plan the grant was sized from, so the two cannot disagree.
+	durationMs, kbps = ThroughputPlan(durationMs, kbps)
+	if sizeBytes < HeaderSize+24 {
+		sizeBytes = 1200 // a size that survives every common path unfragmented
+	}
+	if sizeBytes > 1472 {
+		sizeBytes = 1472
+	}
+
+	// Nanoseconds between packets to hit the target rate.
+	perPacketNs := int64(sizeBytes) * 8 * 1_000_000 / int64(kbps)
+	if perPacketNs < 1_000 {
+		perPacketNs = 1_000
+	}
+
+	payload := make([]byte, sizeBytes-HeaderSize)
+	// [8:16]: the action id, as on every granted packet (spec §5 correlation). The send
+	// timestamp lives past it at [16:24]; the client reads only header fields today, so
+	// reserving the slot costs nothing and keeps one layout rule across granted types.
+	putActionID(payload, g.ActionID)
+	deadline := time.Now().Add(time.Duration(durationMs) * time.Millisecond)
+	start := time.Now()
+	next := start
+
+	var seq uint32
+	for time.Now().Before(deadline) {
+		ok, why := g.TryAllow(sizeBytes)
+		if !ok {
+			if why == session.RefusalRate {
+				// Transient: the bucket is momentarily empty. Wait for the next slot and carry
+				// on. Ending the run here would report a rate measured over a fraction of a
+				// second, which is worse than reporting no rate at all.
+				res.LimitedBy = "rate"
+				time.Sleep(time.Duration(perPacketNs))
+				continue
+			}
+			// Terminal: the budget is spent, or the grant expired.
+			res.LimitedBy = why
+			break
+		}
+		// Reaching here means the run is progressing normally; the clock will end it.
+		res.LimitedBy = "duration"
+		binary.BigEndian.PutUint32(payload[0:4], seq)
+		binary.BigEndian.PutUint64(payload[16:24], uint64(time.Since(s.start).Nanoseconds()))
+		if err := s.sendErr(conn, target, sess, TypeThroughputData, seq, payload); err != nil {
+			// A send error mid-run is a local condition (buffer full, route gone). Stop and
+			// report what got out rather than pretending the rest was lost on the path.
+			res.LimitedBy = "send_error"
+			break
+		}
+		res.Packets++
+		res.Bytes += int64(sizeBytes)
+		seq++
+
+		// Absolute schedule, not sleep-per-packet: sleeping a fixed interval accumulates the
+		// scheduler's error and drifts the achieved rate below the target over a 10-second run.
+		next = next.Add(time.Duration(perPacketNs))
+		if d := time.Until(next); d > 0 {
+			time.Sleep(d)
+		}
+	}
+
+	elapsed := time.Since(start)
+	res.DurationMs = elapsed.Milliseconds()
+	// bits per millisecond is kilobits per second, so no scaling constant is needed - and none
+	// can be got wrong. Guarded because a run that ends inside a millisecond has no rate.
+	if res.DurationMs > 0 {
+		res.Kbps = int(res.Bytes * 8 / res.DurationMs)
+	}
+	return res, nil
+}

server/internal/dataplane/throughput_test.go

@@ -0,0 +1,99 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package dataplane
+
+import (
+	"testing"
+	"time"
+)
+
+// The grant has to be big enough that the *clock* ends a throughput run, not the byte budget. Get
+// this wrong and the test still "works": it stops early, reports a rate computed over a truncated
+// window, and nothing anywhere says the number is meaningless. So the sizing is pinned.
+func TestThroughputBudgetOutlastsTheRequestedRun(t *testing.T) {
+	cases := []struct{ durationMs, kbps int }{
+		{1_000, 1_000},
+		{10_000, 50_000},
+		{10_000, 200_000},
+		{30_000, 100_000},
+	}
+	for _, c := range cases {
+		// Against the *planned* duration, which is what will actually be run: a request the
+		// server shortens is answered with the shorter number, not truncated halfway.
+		planMs, planKbps := ThroughputPlan(c.durationMs, c.kbps)
+		lim := ThroughputLimits(c.durationMs, c.kbps)
+		needed := int64(planKbps) * 1000 / 8 * int64(planMs) / 1000
+		if lim.MaxBytes < needed {
+			t.Errorf("%d ms at %d kbps (planned %d ms) needs %d bytes, budget is %d - the run "+
+				"would stop early and report a rate over a truncated window",
+				c.durationMs, c.kbps, planMs, needed, lim.MaxBytes)
+		}
+	}
+}
+
+// The pacer should control the rate and the grant should be the safety net. If the grant's
+// ceiling equals the pacing target, ordinary scheduling jitter trips it and cuts the run short
+// for no real reason.
+func TestGrantRateCeilingSitsAboveThePacingTarget(t *testing.T) {
+	lim := ThroughputLimits(10_000, 50_000)
+	if lim.MaxKbps <= 50_000 {
+		t.Fatalf("grant ceiling %d kbps is not above the 50000 kbps pacing target", lim.MaxKbps)
+	}
+}
+
+// A client asking for more than the server will do must get the server's number, not its own.
+func TestThroughputRequestsAreClamped(t *testing.T) {
+	lim := ThroughputLimits(10*60*1000, 10_000_000) // ten minutes at 10 Gbps
+	if lim.MaxBytes > maxThroughputBytes {
+		t.Errorf("byte budget %d exceeds the hard cap %d", lim.MaxBytes, maxThroughputBytes)
+	}
+	if lim.MaxKbps > maxThroughputKbps*12/10 {
+		t.Errorf("rate ceiling %d exceeds the hard cap", lim.MaxKbps)
+	}
+	// The hold has to outlast the planned run, or the grant expires mid-send and the run is
+	// reported as rate-limited when it was really time-limited.
+	planMs, _ := ThroughputPlan(10*60*1000, 10_000_000)
+	if lim.MaxHold < time.Duration(planMs)*time.Millisecond {
+		t.Errorf("hold %v is shorter than the planned run of %d ms", lim.MaxHold, planMs)
+	}
+}
+
+// When the byte cap binds before the clock does, the server shortens the run and says so, rather
+// than accepting thirty seconds and delivering twenty-one. Same information, no surprise - and it
+// keeps "the clock ended the run" as the normal case, which is the only case where the resulting
+// rate is a clean property of the network.
+func TestAnOversizedRequestComesBackShorterRatherThanTruncated(t *testing.T) {
+	const kbps = 200_000
+	askedMs := 30_000
+	planMs, planKbps := ThroughputPlan(askedMs, kbps)
+
+	if planKbps != kbps {
+		t.Errorf("rate was reduced to %d; the duration should absorb the cap, not the rate", planKbps)
+	}
+	if planMs >= askedMs {
+		t.Fatalf("plan kept the full %d ms at %d kbps, which exceeds the %d byte cap",
+			askedMs, kbps, maxThroughputBytes)
+	}
+	// And what it does promise must fit.
+	if got := int64(planKbps) * 1000 / 8 * int64(planMs) / 1000; got > maxThroughputBytes {
+		t.Errorf("planned run needs %d bytes, over the %d cap", got, maxThroughputBytes)
+	}
+}
+
+// A short, ordinary request must come back untouched - the clamping only exists for the extremes.
+func TestAnOrdinaryRequestIsNotRewritten(t *testing.T) {
+	planMs, planKbps := ThroughputPlan(10_000, 50_000)
+	if planMs != 10_000 || planKbps != 50_000 {
+		t.Errorf("10 s at 50 Mbps was rewritten to %d ms at %d kbps", planMs, planKbps)
+	}
+}
+
+// Every action other than throughput stays on the small default budget. Throughput needs a big
+// one; raising the global default to suit it would quietly unbound everything else.
+func TestOnlyThroughputGetsTheLargeBudget(t *testing.T) {
+	big := ThroughputLimits(10_000, 50_000)
+	if big.MaxBytes <= 8<<20 {
+		t.Fatalf("throughput budget %d is no larger than the default action budget", big.MaxBytes)
+	}
+}

server/internal/dataplane/tos_linux.go

@@ -0,0 +1,55 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+//go:build linux
+
+package dataplane
+
+import (
+	"net"
+	"syscall"
+)
+
+// withTOS runs fn with the socket's TOS/traffic class set to dscp<<2 (ECN bits left zero — the
+// test is about DSCP survival, and claiming ECN capability we do not use would pollute it), then
+// restores what was there before.
+//
+// Same borrow discipline as withDF: the socket is shared by every session on that family, so the
+// caller must hold Server.dfMu for the whole window or a concurrent burst rides along with — or
+// clears — someone else's marking.
+func withTOS(conn *net.UDPConn, dscp int, fn func() error) error {
+	raw, err := conn.SyscallConn()
+	if err != nil {
+		return err
+	}
+	v4 := conn.LocalAddr().(*net.UDPAddr).IP.To4() != nil
+	level, opt := syscall.IPPROTO_IPV6, syscall.IPV6_TCLASS
+	if v4 {
+		level, opt = syscall.IPPROTO_IP, syscall.IP_TOS
+	}
+
+	var setErr error
+	prev := 0
+	if err := raw.Control(func(fd uintptr) {
+		if p, e := syscall.GetsockoptInt(int(fd), level, opt); e == nil {
+			prev = p
+		}
+		setErr = syscall.SetsockoptInt(int(fd), level, opt, dscp<<2)
+	}); err != nil {
+		return err
+	}
+	if setErr != nil {
+		return setErr
+	}
+	defer func() {
+		_ = raw.Control(func(fd uintptr) {
+			_ = syscall.SetsockoptInt(int(fd), level, opt, prev)
+		})
+	}()
+	return fn()
+}
+
+// TOSSupported reports whether withTOS can actually mark packets here. Exported so the control
+// plane can tell the client up front that its dscp request will not be honored, instead of the
+// client measuring an unmarked burst and concluding the network stripped the marking.
+const TOSSupported = true

server/internal/dataplane/tos_other.go

@@ -0,0 +1,16 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+//go:build !linux
+
+package dataplane
+
+import "net"
+
+// Setting DSCP per-burst uses IP_TOS/IPV6_TCLASS under the same fd-borrow pattern as withDF,
+// which is only exercised on Linux deployments. Elsewhere the burst goes out with the default
+// class and TOSSupported lets the action response say so — an unmarked burst reported as marked
+// would read as "the network stripped DSCP", the exact wrong conclusion.
+func withTOS(_ *net.UDPConn, _ int, fn func() error) error { return fn() }
+
+const TOSSupported = false

server/internal/dataplane/train.go

@@ -0,0 +1,122 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package dataplane
+
+// Upstream trains (spec §3.2, types 0x03–0x05). The client blasts TRAIN_DATA at the server and
+// the server answers nothing per-packet — a reply would double the traffic and measure the
+// return path at the same time. Afterwards the client asks for the server's received view with
+// TRAIN_REPORT_REQ, and gets it back columnar, split across as many TRAIN_REPORT datagrams as
+// it takes to stay under a safe size.
+//
+// Both TRAIN_DATA and TRAIN_REPORT_REQ carry the train id in payload[0:4]; the id is the
+// client's to choose, unique within the session.
+
+import (
+	"encoding/binary"
+	"net"
+	"net/netip"
+
+	"echo-lot.app/server/internal/session"
+)
+
+const (
+	// trainReportMaxDatagram caps one TRAIN_REPORT datagram at a size that survives every common
+	// path unfragmented. A report about loss must not itself be lost to MTU.
+	trainReportMaxDatagram = 1200
+	trainReportHeader      = 16
+	trainReportRow         = 17 // 4 seq + 8 t_rx_ns + 2 size + 1 ttl + 1 dscp + 1 ecn
+)
+
+// recordTrain buffers one TRAIN_DATA packet into its train (session-side, bounded — see
+// session/train.go). A payload too short to carry the id is unreportable and stays only in the
+// flat packet log, which already recorded it.
+func recordTrain(sess *session.Session, payload []byte, seq uint32, size int, tRxNs int64, meta pktMeta) {
+	if len(payload) < 4 {
+		return
+	}
+	sess.RecordTrainPacket(binary.BigEndian.Uint32(payload[0:4]), session.TrainEntry{
+		Seq: seq, TRxNs: tRxNs, Size: uint16(min(size, 0xFFFF)),
+		TTL: meta.TTL, DSCP: meta.dscp(), ECN: meta.ecn(),
+	})
+}
+
+// trainReport answers one TRAIN_REPORT_REQ with the full columnar report.
+//
+// Grant-free on purpose. §3.4 caps ungranted responses at the request size, and a multi-part
+// report is larger than the single REPORT_REQ that asked for it — but it cannot amplify: every
+// 17-byte row accounts for one HMAC-valid TRAIN_DATA packet of at least HeaderSize+4 bytes this
+// session already delivered here, so the whole report is a strict fraction of the traffic it
+// describes, and it only ever goes to the session's verified source address. An unknown id gets
+// a single zero-row report rather than silence — "nothing arrived" IS the measurement.
+func (s *Server) trainReport(conn *net.UDPConn, raddr netip.AddrPort, sess *session.Session, payload []byte) {
+	if len(payload) < 4 {
+		return
+	}
+	id := binary.BigEndian.Uint32(payload[0:4])
+	train, _ := sess.TrainView(id)
+	train.ID = id
+	for i, part := range buildTrainReport(train) {
+		s.send(conn, raddr, sess, TypeTrainReport, uint32(i), part)
+	}
+}
+
+// buildTrainReport lays a train's received view out columnar and cuts it into datagram-sized
+// payloads. Layout (mirrored by the client; all big-endian):
+//
+//	 0  4  train_id
+//	 4  4  received       (every packet counted, buffered or not — loss math uses this)
+//	 8  2  part           (0-based)
+//	10  2  parts
+//	12  1  flags          (bit0: buffer overflowed; rows beyond the cap were counted, not kept —
+//	                       the schema's evidence_truncated honesty, on the wire)
+//	13  1  reserved
+//	14  2  n              (rows in this part)
+//	16     n×4 seq, n×8 t_rx_ns, n×2 size, n×1 ttl, n×1 dscp, n×1 ecn (columns contiguous)
+func buildTrainReport(t session.Train) [][]byte {
+	perPart := (trainReportMaxDatagram - HeaderSize - trainReportHeader) / trainReportRow
+	parts := (len(t.Entries) + perPart - 1) / perPart
+	if parts == 0 {
+		parts = 1 // an empty train still gets its "received: 0" answer
+	}
+	out := make([][]byte, 0, parts)
+	for p := 0; p < parts; p++ {
+		rows := t.Entries[p*perPart : min((p+1)*perPart, len(t.Entries))]
+		n := len(rows)
+		b := make([]byte, trainReportHeader+n*trainReportRow)
+		binary.BigEndian.PutUint32(b[0:4], t.ID)
+		binary.BigEndian.PutUint32(b[4:8], uint32(t.Received))
+		binary.BigEndian.PutUint16(b[8:10], uint16(p))
+		binary.BigEndian.PutUint16(b[10:12], uint16(parts))
+		if t.Truncated {
+			b[12] = 1
+		}
+		binary.BigEndian.PutUint16(b[14:16], uint16(n))
+		off := trainReportHeader
+		for i, r := range rows {
+			binary.BigEndian.PutUint32(b[off+i*4:], r.Seq)
+		}
+		off += n * 4
+		for i, r := range rows {
+			binary.BigEndian.PutUint64(b[off+i*8:], uint64(r.TRxNs))
+		}
+		off += n * 8
+		for i, r := range rows {
+			binary.BigEndian.PutUint16(b[off+i*2:], r.Size)
+		}
+		off += n * 2
+		for i, r := range rows {
+			b[off+i] = r.TTL
+		}
+		off += n
+		for i, r := range rows {
+			b[off+i] = r.DSCP
+		}
+		off += n
+		for i, r := range rows {
+			b[off+i] = r.ECN
+		}
+		out = append(out, b)
+	}
+	return out
+}

server/internal/dataplane/train_test.go

@@ -0,0 +1,179 @@
+// SPDX-FileCopyrightText: 2026 Echolot contributors
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+package dataplane
+
+import (
+	"encoding/binary"
+	"net"
+	"net/netip"
+	"testing"
+	"time"
+
+	"echo-lot.app/server/internal/session"
+)
+
+// parseReportPart decodes one TRAIN_REPORT payload back into rows, checking the header.
+func parseReportPart(t *testing.T, b []byte) (id uint32, received int, part, parts int, truncated bool, rows []session.TrainEntry) {
+	t.Helper()
+	if len(b) < trainReportHeader {
+		t.Fatalf("report part shorter than its header: %d", len(b))
+	}
+	id = binary.BigEndian.Uint32(b[0:4])
+	received = int(binary.BigEndian.Uint32(b[4:8]))
+	part = int(binary.BigEndian.Uint16(b[8:10]))
+	parts = int(binary.BigEndian.Uint16(b[10:12]))
+	truncated = b[12]&1 != 0
+	n := int(binary.BigEndian.Uint16(b[14:16]))
+	if want := trainReportHeader + n*trainReportRow; len(b) != want {
+		t.Fatalf("part length %d, want %d for %d rows", len(b), want, n)
+	}
+	off := trainReportHeader
+	rows = make([]session.TrainEntry, n)
+	for i := range rows {
+		rows[i].Seq = binary.BigEndian.Uint32(b[off+i*4:])
+	}
+	off += n * 4
+	for i := range rows {
+		rows[i].TRxNs = int64(binary.BigEndian.Uint64(b[off+i*8:]))
+	}
+	off += n * 8
+	for i := range rows {
+		rows[i].Size = binary.BigEndian.Uint16(b[off+i*2:])
+	}
+	off += n * 2
+	for i := range rows {
+		rows[i].TTL = b[off+i]
+	}
+	off += n
+	for i := range rows {
+		rows[i].DSCP = b[off+i]
+	}
+	off += n
+	for i := range rows {
+		rows[i].ECN = b[off+i]
+	}
+	return
+}
+
+func TestBuildTrainReportSplitsAndRoundTrips(t *testing.T) {
+	const count = 250 // enough to need several parts
+	train := session.Train{ID: 42, Received: count, Truncated: true}
+	for i := 0; i < count; i++ {
+		train.Entries = append(train.Entries, session.TrainEntry{
+			Seq: uint32(i), TRxNs: int64(i) * 1_000_000, Size: uint16(100 + i),
+			TTL: 64, DSCP: 46, ECN: 1,
+		})
+	}
+
+	parts := buildTrainReport(train)
+	if len(parts) < 2 {
+		t.Fatalf("250 rows should not fit one ≤%d-byte datagram", trainReportMaxDatagram)
+	}
+	var got []session.TrainEntry
+	for i, p := range parts {
+		if HeaderSize+len(p) > trainReportMaxDatagram {
+			t.Fatalf("part %d would be a %d-byte datagram, cap is %d", i, HeaderSize+len(p), trainReportMaxDatagram)
+		}
+		id, received, part, total, truncated, rows := parseReportPart(t, p)
+		if id != 42 || received != count || part != i || total != len(parts) || !truncated {
+			t.Fatalf("part %d header: id=%d received=%d part=%d/%d truncated=%v",
+				i, id, received, part, total, truncated)
+		}
+		got = append(got, rows...)
+	}
+	if len(got) != count {
+		t.Fatalf("round-tripped %d rows, want %d", len(got), count)
+	}
+	for i, r := range got {
+		want := train.Entries[i]
+		if r != want {
+			t.Fatalf("row %d = %+v, want %+v", i, r, want)
+		}
+	}
+}
+
+func TestBuildTrainReportEmptyTrainStillAnswers(t *testing.T) {
+	parts := buildTrainReport(session.Train{ID: 9})
+	if len(parts) != 1 {
+		t.Fatalf("empty train: %d parts, want 1 — 'nothing arrived' is the answer, not silence", len(parts))
+	}
+	id, received, _, total, _, rows := parseReportPart(t, parts[0])
+	if id != 9 || received != 0 || total != 1 || len(rows) != 0 {
+		t.Fatalf("empty report: id=%d received=%d parts=%d rows=%d", id, received, total, len(rows))
+	}
+}
+
+func TestTrainDataThenReportOverTheWire(t *testing.T) {
+	mgr, addr := startServer(t)
+	sess, _, err := mgr.New("dev1", "credential-ikm", netip.MustParseAddr("127.0.0.1"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	client, err := net.DialUDP("udp", nil, net.UDPAddrFromAddrPort(addr))
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer client.Close()
+	client.SetDeadline(time.Now().Add(3 * time.Second))
+
+	// A short train: id 5 in payload[0:4], plus padding.
+	const trainID, count = 5, 4
+	for i := 0; i < count; i++ {
+		payload := make([]byte, 60)
+		binary.BigEndian.PutUint32(payload[0:4], trainID)
+		if _, err := client.Write(craft(t, sess, TypeTrainData, uint32(i+1), payload)); err != nil {
+			t.Fatal(err)
+		}
+	}
+	// TRAIN_DATA must be silent (spec §3.2).
+	client.SetReadDeadline(time.Now().Add(300 * time.Millisecond))
+	if _, err := client.Read(make([]byte, 1500)); err == nil {
+		t.Fatal("TRAIN_DATA got a response, want none")
+	}
+
+	// Ask for the report.
+	reqPayload := make([]byte, 4)
+	binary.BigEndian.PutUint32(reqPayload, trainID)
+	client.SetDeadline(time.Now().Add(3 * time.Second))
+	if _, err := client.Write(craft(t, sess, TypeTrainReportReq, 100, reqPayload)); err != nil {
+		t.Fatal(err)
+	}
+	buf := make([]byte, 2000)
+	n, err := client.Read(buf)
+	if err != nil {
+		t.Fatalf("no TRAIN_REPORT: %v", err)
+	}
+	if buf[4] != TypeTrainReport {
+		t.Fatalf("type = %#x, want TRAIN_REPORT", buf[4])
+	}
+	id, received, part, parts, truncated, rows := parseReportPart(t, buf[HeaderSize:n])
+	if id != trainID || received != count || part != 0 || parts != 1 || truncated {
+		t.Fatalf("report header: id=%d received=%d part=%d/%d truncated=%v", id, received, part, parts, truncated)
+	}
+	if len(rows) != count {
+		t.Fatalf("%d rows, want %d", len(rows), count)
+	}
+	for i, r := range rows {
+		if r.Seq != uint32(i+1) {
+			t.Fatalf("row %d seq = %d, want %d", i, r.Seq, i+1)
+		}
+		if r.Size != HeaderSize+60 {
+			t.Fatalf("row %d size = %d, want %d", i, r.Size, HeaderSize+60)
+		}
+	}
+
+	// Unknown train id: one zero-row report, not silence.
+	binary.BigEndian.PutUint32(reqPayload, 999)
+	if _, err := client.Write(craft(t, sess, TypeTrainReportReq, 101, reqPayload)); err != nil {
+		t.Fatal(err)
+	}
+	n, err = client.Read(buf)
+	if err != nil {
+		t.Fatalf("no report for unknown train: %v", err)
+	}
+	id, received, _, _, _, rows = parseReportPart(t, buf[HeaderSize:n])
+	if id != 999 || received != 0 || len(rows) != 0 {
+		t.Fatalf("unknown-train report: id=%d received=%d rows=%d", id, received, len(rows))
+	}
+}

server/internal/dataplane/udp.go

@@ -2,15 +2,15 @@
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // Package dataplane implements the binary UDP probe protocol (spec §3):
-// 32-byte header, HMAC gate, anti-replay, ECHO with observation block.
-// Skeleton scope: ECHO_REQ/ECHO_RESP and TIMESYNC only; trains, MTU probes
-// and delayed echo land with the corresponding client tests.
+// 32-byte header, HMAC gate, anti-replay, ECHO with observation block,
+// upstream trains with columnar reports, and the granted server->client sends.
 package dataplane
 
 import (
 	"crypto/hmac"
 	"crypto/sha256"
 	"encoding/binary"
+	"encoding/hex"
 	"fmt"
 	"log/slog"
 	"net"
@@ -18,6 +18,7 @@ import (
 	"sync"
 	"time"
 
+	"echo-lot.app/server/internal/ratelimit"
 	"echo-lot.app/server/internal/session"
 )
 
@@ -27,6 +28,11 @@ const (
 
 	TypeEchoReq  = 0x01
 	TypeEchoResp = 0x02
+	// Upstream trains (spec §3.2): DATA gets no per-packet response; REPORT_REQ fetches the
+	// server's received view as one or more REPORT datagrams (train.go).
+	TypeTrainData      = 0x03
+	TypeTrainReportReq = 0x04
+	TypeTrainReport    = 0x05
 	TypeTimesyncReq    = 0x07
 	TypeTimesyncRsp = 0x08
 	TypeMtuProbe    = 0x09
@@ -37,10 +43,22 @@ const (
 	TypeBigSend       = 0x0C
 	// TypeFragData is delivered only after IP reassembly, so its arrival IS the measurement.
 	TypeFragData = 0x0D
+	// TypeThroughputData is one packet of a sustained-rate downstream run.
+	TypeThroughputData = 0x0E
+	// TypeThroughputUp is one packet of a client-driven upstream run. The server counts it and
+	// deliberately does not answer: a reply would double the traffic and measure the return
+	// path at the same time, which is the one thing this test is trying not to do.
+	TypeThroughputUp = 0x0F
 )
 
 type Server struct {
 	Sessions *session.Manager
+	// Spec §2.5 ceilings on verified traffic, silent-drop (nil = no ceiling). Charged after the
+	// HMAC gate so an unauthenticated flood cannot spend anyone's budget, keyed per source
+	// address AND per device credential so neither one hot address nor one hot credential can
+	// crowd out the rest.
+	PacketRate *ratelimit.Limiter // tokens are packets
+	ByteRate   *ratelimit.Limiter // tokens are bytes
 	// Epoch for server-side t_rx/t_tx: process start; observation consumers
 	// only need differences plus the timesync exchange, not absolute time.
 	start time.Time
@@ -53,6 +71,34 @@ type Server struct {
 	dfMu sync.Mutex
 }
 
+// pktMeta is what the kernel told us about one received datagram beyond its bytes (spec §3.3:
+// received TTL, DSCP, ECN). 0xFF means "not observed": non-Linux hosts and datagrams whose
+// cmsg never arrived keep the sentinel rather than inventing a value.
+type pktMeta struct {
+	TTL uint8
+	TOS uint8 // the whole DSCP/ECN byte; DSCP = TOS>>2, ECN = TOS&3
+}
+
+const metaUnavailable = 0xFF
+
+func (m pktMeta) dscp() uint8 {
+	if m.TOS == metaUnavailable {
+		return metaUnavailable
+	}
+	return m.TOS >> 2
+}
+
+func (m pktMeta) ecn() uint8 {
+	if m.TOS == metaUnavailable {
+		return metaUnavailable
+	}
+	return m.TOS & 0x3
+}
+
+// oobCap fits the two cmsgs (TTL + TOS, each ≤ CMSG_SPACE(4)) with headroom for whatever else
+// the kernel decides to attach.
+const oobCap = 64
+
 // Serve runs the read loop for one socket; call once per bound address.
 // The socket is retained so actions (delayed echo) can pick a family-matching
 // sender later.
@@ -63,14 +109,16 @@ func (s *Server) Serve(conn *net.UDPConn) error {
 	}
 	s.conns = append(s.conns, conn)
 	s.mu.Unlock()
+	enableRecvMeta(conn)
 	buf := make([]byte, 65535)
+	oob := make([]byte, oobCap)
 	for {
-		n, raddr, err := conn.ReadFromUDPAddrPort(buf)
+		n, oobn, _, raddr, err := conn.ReadMsgUDPAddrPort(buf, oob)
 		if err != nil {
 			return err
 		}
 		tRx := time.Since(s.start).Nanoseconds()
-		s.handle(conn, raddr, buf[:n], tRx)
+		s.handle(conn, raddr, buf[:n], tRx, parseMeta(oob[:oobn]))
 	}
 }
 
@@ -120,7 +168,7 @@ func (s *Server) SendDelayedEcho(sess *session.Session, actionID string) error {
 
 // handle enforces spec §3.1/§3.4: unknown prefix, bad HMAC, expired session,
 // replayed seq → silent drop, never a response.
-func (s *Server) handle(conn *net.UDPConn, raddr netip.AddrPort, pkt []byte, tRxNs int64) {
+func (s *Server) handle(conn *net.UDPConn, raddr netip.AddrPort, pkt []byte, tRxNs int64, meta pktMeta) {
 	if len(pkt) < HeaderSize || string(pkt[0:4]) != Magic {
 		return
 	}
@@ -143,6 +191,12 @@ func (s *Server) handle(conn *net.UDPConn, raddr netip.AddrPort, pkt []byte, tRx
 	if !hmac.Equal(mac.Sum(nil)[:4], pkt[28:32]) {
 		return
 	}
+	// Spec §2.5: over-ceiling traffic is silently dropped (probes tolerate loss by design).
+	// After the HMAC gate so a spoofed flood cannot drain a victim's budget; before the replay
+	// window so a dropped packet's seq stays usable for a resend.
+	if !s.allowUDP(raddr, sess.Device, len(pkt)) {
+		return
+	}
 	if !sess.CheckSeq(seq) {
 		return
 	}
@@ -150,14 +204,29 @@ func (s *Server) handle(conn *net.UDPConn, raddr netip.AddrPort, pkt []byte, tRx
 	if la, ok := conn.LocalAddr().(*net.UDPAddr); ok {
 		sess.NoteDataLocal(la.AddrPort())
 	}
+
+	// Upstream throughput short-circuits before the observation log. Recording one struct per
+	// packet here would mean tens of thousands of allocations for a single run; the counter is
+	// all anyone needs, since the client holds the send-side record.
+	if typ == TypeThroughputUp {
+		sess.CountUpstream(len(pkt), tRxNs)
+		return
+	}
 	sess.RecordUDP(session.UDPObservation{
 		Seq: seq, TRxNs: tRxNs, TTxNs: time.Since(s.start).Nanoseconds(),
 		Src: raddr.String(), Size: len(pkt), Type: typ,
 	})
 
+	payload := pkt[HeaderSize : HeaderSize+int(payloadLen)]
 	switch typ {
 	case TypeEchoReq:
-		s.echoResp(conn, raddr, sess, pkt, seq, tRxNs)
+		s.echoResp(conn, raddr, sess, pkt, seq, tRxNs, meta)
+	case TypeTrainData:
+		// No response (spec §3.2): the train is upstream-only; its received view is fetched
+		// afterwards via TRAIN_REPORT_REQ or the observations API.
+		recordTrain(sess, payload, seq, len(pkt), tRxNs, meta)
+	case TypeTrainReportReq:
+		s.trainReport(conn, raddr, sess, payload)
 	case TypeTimesyncReq:
 		s.timesyncResp(conn, raddr, sess, pkt, seq, tRxNs)
 	case TypeMtuProbe:
@@ -167,6 +236,16 @@ func (s *Server) handle(conn *net.UDPConn, raddr netip.AddrPort, pkt []byte, tRx
 	}
 }
 
+// allowUDP charges the §2.5 packet and byte buckets, per source address and per credential.
+func (s *Server) allowUDP(raddr netip.AddrPort, device string, size int) bool {
+	ipKey, credKey := "ip:"+raddr.Addr().String(), "cred:"+device
+	okA, _ := s.PacketRate.Allow(ipKey)
+	okC, _ := s.PacketRate.Allow(credKey)
+	okAB, _ := s.ByteRate.AllowN(ipKey, float64(size))
+	okCB, _ := s.ByteRate.AllowN(credKey, float64(size))
+	return okA && okC && okAB && okCB
+}
+
 // mtuAck replies to an MTU_PROBE with a small MTU_ACK carrying the total
 // datagram size the server actually received (spec §3.2). The client sends
 // DF-flagged probes of increasing size and binary-searches the path MTU / a
@@ -184,28 +263,43 @@ func (s *Server) mtuAck(conn *net.UDPConn, raddr netip.AddrPort, sess *session.S
 //	 8  8  t_tx_ns
 //	16 16  observed source IP (v4-mapped when v4)
 //	32  2  observed source port
-//	34  1  received TTL   (0xFF = not observed yet; needs recvmsg cmsgs)
+//	34  1  received TTL   (0xFF = not observed; cmsgs unavailable on this host)
 //	35  1  received DSCP/ECN byte (0xFF = not observed)
 //	36  4  received size
-func observation(tRxNs, tTxNs int64, src netip.AddrPort, rcvd int) []byte {
+func observation(tRxNs, tTxNs int64, src netip.AddrPort, rcvd int, meta pktMeta) []byte {
 	b := make([]byte, 40)
 	binary.BigEndian.PutUint64(b[0:8], uint64(tRxNs))
 	binary.BigEndian.PutUint64(b[8:16], uint64(tTxNs))
 	a16 := src.Addr().As16()
 	copy(b[16:32], a16[:])
 	binary.BigEndian.PutUint16(b[32:34], src.Port())
-	b[34], b[35] = 0xFF, 0xFF
+	b[34], b[35] = meta.TTL, meta.TOS
 	binary.BigEndian.PutUint32(b[36:40], uint32(rcvd))
 	return b
 }
 
+// putActionID writes a grant's action id into payload[8:16] — the correlation the spec promises
+// (§5: "an action_id echoed in resulting data-plane packets"), consumed by the client as
+// test.params.action_id (§9). Bytes [0:8] stay with the packet type; [8:16] is reserved for this
+// across every granted type, so the client needs one rule, not five.
+func putActionID(payload []byte, actionID string) {
+	if len(payload) < 16 {
+		return
+	}
+	raw, err := hex.DecodeString(actionID)
+	if err != nil || len(raw) != 8 {
+		return // a malformed id yields zero bytes, not a crash mid-burst
+	}
+	copy(payload[8:16], raw)
+}
+
 // echoResp mirrors the request header (type flipped), appends the observation
 // block, and re-HMACs with the session key. Anti-amplification: the response
 // is capped at the request size (spec §3.4) — the observation block replaces
 // padding rather than growing the datagram; if the request was smaller than
 // header+observation, the block is truncated to fit.
-func (s *Server) echoResp(conn *net.UDPConn, raddr netip.AddrPort, sess *session.Session, req []byte, seq uint32, tRxNs int64) {
-	obs := observation(tRxNs, time.Since(s.start).Nanoseconds(), raddr, len(req))
+func (s *Server) echoResp(conn *net.UDPConn, raddr netip.AddrPort, sess *session.Session, req []byte, seq uint32, tRxNs int64, meta pktMeta) {
+	obs := observation(tRxNs, time.Since(s.start).Nanoseconds(), raddr, len(req), meta)
 	max := len(req)
 	if max < HeaderSize {
 		return