docs(review): row 45 round 1 review record, changes (filbert)

Co-Authored-By: Claude Opus 5.5 <[email protected]>
2026-10-09 08:32:38 -05:00
parent 6e4af4b7c2
commit 46baf3f854
53 changed files with 3755 additions and 0 deletions
@@ -0,0 +1,13 @@
 #!/usr/bin/env bash
 # Sequential gate; $1 = tree, $2 = out prefix. Each output teed to a file.
 T="$1"; P="$2"; O=~/filbert-scratch/r45/out; cd "$T"
 for d in packages/*/tests; do
  p=$(basename "$(dirname "$d")")
  env -u NODE_TEST_CONTEXT DOCKER_HOST=unix:///nonexistent-filbert-r45.sock timeout 900 node --test "packages/$p/tests/*.test.mjs" 2>&1 | tee "$O/$P-node-$p.txt" > /dev/null
  echo "$P node-$p exit ${PIPESTATUS[0]} $(grep -E '^# (pass|fail|cancelled)' "$O/$P-node-$p.txt" | tr '\n' ' ')"
 done
 for s in scripts/test-*.sh; do
  n=$(basename "$s" .sh); n=${n#test-}
  env -u NODE_TEST_CONTEXT DOCKER_HOST=unix:///nonexistent-filbert-r45.sock timeout 900 bash "$s" 2>&1 | tee "$O/$P-suite-$n.txt" > /dev/null
  echo "$P suite-$n exit ${PIPESTATUS[0]} load $(cut -d" " -f1 /proc/loadavg)"
 done
@@ -0,0 +1,57 @@
 #!/usr/bin/env bash
 # Row 45 round 1 mutants (Filbert): Rocko's 18 (agents/rocko/work/s4-follow-up/mutants.sh)
 # and X1-X14. One perl substitution each, restored after its run. A run counts
 # as killed when any test fails or is cancelled, or the outer timeout fires.
 # Usage: mutants.sh <tree> <outdir>
 set -u
 T="$1"; O="$2"; cd "$T"
 run() {
  local id="$1" file="$2" expr="$3"
  cp "$file" "$file.orig"
  perl -0pi -e "$expr" "$file"
  if cmp -s "$file" "$file.orig"; then echo "$id NOT-APPLIED $file"; mv "$file.orig" "$file"; return; fi
  env -u NODE_TEST_CONTEXT DOCKER_HOST=unix:///nonexistent-filbert-r45.sock timeout 900 node --test --test-timeout=90000 'packages/cli/tests/*.test.mjs' 'packages/discord/tests/*.test.mjs' > "$O/mut-$id.txt" 2>&1
  local rc=$? f c
  f=$(grep -E '^ℹ fail ' "$O/mut-$id.txt" | awk '{print $3}')
  c=$(grep -E '^ℹ cancelled ' "$O/mut-$id.txt" | awk '{print $3}')
  if [ "$rc" = 124 ]; then echo "$id killed (hang, outer timeout)"
  elif [ "${f:-?}" = 0 ] && [ "${c:-?}" = 0 ]; then echo "$id SURVIVED"
  else echo "$id killed (fail ${f:-?}, cancelled ${c:-?})"; fi
  mv "$file.orig" "$file"
 }
 NT=packages/cli/src/notifier.mjs
 HS=packages/cli/src/host.mjs
 # Rocko's set, verbatim.
 run N2   $NT 's/lstatSync, mkdirSync/lstatSync, statSync, mkdirSync/; s/const ds = lstatSync\(dir\);/const ds = statSync(dir);/'
 run N4   $NT 's/O_WRONLY \| O_APPEND \| O_NOFOLLOW\)/O_WRONLY | O_APPEND)/'
 run N5   $HS 's/if \(child\.exitCode !== null \|\| child\.signalCode !== null\) onDeath/if (child.exitCode !== null) onDeath/'
 run M28  $HS 's/  if \(prior\?\.live\) throw new CliError\([^\n]*\n//'
 run G150 $HS 's/(if \(ok\?\.ok !== true\) \{\n\s*await ended\(notify, 5000\);\n\s*)if \(broker\.connected\) /$1/'
 run G144 $HS 's/(notify\.kill\("SIGTERM"\);\n\s*await ended\(notify, 5000\);\n\s*)if \(broker\.connected\) /$1/'
 run F2a  $NT 's/export const REFUSAL_LIMIT = 5;/export const REFUSAL_LIMIT = 6;/'
 run F2b  $NT 's/ && r\.status !== 429;/;/'
 run F2c  $NT 's/ \|\| journal\.gaveUp\.has\(d\.id\)\) continue;/) continue;/'
 run F2d  $NT 's/if \(\(journal\.refusals\.get\(d\.id\) \?\? 0\) >= REFUSAL_LIMIT\) \{/if (false) {/'
 run F2e  $NT 's/ : dmGaveUp\(d\.id\) \? "\[blocking, DM refused, not retried\] "//'
 run F2f  $NT 's/      if \(record\.kind === "dm" && \(journal\.refusals\.get\(record\.decision\) \?\? 0\) >= REFUSAL_LIMIT\) \{/      if (false) {/'
 run J4a  $NT 's/ \|\| new Date\(r\.at\)\.toISOString\(\) !== r\.at\) return "at";/) return "at";/'
 run J4b  $NT 's/typeof r\.decision !== "string" \|\| r\.decision === ""/false/'
 run J4c  $NT 's/if \(r\.status !== undefined && !Number\.isInteger\(r\.status\)\) return "status";//'
 run J4d  $NT 's/r\.outcome === "confirmed" \? typeof r\.messageId !== "string" : //'
 run E1   $NT 's/accessSync\(dir, constants\.W_OK \| constants\.X_OK\);//'
 run E2   $NT 's/  if \(ds\.isSymbolicLink\(\)\) throw[^\n]*\n//'
 # Filbert's set.
 run X1   $NT 's/r\.outcome === "refused" && Number\.isInteger\(r\.status\) && r\.status >= 400 && r\.status < 500 && r\.status !== 429;/r.outcome === "refused";/'
 run X2   $NT 's/r\.status >= 400 && r\.status < 500 &&/r.status >= 400 \&\& r.status < 600 \&\&/'
 run X3   $NT 's/      if \(bad\) throw new CliError\(`notify journal line[^\n]*\n      count\(r\);/$&\n      if (definite(r)) refusals.delete(r.decision);/'
 run X4   $NT 's/r\.kind === "digest" \? typeof r\.day !== "string" \|\| !DAY\.test\(r\.day\) :/r.kind === "digest" ? false :/'
 run X5   $NT 's/ \|\| \(r\.outcome === "gave-up" && r\.kind !== "dm"\)\) return "outcome";/) return "outcome";/'
 run X6   $NT 's/    if \(UNWRITABLE\.includes\(e\.code\)\) throw new CliError\(`notify journal directory cannot be created[^\n]*\n//'
 run X7   $NT 's/ : r\.decision !== null && r\.decision !== undefined\) return "decision";/ : false) return "decision";/'
 run X8   $NT 's/    if \(e\.code === "ELOOP"\) throw new CliError\(`notify journal must not be a symlink[^\n]*\n//'
 run X9   $NT 's/    if \(UNWRITABLE\.includes\(e\.code\)\) throw new CliError\(`notify journal is not writable[^\n]*\n//'
 run X10  $NT 's/: r\.messageId !== null && typeof r\.messageId !== "string"\) return "messageId";/: false) return "messageId";/'
 run X11  $NT 's/if \(!\["dm", "digest"\]\.includes\(r\.kind\)\) return "kind";//'
 run X12  $NT 's/if \(r\.kind === "dm" && r\.outcome === "gave-up"\) gaveUp\.add\(r\.decision\);/if (false) gaveUp.add(r.decision);/'
 run X13  $NT 's/(function giveUp\(decision, t\) \{\n\s*)journal\.append\(/$1if (0) journal.append(/'
 run X14  scripts/bus-service.sh 's/^[^\n]*mkdir -m 0700 -p[^\n]*\n//m'
@@ -0,0 +1,109 @@
 // Row 45 probes (Filbert). Run: node --test probe.test.mjs, with TREE set.
 import { test } from "node:test";
 import assert from "node:assert/strict";
 import { appendFileSync, readFileSync } from "node:fs";
 const T = process.env.TREE;
 const { createNotifier, journalPath, openJournal, POLL_MS } = await import(`${T}/packages/cli/src/notifier.mjs`);
 const { RestOutcome } = await import(`${T}/packages/discord/src/rest.mjs`);
 const { broker, tmp } = await import(`${T}/packages/cli/tests/helpers.mjs`);
 const log = (...a) => console.log("PROBE", ...a);
 function rig(t, iso, answer) {
  const bus = broker(t);
  const dataRoot = tmp(t);
  const clock = { t: new Date(iso) };
  const sends = [];
  const logs = [];
  const direct = { async send(m) { sends.push({ at: clock.t.toISOString(), nonce: m.nonce }); return answer(clock.t); } };
  const make = () => createNotifier({ business: "demo", dataRoot, inbox: async () => bus.read("inbox"), direct, now: () => clock.t, log: (l) => logs.push(l) });
  const journal = () => readFileSync(journalPath(dataRoot, "demo"), "utf8").trim().split("\n").filter(Boolean).map((l) => JSON.parse(l));
  return { ...bus, dataRoot, clock, sends, logs, make, journal };
 }
 const refuse403 = () => { throw new RestOutcome("refused", "dm: refused", { status: 403 }); };
 // F2 timing: a binding typo (every DM 403s), polled every POLL_MS.
 test("F2-T1 time from first refusal to gave-up, polling every POLL_MS", async (t) => {
  const s = rig(t, "2026-10-08T05:00:00Z", refuse403);
  s.raise("git.push.protected", { target: "refactor", blocking: true, task_ref: "vikunja:1/7" });
  const n = s.make();
  const start = s.clock.t.getTime();
  for (let i = 0; i < 2000 && !s.journal().some((r) => r.outcome === "gave-up"); i++) {
    await n.tick();
    s.clock.t = new Date(s.clock.t.getTime() + POLL_MS);
  }
  const gave = s.journal().find((r) => r.outcome === "gave-up");
  log("T1 sends at", s.sends.map((x) => `+${(Date.parse(x.at) - start) / 1000}s`).join(" "));
  log("T1 gave-up at", `+${(Date.parse(gave.at) - start) / 1000}s`, "=", ((Date.parse(gave.at) - start) / 60000).toFixed(1), "min");
 });
 // The typo is fixed an hour after the first refusal: does the DM land?
 test("F2-T2 binding fixed after 60 min: the DM never lands", async (t) => {
  const t0 = Date.parse("2026-10-08T05:00:00Z");
  const s = rig(t, "2026-10-08T05:00:00Z", (now) => (now.getTime() - t0 < 60 * 60_000 ? refuse403() : { messageId: "123456789012345678" }));
  s.raise("git.push.protected", { target: "refactor", blocking: true, task_ref: "vikunja:1/7" });
  const n = s.make();
  for (let i = 0; i < 4 * 120; i++) {
    await n.tick();
    s.clock.t = new Date(s.clock.t.getTime() + POLL_MS);
  }
  log("T2 after 4 h: outcomes", s.journal().filter((r) => r.kind === "dm").map((r) => r.outcome).join(","));
  log("T2 sends", s.sends.filter((x) => x.nonce.startsWith("dm")).length);
 });
 // Restart after two refusals: the in-memory backoff is lost.
 test("F2-T3 restart drops the backoff: next attempt is immediate", async (t) => {
  const s = rig(t, "2026-10-08T05:00:00Z", refuse403);
  s.raise("git.push.protected", { target: "refactor", blocking: true, task_ref: "vikunja:1/7" });
  let n = s.make();
  await n.tick();
  s.clock.t = new Date(s.clock.t.getTime() + 31_000);
  await n.tick();
  const before = s.sends.length;
  n = s.make();
  await n.tick();
  log("T3 sends before restart", before, "after one tick post-restart", s.sends.length, "(same clock second)");
 });
 // Unknown outcomes and 429 never give up.
 test("F2-T4 500s for 6 h: no gave-up", async (t) => {
  const s = rig(t, "2026-10-08T05:00:00Z", () => { throw new RestOutcome("unknown", "dm: unknown", { status: 500 }); });
  s.raise("git.push.protected", { target: "refactor", blocking: true, task_ref: "vikunja:1/7" });
  const n = s.make();
  for (let i = 0; i < 6 * 120; i++) { await n.tick(); s.clock.t = new Date(s.clock.t.getTime() + POLL_MS); }
  log("T4 dm sends", s.sends.filter((x) => x.nonce.startsWith("dm")).length, "gave-up", s.journal().some((r) => r.outcome === "gave-up"));
 });
 // A digest 403 forever: the digest has no limit (only DMs give up).
 test("F2-T5 digest refusals: retried, never gave-up", async (t) => {
  const s = rig(t, "2026-10-08T13:00:00Z", refuse403);
  const n = s.make();
  for (let i = 0; i < 120; i++) { await n.tick(); s.clock.t = new Date(s.clock.t.getTime() + POLL_MS); }
  log("T5 digest sends in 1 h", s.sends.length, "gave-up", s.journal().some((r) => r.outcome === "gave-up"));
 });
 // J4 on append: a non-string messageId is written, and the next open refuses.
 test("J4-A1 append does not type-check: a numeric messageId bricks the next open", async (t) => {
  const s = rig(t, "2026-10-08T05:00:00Z", () => ({ messageId: 123 }));
  s.raise("git.push.protected", { target: "refactor", blocking: true, task_ref: "vikunja:1/7" });
  const r = await s.make().tick();
  log("A1 tick", JSON.stringify(r), "line", JSON.stringify(s.journal()[0]));
  try { s.make(); log("A1 reopen ok"); } catch (e) { log("A1 reopen", e.exitCode, e.message.replace(s.dataRoot, "<root>")); }
 });
 // J4: a journal written by the base (row 39) code still opens.
 test("J4-A2 a line the row 39 writer produced still opens", (t) => {
  const s = rig(t, "2026-10-08T05:00:00Z", refuse403);
  const file = journalPath(s.dataRoot, "demo");
  openJournal(file);
  const at = new Date("2026-10-08T05:00:00Z").toISOString();
  appendFileSync(file, [
    { at, kind: "dm", decision: "d1", outcome: "confirmed", messageId: "1" },
    { at, kind: "dm", decision: "d2", outcome: "refused", messageId: null, status: 403 },
    { at, kind: "dm", decision: "d3", outcome: "unknown", messageId: null },
    { at, kind: "digest", decision: null, day: "2026-10-08", outcome: "confirmed", messageId: "2" },
    { at, kind: "digest", decision: null, day: "2026-10-09", outcome: "refused", messageId: null, status: 429 },
  ].map((r) => `${JSON.stringify(r)}\n`).join(""));
  const j = openJournal(file);
  log("A2 ok sent", [...j.sent].join(","), "days", [...j.days].join(","), "refusals", JSON.stringify([...j.refusals]));
 });
@@ -0,0 +1,22 @@
 // Filbert, row 45: the host.mjs:144/150 window. The broker is SIGKILLed and
 // dead, but this process has not yet run the disconnect, so connected is
 // still true. Does send() with no callback raise an unhandled 'error'?
 // argv[2]: "nocb" | "cb"
 import { fork, spawnSync } from "node:child_process";
 import { once } from "node:events";
 import { fileURLToPath } from "node:url";
 if (process.argv[2] === "child") { process.send("up"); setInterval(() => {}, 1000); }
 else {
  const mode = process.argv[2];
  const c = fork(fileURLToPath(import.meta.url), ["child"], { stdio: ["ignore", "inherit", "inherit", "ipc"] });
  await once(c, "message");
  process.on("exit", (code) => console.log(mode, "process exit code", code));
  c.kill("SIGKILL");
  spawnSync("sleep", ["0.3"]); // block the loop: the child is dead, the disconnect not yet seen
  console.log(mode, "connected before send", c.connected);
  if (c.connected) {
    if (mode === "cb") c.send({ op: "close" }, (e) => console.log(mode, "callback got", e?.code ?? "no error"));
    else c.send({ op: "close" });
  }
  process.exitCode = 3;
 }
@@ -0,0 +1,75 @@
 ✔ launch identity is stamped, payload identity is refused and stale holder cannot send (24.241496ms)
 ✔ decision classes route from policy; gated resolution is human-only, choice and target must match (27.401376ms)
 ✔ claim exclusion, holder release, gated revoke and rerouting to a new holder are atomic (27.377079ms)
 ✔ launch events require a human CLI capability; generic emit cannot forge authority events (18.540838ms)
 ✔ within-role decisions close atomically and invalid options or blocking omissions refuse (15.718237ms)
 ✔ observer capabilities read human inbox but cannot mutate or forge launch identity (17.673868ms)
 ✔ task action subjects and linked decision trail are complete and ordered (26.350248ms)
 ✔ launch binding is durable and reconnecting requires the identical trusted record (12.074309ms)
 ✔ business isolation includes inherited object names and cross-business message references (17.714459ms)
 ✔ authority never transfers between action, run, target, unresolved or replaced role holder (23.654992ms)
 ✔ task projection uses schema current view, skipping earlier and equal-start polls (12.601793ms)
 ✔ revocation permanently bars the old run from reclaiming first, including after broker restart (21.489231ms)
 ✔ empty message references refuse before storage; refusal-evidence failure stays a typed error (12.121438ms)
 ✔ both arbiters require human resolution when their cross-role route is themselves (18.100019ms)
 ✔ S1 adapter takes resolved limits and refs, rejects mismatched instance, never mutates input (2.663501ms)
 ✔ only validated broker references load; returned data and exceptions cannot expose a known token (7.494723ms)
 ✔ bad file modes, symlinks, repository/data paths, malformed tokens and missing dates refuse (2.614846ms)
 ✔ expiry refuses use and env references never become client data (0.953331ms)
 ✔ S1 parsed service refs work, service mismatch refuses, Gitea rotation due is a warning state (1.997646ms)
 ✔ opaque tokens shorter than 16 characters refuse before use (0.426004ms)
 ✔ human proof binds CLI entry, process start and nonce; agents and incomplete ancestry refuse (2.639842ms)
 ✔ process reader gets own kernel identity without exposing environment values (0.796532ms)
 ✔ EACCES ancestor environments skip only markers; commands and registered launches still refuse (1.31124ms)
 ✔ real pid 1 remains inspectable when its environment is protected (0.492702ms)
 ✔ within-role sends cite an open gated launch decision without spending it or naming it in grants (30.750778ms)
 ✔ missing and foreign-business citations refuse and roll back message and grant (19.639241ms)
 ✔ cross-role sends still need a matching resolved decision and consume it once (25.365852ms)
 ✔ broker process binds trusted launches, offers reader capabilities, refuses human mutation, closes cleanly (90.577896ms)
 ✔ startup token refusal returns safe code without value or partial listening broker (57.911085ms)
 ✔ loaded fixture token is absent from socket replies and SQLite, including refusal evidence (81.062244ms)
 ✔ killed broker leaves an explicit stale lock; another process cannot silently reclaim it (93.951236ms)
 ✔ trusted host registers later launches; socket clients never have a registration verb (54.687669ms)
 ✔ runtime excludes declared project roots even when host supplies no repoRoots (37.187253ms)
 ✔ a refused launch binding leaves the broker and existing capabilities alive; bad protocol stops it (57.471264ms)
 ✔ v3b prototype refusals, views and append-only mutations (85.698753ms)
 ✔ gated approval authorizes once, survives store reopen, and fresh approval works (32.966804ms)
 ✔ another run cannot consume an approval; a failed check leaves it usable (23.112882ms)
 ✔ two scheduled callers have exactly one grant and one consumed refusal (18.143623ms)
 ✔ failed commit rolls consumption back; cross-role consumes and within-role stays reusable (30.935331ms)
 ✔ class drift gated to cross-role refuses before consumption (22.429967ms)
 ✔ class drift cross-role to gated refuses before consumption (23.526473ms)
 ✔ class drift gated to within-role refuses before consumption (18.94584ms)
 ✔ class drift cross-role to within-role refuses before consumption (20.00439ms)
 ✔ class drift within-role to gated refuses before consumption (20.09949ms)
 ✔ class drift within-role to cross-role refuses before consumption (20.004313ms)
 ✔ message.send consumes approval and prevents a later send or authorize (20.507454ms)
 ✔ role.revoke consumes approval and prevents a later revoke or authorize (24.840891ms)
 ✔ creates private WAL store and excludes a second writer until explicit close (17.620181ms)
 ✔ rollback is atomic and schema metadata is checked against trusted DDL, not just itself (16.372243ms)
 ✔ existing empty database and symlink runtime directory refuse, never initialize over damage (17.06366ms)
 ✔ crash during a transaction recovers no partial event after explicit fixture-only lock removal (61.104182ms)
 ✔ writer refuses mixed at/read_at forms atomically, even through trusted SQL helpers (11.117121ms)
 ✔ async transactions refuse before invoking their function (8.582549ms)
 ✔ recordTask keeps sync reads and a role write apart (26.654451ms)
 ✔ read_at must be one canonical UTC format, so the projection compares strings safely (10.565667ms)
 ✔ a bad entry refuses the whole record (10.346049ms)
 ✔ taskView reads the projection for one business (14.743855ms)
 ✔ requestTask hands only a holder and a task verb to the handler, and records refusals (23.223528ms)
 ✔ the server sends task verbs to the adapter with its own timeout; other verbs stay synchronous (278.8713ms)
 ✔ without an adapter the server refuses every task verb (14.399987ms)
 ✔ the runtime refuses an invalid adapter and closes a valid one (16.003453ms)
 ✔ the process loads the S3 adapter from plain-data trackers (104.50676ms)
 ✔ socket capability stamps launch identity; shared views use wire, no SQL client (30.89905ms)
 ✔ two wire claims serialize; a lost reply never automatically retries (25.599399ms)
 ✔ malformed, oversized and identity-forging envelopes refuse without echoing input (13.8714ms)
 ✔ client preserves UTF-8 when a response divides a multibyte character (12.484692ms)
 ✔ committed mutation followed by dropped reply reports unknown and is never retried (16.628757ms)
 ℹ tests 67
 ℹ suites 0
 ℹ pass 67
 ℹ fail 0
 ℹ cancelled 0
 ℹ skipped 0
 ℹ todo 0
 ℹ duration_ms 620.262652
@@ -0,0 +1,68 @@
 ✔ config directory and file path follow MOSAIC_CONFIG (1.867049ms)
 ✔ the fixture business validates and comes back frozen (6.613216ms)
 ✔ two instances may share a definition (1.553281ms)
 ✔ top-level refusals (4.394638ms)
 ✔ arbiters and projects (6.806464ms)
 ✔ role instances (4.007157ms)
 ✔ Vikunja bots (10.644466ms)
 ✔ a role without Vikunja takes no tracker block (3.567703ms)
 ✔ credential references match the definition's services (4.264474ms)
 ✔ launch (12.958131ms)
 ✔ loadBusiness: file checks (2.079948ms)
 ✔ loadBusiness: not a regular file (45.475837ms)
 ✔ loading writes nothing (1.248191ms)
 ✔ names that are Object.prototype properties don't count as declared (2.636431ms)
 ✔ the shipped example refuses as written and validates once filled in (0.938997ms)
 ✔ usage errors exit 4 (306.920279ms)
 ✔ validate: a good business exits 0 and prints instance digests (75.941648ms)
 ✔ validate: project files (372.761359ms)
 ✔ validate: missing files and a broken system config (286.961712ms)
 ✔ validate: credential reference problems exit 2 and name each one (92.078302ms)
 ✔ validate: a token file inside the repository is refused (69.525005ms)
 ✔ validate: role definitions come from MOSAIC_ROLES_DIR (212.064917ms)
 ✔ resolve: prints one instance's record (225.62473ms)
 ✔ resolve: refusals (438.939536ms)
 ✔ parse: exactly one of file or env, plus the service's date (2.990356ms)
 ✔ check: a good file has no problems (0.88057ms)
 ✔ check never opens the file: a write-only token passes (0.344506ms)
 ✔ check: file problems (1.133557ms)
 ✔ check: token files can't live in the repository or dataRoot, even through a linked directory (1.180256ms)
 ✔ check: dates and environment references (0.53199ms)
 ✔ path and load (2.600044ms)
 ✔ refusals (1.590197ms)
 ✔ systemVars flattens the validated config (2.334227ms)
 ✔ precedence: system, business, project, project role, agent (5.382138ms)
 ✔ limits narrow the definition and never widen it (2.738644ms)
 ✔ role.launch stays within-role only for the instance the launch block names (5.289806ms)
 ✔ limits.authority without role.launch leaves the launcher with no launch block (2.998882ms)
 ✔ limits.authority narrows cross-role actions too (1.66732ms)
 ✔ classify (1.855641ms)
 ✔ the record carries what the broker and launcher need (1.587725ms)
 ✔ digest: key order doesn't matter, any value change does (10.439567ms)
 ✔ refusals (3.815951ms)
 ✔ the four shipped version 2 roles load (3.112119ms)
 ✔ shipped role scopes match addendum B section 2 and the SR runbook (1.576915ms)
 ✔ shipped authority follows the note's table (0.5797ms)
 ✔ version 1 files keep loading with no authority (1.255039ms)
 ✔ the conductor policy isn't a role (0.287754ms)
 ✔ a missing role file is exit 4, a symbolic link too (0.465952ms)
 ✔ version 2 refusals (1.508511ms)
 ✔ authority: closed vocabulary, no gated-only action, no overlap (2.698625ms)
 ✔ credentials: Gitea scopes (0.935475ms)
 ✔ credentials: Vikunja scopes are a group-to-verbs map from the grantable list (1.273969ms)
 ✔ credentials: services (0.668012ms)
 ✔ contract: a non-empty regular Markdown file beside the role file (0.725598ms)
 ✔ every key names known layers and a merge rule (0.99936ms)
 ✔ unknown keys and wrong layers refuse (0.880079ms)
 ✔ types (2.195618ms)
 ✔ merge: defaults, then the most specific layer wins (0.322584ms)
 ✔ merge: limits only narrow, and provenance lists each source (0.431596ms)
 ✔ merge doesn't change its inputs (0.188001ms)
 ℹ tests 60
 ℹ suites 0
 ℹ pass 60
 ℹ fail 0
 ℹ cancelled 0
 ℹ skipped 0
 ℹ todo 0
 ℹ duration_ms 2158.961409
@@ -0,0 +1,57 @@
 ✔ inbox lists only decisions routed to the human, with what approving authorizes and how to decide (18.683142ms)
 ✔ decide resolves by id prefix with --yes, and the trail shows it in broker order with the task to follow (23.175532ms)
 ✔ decide refuses without a terminal or --yes, on an unknown option and on a short reference (12.54802ms)
 ✔ decide prints a declining choice as declining (12.654365ms)
 ✔ an unknown outcome is reported once and never resent (12.795995ms)
 ✔ a decision closed before the answer arrives exits 2 and points at its trail (10.791338ms)
 ✔ a prefix that matches two open decisions exits 2 and resolves neither (12.657588ms)
 ✔ without --business a command uses the live host's business, and a stale host.json is not a host (10.224266ms)
 ✔ every human command refuses inside an agent run before it touches the bus (11.932399ms)
 ✔ usage errors exit 4; no business and no host is a usage error (10.850137ms)
 ✔ agents and tasks print through the broker (9.820295ms)
 ✔ notify.json: missing, loose, malformed and extra keys refuse; a binding or null passes (1.459087ms)
 ✔ bootConfig builds the broker's boot message for one business, with no trackers key when no project names one (47.077328ms)
 ✔ trackers come from the tracker.* variables of the one project that names a tracker project (41.209336ms)
 ✔ with two projects, the one that sets tracker.project supplies the tracker and the other is no conflict (36.360521ms)
 ✔ two projects that each name a tracker project refuse, since the boot shape holds one (37.183043ms)
 ✔ a business without tracker.baseUrl gets no trackers entry (34.767644ms)
 ✔ an unknown business and a broken system config refuse with exit 3 (65.444872ms)
 ✔ empty views say so (0.866ms)
 ✔ the trail keeps the broker's order and names a decision's task without its rows (1.026474ms)
 ✔ tasks print the tracker fields the snapshot carries (0.169169ms)
 ✔ the host boots the broker, binds a launch in process, and the notifier DMs a blocking decision exactly once (868.369289ms)
 ✔ a notifier that dies takes the host down with exit 1, so the unit restarts the pair (172.365188ms)
 ✔ a notifier that refuses stops the broker and the host refuses with exit 3 (147.095184ms)
 ✔ watchChildren reports a child that died before it was called, and one that dies later (31.864675ms)
 ✔ bus stop refuses to signal a live pid that is not a bus host (203.54711ms)
 ✔ bus start refuses with exit 3 and the code when the broker refuses to boot; bus status names the lock (257.814577ms)
 ✔ bus start refuses with exit 3 without a notifier config (111.480249ms)
 ✔ bus start runs until bus stop; status reports it while it runs (729.54278ms)
 ✔ bus-service.sh renders the unit and installs it into a given directory (33.560404ms)
 ✔ zoned uses the IANA zone across DST (25.853917ms)
 ✔ each open blocking decision is DM'd once, across polls and a restart; non-blocking ones are not (18.485699ms)
 ✔ two blocking decisions get two DMs with different nonces (12.527143ms)
 ✔ the digest nonce differs per business and per day and fits Discord's 25 characters (0.235022ms)
 ✔ a failed DM is journaled, backs off, and is retried until it lands (13.460085ms)
 ✔ the digest goes at 08:00 Chicago once a day, with blocking ones marked as DM'd (16.722078ms)
 ✔ a late start with no digest for the day sends one at once; an empty inbox gets one line (12.203833ms)
 ✔ an inbox read failure is logged and the next poll retries (0.835125ms)
 ✔ no Discord id reaches the journal or the log (13.241518ms)
 ✔ the journal: a torn tail is copied out and truncated, so an append after it reopens cleanly (2.243073ms)
 ✔ the journal: a crash between the copy and the truncate leaves a tail the next open repairs (3.159201ms)
 ✔ the journal: a whole file that is one torn line truncates to empty (1.998131ms)
 ✔ the journal: a malformed complete line refuses and leaves the file and any torn tail alone (0.793662ms)
 ✔ the journal: a loose file mode, a loose directory or a symlinked journal refuses (0.743756ms)
 ✔ digest content stays within Discord's 2000 characters (0.392904ms)
 ✔ runLoop never overlaps ticks and stops after the one in flight (110.519055ms)
 ✔ the transport writes {business, verb, args} to the child and reads its JSON (34.393972ms)
 ✔ a bus code on stderr becomes the exit code; garbage and timeouts are outcome problems (2176.092888ms)
 ✔ busExit and refuseInsideAgent (0.551635ms)
 ℹ tests 49
 ℹ suites 0
 ℹ pass 49
 ℹ fail 0
 ℹ cancelled 0
 ℹ skipped 0
 ℹ todo 0
 ℹ duration_ms 2660.213546
@@ -0,0 +1,132 @@
 ✔ explicit request, Seen, ordinary completion and a new request have distinct attention states (6.010669ms)
 ✔ attention convention ignores reasoning/quoted examples and permits leading blank lines (0.427305ms)
 ✔ completed smoke replies and ordinary questions are idle, not human blockers (1.380473ms)
 ✔ only an explicit first-line input request makes a finished reply waiting (0.350763ms)
 ✔ tool activity, user text, errors and unfinished turns override attention text (0.244757ms)
 ✔ STOP access failure is unknown, not absence, under a non-root identity (46.560522ms)
 ✔ connector Task never inherits Discord routing envelopes; ordinary Task still uses user text (4.628469ms)
 ✔ connector discovery keeps only safe identity; rejects modes, mismatches, links and traversal (2.347115ms)
 ✔ canonical owner identity and STOP are independent; no tmux fallback or forged registration (8.011484ms)
 ✔ connector reply refusal precedes forged live tmux registration; ordinary agent still sends (1.083299ms)
 ✔ server rescans connector discovery and refuses HTTP reply without transport (44.367227ms)
 ✔ connector session links and linked directories are not read (1.464268ms)
 ✔ newer live matching launch marks old activity, preserves history/attention/attribution, then clears on new activity (3.99754ms)
 ✔ CLI print uses the relaunch notice instead of old current preview (95.092586ms)
 ✔ connector owner and fixed task never inherit a native relaunch notice (3.122909ms)
 ✔ equality, stale/unknown/offline, mismatched registration and unknown activity do not assert relaunch (1.855835ms)
 ✔ loadConfig: missing file throws ConfigError (2.170209ms)
 ✔ loadConfig: invalid JSON throws ConfigError (0.404033ms)
 ✔ loadConfig: missing dataRoot throws ConfigError (0.309222ms)
 ✔ loadConfig: relative dataRoot throws ConfigError (0.423873ms)
 ✔ loadConfig: valid config returns dataRoot (0.828693ms)
 ✔ findNewestSession: picks the newest by mtime among two files (0.711595ms)
 ✔ findNewestSession: finds files in nested subdirectories (0.463025ms)
 ✔ findNewestSession: returns null for a missing dir (0.166929ms)
 ✔ readSession: extracts fields, collapses/truncates text, counts a truncated final line (0.91581ms)
 ✔ readSession: model and provider follow the latest model_change entry or assistant turn; null when the log names neither; scanAgent carries them (1.74382ms)
 ✔ readSession: lastError carries the assistant errorMessage only when the last assistant turn errored (0.68941ms)
 ✔ findNewestSession/scan: never read sibling auth or secrets next to a sessions dir (1.513219ms)
 ✔ deriveState: full state table (0.217441ms)
 ✔ rule: newest entry is an assistant message with a tool call, after a question-looking text, is working (0.584263ms)
 ✔ rule: newest entry is a tool result with no assistant text after it is working (0.4952ms)
 ✔ rule: a finished ordinary turn is idle, even if it says your move (0.461533ms)
 ✔ task: the first user message of the session, from text blocks (0.358351ms)
 ✔ task: a plain-string user content is accepted, whitespace collapsed and long text capped (0.40596ms)
 ✔ task: no user message in the log means null (shown as unknown), never a guess (0.36415ms)
 ✔ workspace: the live tmux pane path wins; the session cwd is the fallback; neither means null (0.529594ms)
 ✔ activeProject: basename of the nearest .git directory or .git file above the workspace; none means null (0.638536ms)
 ✔ scan: the written record carries task, workspace and activeProject (0.93127ms)
 ✔ registration: overrides task, project and workspace; every source says registration; registered carries the launch fields; the grouping column is untouched (0.9256ms)
 ✔ registration: empty task and null project/workspace leave the derived values in place; registered is still non-null (0.674264ms)
 ✔ registration: a record whose pid is gone is stale; derived values win, sources say derived, registered stays with alive false; a pid the probe cannot decide is not stale; pidAlive itself (1.455794ms)
 ✔ registration: no registration leaves the Gate A fields exactly as before, and registered is null (0.591052ms)
 ✔ loadRegistrations: a missing seatsDir gives empty lists (0.216108ms)
 ✔ loadRegistrations: one good record, one malformed JSON, one with an unknown field; a stray file under seatsDir is ignored (1.142905ms)
 ✔ matchRegistration: matches by sessionsDir, and by realpath through a symlink; sessionsDir null never matches; same seat name with a different sessionsDir does not match (fleet vs repo darkwing) (0.43043ms)
 ✔ scan: writes the registration override to disk; index.json carries registered and registrationErrors (1.014589ms)
 ✔ scan: a relative seatsDir throws ConfigError; an omitted seatsDir behaves as before (0.775468ms)
 ✔ scanAgent: waitingOnYou is true for waiting/error and false otherwise (0.638485ms)
 ✔ scanAgent: ageSeconds is computed from the injected now (0.325922ms)
 ✔ scanAgent: sessionFile null and state idle when sessions dir is empty but alive (0.245852ms)
 ✔ discoverRepoAgents: finds agents with a sessions dir, skips those without, sorted by name (0.434228ms)
 ✔ discoverFleetAgents: finds agents with a sessions dir, sorted by name, fleet tmux fields (0.552141ms)
 ✔ scan: writes per-agent files and index.json, rerun overwrites, no leftover tmp files (1.187081ms)
 ✔ scan: relative boardDir throws ConfigError (0.145247ms)
 ✔ CLI: scan with assume-alive liveness exits 0, prints board summary, writes board files (86.948336ms)
 ✔ CLI: missing config exits 2 with a refused: message (60.852136ms)
 ✔ CLI: unknown command exits 2 (70.857596ms)
 ✔ CLI: unknown --liveness value exits 2 (77.483179ms)
 ✔ panesRunPi: true when any trimmed line equals 'pi' (0.263144ms)
 ✔ panesRunPi: false for bash-only, claude, empty, or node-pi-style lines (0.09522ms)
 ✔ tmuxIsAlive: a pane running pi is alive (0.206673ms)
 ✔ tmuxIsAlive: session exists but pi has exited is not alive (0.10368ms)
 ✔ tmuxIsAlive: no such tmux session is not alive (0.058347ms)
 ✔ tmuxIsAlive: tmux could not be run at all is unknown (null), never assumed alive (0.067732ms)
 ✔ tmuxIsAlive: passes -L <socket> only when a socket is given (0.105159ms)
 ✔ parsePanes: one pane per line, command and optional tab-separated path (0.08465ms)
 ✔ tmuxInspect: reports the path of the pane running pi, not of a shell pane (0.085903ms)
 ✔ tmuxInspect: no pi pane, no session, or no tmux gives no workspace and the matching liveness (0.102332ms)
 ✔ loadSeen: missing file returns {} (0.17577ms)
 ✔ loadSeen: invalid JSON throws ConfigError (0.222621ms)
 ✔ loadSeen: a JSON array throws ConfigError (0.215859ms)
 ✔ loadSeen: a non-string value throws ConfigError (0.197642ms)
 ✔ markSeen: seen true adds the key and writes seen.json mode 0600, no leftover tmp files (0.474763ms)
 ✔ markSeen: seen false deletes the key (0.295837ms)
 ✔ markSeen: missing, empty, or non-string fields throw ConfigError (0.214289ms)
 ✔ markSeen: project containing '/' throws ConfigError (0.663767ms)
 ✔ markSeen: non-boolean seen throws ConfigError (0.163287ms)
 ✔ scanAgent: a seen mark matching the waiting session's lastTimestamp clears waitingOnYou (0.366334ms)
 ✔ scanAgent: a stale mark (agent wrote something newer) is not seen and waitingOnYou is true (0.291198ms)
 ✔ scanAgent: a working session with a matching mark is not seen (marks only apply to waiting/error) (0.267725ms)
 ✔ scanAgent: an error-state session with a matching mark is seen (0.310516ms)
 ✔ scan: index.seen and waitingOnYou reflect seen.json, which scan never rewrites or deletes (0.613896ms)
 ✔ scan: a corrupt seen.json makes scan throw ConfigError (fail closed) (0.22487ms)
 ✔ taskSetBy: a registered task carries the record's setter; a record without the field (pre-#1511) reads unknown; the value is not copied into registered (0.830851ms)
 ✔ taskSetBy: null whenever the task shown is not the registered one: no registration, an empty registered task, a stale registration; the field is always present (0.914619ms)
 ✔ taskSetBy: scan() reads the field from disk through the seat package (bounded there), writes it to the per-agent record and index, and an invalid on-disk value is a registrationError, never a row value (1.340662ms)
 ✔ isLoopbackHost: recognizes loopback hosts (1.861483ms)
 ✔ isLoopbackHost: rejects non-loopback hosts (6.326495ms)
 ✔ startServer: refuses a non-loopback host with ConfigError, never opens a socket (4.294971ms)
 ✔ startServer: serves page, healthz, and a rescanning /api/board (52.193955ms)
 ✔ startServer: a seatsDir registration overrides the row and index.registered reflects it (10.198557ms)
 ✔ startServer: /api/board returns 500 JSON with an error field when scan throws (3.653617ms)
 ✔ CLI: serve refuses a non-loopback host with exit 2 and a refused: message (67.059083ms)
 ✔ CLI: serve rejects a non-numeric --port with exit 2 (72.917415ms)
 ✔ CLI: scan still works after the async cli refactor (71.064375ms)
 ✔ CLI: live serve prints its URL and answers /healthz (74.988166ms)
 ✔ page.html: esc() escapes every HTML-significant character (0.657462ms)
 ✔ POST /api/seen marks a row; GET /api/board still shows it seen; seen:false clears it (8.813455ms)
 ✔ POST /api/seen without a JSON content-type returns 400 and does not write a mark (2.279532ms)
 ✔ POST /api/seen with invalid JSON returns 400 (2.984055ms)
 ✔ POST /api/seen with a body over 4096 bytes returns 400 (or resets the connection) and writes no mark (2.586391ms)
 ✔ POST /api/seen with a missing agent returns 400 (1.774095ms)
 ✔ POST /api/board returns 405; PUT /api/seen returns 405 (1.963635ms)
 ✔ CLI: scan --print marks a seen row with 's' and the summary line ends with 'N seen)' (62.667316ms)
 ✔ page.html: seenControl() escapes rec.project/agent/lastActivity, and the POST uses a JSON content-type (0.392194ms)
 ✔ page.html: has a collapsed Seen section that lists seen rows with the shared row builder (0.385316ms)
 ✔ page.html: each project has a Hide seen checkbox (default on) beside Hide offline, with a hidden-count note (0.239229ms)
 ✔ page.html: a project header reads "N of N" only while a checkbox hides rows (0.159999ms)
 ✔ page.html: every row shows Task and Active project, derived or the word unknown, with the workspace in the detail (0.377441ms)
 ✔ page.html: task and active project cells show their source via sourceTag(); the detail has a Registered row via registeredText(); SOURCE_LABEL maps registration to registered; every dynamic value in sourceTag/fromSource/registeredText is escaped (0.779315ms)
 ✔ POST /api/reply: runs agent-send.sh with -s from the registration, -S <host>:control-board, -m text plus the fixed trailer, no -L on the default socket, MOSAIC_TMUX_SOCKET stripped; answers delivered with the exit code and both streams (39.730547ms)
 ✔ POST /api/reply: a registration with a tmux socket adds -L <socket> (36.773894ms)
 ✔ POST /api/reply: a non-zero tool exit is a 200 with delivered false, the exit code and the stderr verbatim (34.067307ms)
 ✔ POST /api/reply: refusals before the tool runs: empty or blank or long text 400, unknown row 404, no registration 409, stale registration 409, no tmux session 409, bad JSON 400; the tool is never called (29.012725ms)
 ✔ POST /api/reply: a missing agent-send.sh is a 500 with the path in the error, not a crash (6.244474ms)
 ✔ replyToRow: DEFAULT_AGENT_SEND is the repository's tools/tmux/agent-send.sh and it is executable (0.212342ms)
 ✔ page.html: the reply box appears only where canReply() holds (live registration with a tmux session), the detail has a Reply row, the submit posts JSON to /api/reply, receipts and drafts survive a refresh, and every receipt value is escaped (0.686283ms)
 ✔ startServer: /api/board carries taskSetBy from a live registration and null for the derived rows (2.987338ms)
 ✔ page.html: the task cell and detail show who set a registered task via setByTag()/setByText(), both escaped, only from rec.taskSetBy; the reply gate does not read it (0.390807ms)
 ✔ Host/Origin guard: GET /api/board and POST /api/reply refuse a foreign Host, a wrong port and a cross-origin Origin with 403 JSON, before any scan or send, and never send CORS headers (9.577638ms)
 ✔ Host/Origin guard: loopback names on this port are accepted, with or without a same-origin Origin (36.075218ms)
 ✔ conversation routes (F16): a foreign Host, a wrong port and a cross-origin Origin get 403 before the reader runs, with no CORS headers (6.60021ms)
 ✔ every refusal code the reader can raise has an HTTP status (3.982216ms)
 ✔ conversation routes: catalogue, first page, next page and follow over HTTP; refusals map to 4xx with their code; nothing is written (49.201664ms)
 ℹ tests 124
 ℹ suites 0
 ℹ pass 124
 ℹ fail 0
 ℹ cancelled 0
 ℹ skipped 0
 ℹ todo 0
 ℹ duration_ms 827.412997
@@ -0,0 +1,205 @@
 ✔ W1: two processes acquire the same pair at once; exactly one claim (208.33252ms)
 ✔ W1: two writers publish the same revision at once: one wins, the other gets null, the winner's record stays (9.811798ms)
 ✔ W1: a revision name appears only after its bytes are synced; before that, only a temp file exists (5.537445ms)
 ✔ W2: acquire while a claim is reserved or active refuses already-active (228.104072ms)
 ✔ W3: acquire while stopping, uncertain, or stopped without proof refuses unsafe-replacement (324.120499ms)
 ✔ W4: same session with another seat tuple, and the reverse, both refuse; a loser on the seat key closes it no-unit (209.598509ms)
 ✔ W4: a hard link of one session under another seat is the same session: the second controller refuses already-active and launches nothing (27.238547ms)
 ✔ W4: a copy of one session under another seat is the same session: the second controller refuses already-active and launches nothing (22.122492ms)
 ✔ W4: a session header ID that changes after construction refuses target; nothing is claimed or launched (3.087845ms)
 ✔ W5: SIGKILL between every publication barrier of acquire and transition; restart never finds two holders or a lost claim (5701.445297ms)
 ✔ W5: SIGKILL between every publication barrier of release; restart finishes or holds the release (22244.010272ms)
 ✔ W6: controller killed mid-turn while the engine lives; restart is uncertain, no launch, prompts refuse (149.81459ms)
 ✔ W12: a live owner paused with SIGSTOP; a second controller refuses already-active and changes nothing (88.345296ms)
 ✔ W13: crash after the engine spawns, before active; restart finds the live unit: uncertain, no second spawn, force stop only (223.831116ms)
 ✔ W14: crash after reservation, before the spawn marker: stopped with a no-unit observation; the pair is free (188.323071ms)
 ✔ W20: crash after the spawn marker, scope collected; uncertain in both runs, the marker is copied, no launch until a boot proof (217.683612ms)
 ✔ W15: crash between the two keys during release; restart finishes it under the same claim ID (32.560634ms)
 ✔ W7: recorded boot ID differs on the same machine: stopped with a boot proof; open tool calls become uncertain (96.660024ms)
 ✔ W8: resume after a proven stop with the same pins: new claim ID, generation +1, same conversation, branch and leaf (32.019907ms)
 ✔ W9: resume with a changed binary, argv digest, branch or leaf is refused and the claim is unchanged (91.969696ms)
 ✔ W11: the controller writes no session file; only the fake engine's own appends appear (22.329944ms)
 ✔ W16: a highest revision that won't parse holds the pair uncertain; the older stopped revision is not reused (54.201621ms)
 ✔ W17: a claim root copied from another host refuses foreign-host and promotes nothing (60.552307ms)
 ✔ G1: a session path or claim root under .pi/state, ~/.claude, the data root or a registration refuses at construction (3.596267ms)
 ✔ G2: a symlink inside the fixture root to a live session file is refused by the real-path check (1.126052ms)
 ✔ G3: a fixture path swapped for a live path after construction is refused at bind (1.770404ms)
 ✖ K1: force stop kills a tool child that called setsid; stopped with a verified proof (2653.677511ms)
 ✔ K2: K1 on the process-group fallback ends uncertain, never stopped (129.308388ms)
 ✖ K3: SIGTERM acknowledged while a member lives: stopping until the kill phase, never stopped from TERM (2165.26786ms)
 ✔ K4: two engines; force stop one; the other survives by independent observation (4287.963097ms)
 ✔ K5: a stop during a tool call leaves the effect uncertain, and it is shown (2182.565146ms)
 ✔ K12: a member forking in a loop: the freeze stops it, enumeration is complete, populated 0 after cgroup.kill (2258.603392ms)
 ✔ K13: a member writing its pid into another cgroup is refused by the namespace; the kill is complete (2153.528491ms)
 ✔ K15: the shim gone, engine/cgroup.events unreadable, or the engine cgroup missing: evidence unavailable, not empty; uncertain (4403.942608ms)
 ✖ K10: controller killed between the TERM and kill phases: restart checks the invocation ID and re-runs from TERM for the same stop (180.585304ms)
 ✔ K11: controller killed after the confirmation is recorded, before TERM: restart checks the invocation ID and re-runs from TERM for the same stop (302.630477ms)
 ✔ K14: a unit with the recorded name but another invocation ID: evidence unavailable, no signals, uncertain (265.388551ms)
 ✔ K6: recover without proof, without confirmation, or with changed pins is refused (69.424274ms)
 ✔ K7: recover after proof, then launch: new claim and execution, generation +1, same leaf; the cancelled prompt is not replayed (38.2806ms)
 ✔ K8: an engine that loads another leaf on resume is refused before admission; it stays claimed until a proven stop (48.757019ms)
 ✔ K9: an interrupt that never settles stays uncertain; force stop stays available; takeover is refused while fenced (3026.567752ms)
 ✔ K16: a claim from another machine ID refuses foreign-host; no boot proof is issued (5.577601ms)
 ✔ K17: two launcher calls with one eligibility record: one launch, the other refuses, no second engine (35.813713ms)
 ✔ K18: the leaf changes after eligibility: launch refused; the reservation stays until released with proof (28.909283ms)
 ✔ S1: `/goal x`, with leading spaces or a tab, refuses text-policy at admission; zero engine bytes (49.842944ms)
 ✔ S2: every prefix pinned Pi interprets is refused, from the list the code uses; the rest reach the engine exactly (37.949131ms)
 ✔ S3: `/goal` on the second line is pinned from the source: Pi checks only index 0, so it is admitted and sent exactly (34.393597ms)
 ✔ S4: a `/` left in the composer is cleared when control transfers and returns; the next submit sends only the new text (55.353224ms)
 ✔ S5: an observer terminal gets a paste then Enter, as send-message.sh does: not admitted: controller, nothing sent (24.11355ms)
 ✔ S6: a mediated-shaped registration (no tmux) passed to the board's replyToRow: 409 no tmux session; exec never runs (0.684512ms)
 ✔ S7: ESC, bracketed-paste markers and U+2028/U+2029 travel as one JSON string; the engine receives the exact text in one record (26.735815ms)
 ✔ P3: a Pi confirm, select, input or editor dialog is shown disabled with a reason and never answered (127.693023ms)
 ✔ E1: send, ack, user, toolCall, toolResult, final answer: shown once, no refresh, draft and reading position kept (43.227988ms)
 ✔ E2: U+2028, U+2029 inside JSON strings and CRLF line ends each parse as one record, on the splitter and through the controller (26.124057ms)
 ✔ E3: a multipart final, two blocks, null request correlation and duplicate delivery (38.762027ms)
 ✔ E4: a page read after message_end but before its entry is persisted: marker at the seam, re-read after run-settled, each message once (38.724253ms)
 ✔ E4: a gap or a new epoch also reconciles; nothing is concatenated across a gap (10.71162ms)
 ✔ E5: an unknown native event gives no client event; evidence records its type and bytes; the terminal count goes up (30.187625ms)
 ✔ E6: a tool result delayed across a pause and a reconnect is reconciled without a manual refresh (48.049709ms)
 ✔ E7: the terminal renders the same stream as the library client, as observer and then as controller, and submits only as controller (43.974693ms)
 ✔ terminal: engine control characters are made visible; a lost connection refuses submit (30.554731ms)
 ✔ terminal: outcome unknown is shown as such, with no resend offer, and nothing is resent (0.71206ms)
 ✔ terminal: text after Enter in the same input chunk starts the next message; it never joins the one submitted (0.501472ms)
 ✔ terminal: a paste-start marker split right after its ESC still opens the paste; the Enter inside it never submits (0.71016ms)
 ✔ terminal: invisible and bidi characters are made visible; head, status and notice lines stay one line (0.205325ms)
 ✔ every record these fixtures produced is a valid CHAT-01 record (E5: no record fails the schema) (440.560715ms)
 ✔ H1: two takeovers with the same expected generation: one wins, +1; the other refuses generation (70.048519ms)
 ✔ H2: the old controller's prompt after a takeover commits is refused with zero engine bytes (89.612495ms)
 ✔ H3: a takeover while a prompt holds the dispatch lock: written under the old actor, or refused; never both (151.740861ms)
 ✔ H4: self-takeover is refused (19.567355ms)
 ✔ H9: Interrupt racing a prompt's dispatch: before the write, dispatch-refused and no-turn; after, §3 rules (126.266639ms)
 ✔ H10: Interrupt and force stop together: one stop chain, force stop supersedes (127.987954ms)
 ✔ H10: an overlap during the pause before the abort: no abort, the stop ends uncertain (30.194643ms)
 ✔ H10: a no-turn Interrupt lifts only its own fence; admission stays closed under force stop, overlap or revocation (90.744705ms)
 ✔ H11: the controller disconnects mid-turn: work continues, the claim is unchanged, control stays put (137.197933ms)
 ✔ H12: an exact retry after reconnecting to the same incarnation returns the same receipt; one dispatch (15.871752ms)
 ✔ H13: a retry with the same request ID and different text is refused (16.248157ms)
 ✔ H14: late stdout from the old engine after a replacement is dropped by incarnation, counted, never rendered (137.840526ms)
 ✔ H15: a revoked connection's command is refused; the revocation fence holds (67.750797ms)
 ✔ H16: a second controller for the same session refuses already-active; the first is untouched (18.408706ms)
 ✔ H10: a second force stop while the first escalation runs refuses fenced; one escalation, and the claim records only the first stop's phases (58.879625ms)
 ✔ H17: a confirmation reused, answered from another connection, or used after the stop changed is refused (54.75082ms)
 ✔ H18: two prompts before any native output: the second refuses busy; one engine write (15.210695ms)
 ✔ H19: the pipe fails mid-line under a large prompt: delivery-unknown transport-unknown, poisoned, no later write (119.912673ms)
 ✔ H19: the link itself never writes again after an unknown outcome, whoever calls it (0.74475ms)
 ✔ H19: the controller dies mid-write of a large line: after restart the outcome is unknown and nothing is resent (526.777789ms)
 ✔ H20: the line is written but the ack is lost when the controller dies: orphan, outcome unknown, nothing resent (406.518664ms)
 ✔ H21: a retry of the exact request with the old token after a crash is stale-incarnation; no second write (372.815504ms)
 ✔ H22: after H21 and a valid recovery, a new request with the new token is admitted (2470.446235ms)
 ✔ H23: requests pending at a restart are not resent; each shows outcome unknown (457.248222ms)
 ✔ a plain conversation: catalogue row, one page, CHAT-01 records (11.82056ms)
 ✔ native entries map to blocks: tools, thinking, bash, notices, ids that do not fit (2.271047ms)
 ✔ F1: a malformed line is an unavailable part at its position, and reading continues (2.730979ms)
 ✔ F1: a missing parent stops the history with a notice that names the unreadable lines (3.52433ms)
 ✔ F1: an unreadable fork is never merged into another branch's history (1.850162ms)
 ✔ F1: a follow stays on its branch when the next entry's parent is unreadable (4.636359ms)
 ✔ F1: a file whose entries are all unreadable shows a notice per line (1.641695ms)
 ✔ F2: a truncated trailing line marks the view incomplete, not an error (2.301832ms)
 ✔ pagination: 100 parts, then the rest; parts concatenate to the whole branch (7.597286ms)
 ✔ F3: a replaced file (new inode) refuses old cursors with reconcile (5.634047ms)
 ✔ F4: a same-inode rewrite of the prefix refuses old cursors with reconcile (7.869533ms)
 ✔ F5: growth between pages keeps the epoch and the page stops at the pinned length (11.185671ms)
 ✔ F6: unknown, foreign and expired cursors refuse and leave the cursor usable (14.562294ms)
 ✔ F7: a symlinked file and a symlinked directory component are refused, never opened (15.568579ms)
 ✔ F8: a file swapped for a symlink after the catalogue is refused (3.87859ms)
 ✔ F9: registrations never add or redirect a root (2.922432ms)
 ✔ F10: a header cwd naming another project is refused (6.505993ms)
 ✔ F11: parentSession renders with a marker and the parent is never opened (1.280591ms)
 ✔ F12: two leaves: the default leaf is shown and the other branch reads alone (8.701968ms)
 ✔ F12: a follow refuses when an appended duplicate id changes the branch's earlier parts (3.047678ms)
 ✔ F12: a second root (Pi's resetLeaf) starts its own branch (2.529818ms)
 ✔ F13: compaction is a marker in place, then the retained content (1.200056ms)
 ✔ F14: long strings split into fragments and parts, reassemble exactly, and pages respect the byte cap (927.08769ms)
 ✔ fragments never cut a surrogate pair and keep an empty string (6.246556ms)
 ✔ F15: a Claude seat is an unsupported-harness placeholder whose directory is never read (1.891982ms)
 ✔ unknown conversations, empty files and non-Pi files refuse (3.354143ms)
 ✔ an unreadable file or root inside the roots is refused per row, not a failed catalogue (1.403634ms)
 ✔ a seat directory without search permission refuses that root, not the catalogue (2.386158ms)
 ✔ every page and cursor is a valid CHAT-01 record (963.005061ms)
 ✔ the engine pin holds for the installed package (4.64981ms)
 ✔ pinned Pi, sealed and without credentials, answers the controller's commands with the shapes the fake models (492.509878ms)
 ✔ pinned Pi appends thinking_level_change at start when the branch lacks one, so the leaf moves (K8 then fails closed) (449.300806ms)
 ✔ N25: ordinary Interrupt reconciles; a non-empty queue_update in the window is O5 (99.699963ms)
 ✔ N1: an extension's follow-up queued after the fence is cleared before any abort; O5, Unknown (56.005591ms)
 ✔ N1: a follow-up queued before the fence is O5 at once; the Interrupt refuses fenced (36.010734ms)
 ✔ N2: with abort first, the fake runs the external item (the ordering guard has teeth) (23.120368ms)
 ✔ N3: the fence lands in preflight, preflight errors, no run: failed, No run, uncertain (40.605984ms)
 ✔ N4: the ack arrives after the first abort and a run starts: clear and abort again; Interrupted (32.198961ms)
 ✔ N5: an input handler takes the prompt: ack, no run, delivery-unknown handled-without-run (129.781085ms)
 ✔ N6: an extension queues between clear_queue and abort: O5 and O6, Unknown (60.975823ms)
 ✔ N7: clear_queue times out: no abort, nativeQueue unknown, force stop still ends it (1540.123798ms)
 ✔ N7: clear_queue answers an error: no abort, nativeQueue unknown, the link not poisoned (17.607252ms)
 ✔ N8: an extension prompt starts a run during Mosaic preflight; the losing settle is O3 (71.219891ms)
 ✔ N9: a run that started before the fence and ends aborted: failed interrupted, Interrupted (14.260687ms)
 ✔ N9: decision 34: a run that ends aborted with no stop in progress: aborted-without-stop, uncertain, outcome unknown (19.355928ms)
 ✔ N9: an aborted that lands after the fence but before any abort is written: aborted-without-stop, Unknown (40.425771ms)
 ✔ N10: fake conformance (34.674337ms)
 ✔ N11: the run fails before any user message_start: delivery-unknown ack-without-start, never failed (45.385869ms)
 ✔ N12: input that starts a run after the final empty clear is O1 and not part of the stop's proof (26.329147ms)
 ✔ N13: agent_start with no slot held is O1; a later prompt refuses with zero engine bytes (70.140045ms)
 ✔ N14: the run completes while clear_queue is in flight: finished, Completed first, uncertain (30.960887ms)
 ✔ N14: the run completes after the abort is written, before Pi applies it: finished, never relabelled (33.997394ms)
 ✔ N15: the fence lands in preflight, then an input handler takes it: handled-without-run, No run (24.416409ms)
 ✔ N16: Interrupt with no slot and no run refuses no-turn: no stop, no bytes, admission open (13.466921ms)
 ✔ N17: the run fails on its own during the exchange: failed, Failed on its own (26.194657ms)
 ✔ N18: no final assistant message_end, or a lost line: working stays working; before working, transport-unknown (105.895233ms)
 ✔ N19: a losing extension prompt settles inside the Mosaic run before its user message: O3, run-overlap (137.573966ms)
 ✔ N20: an extension triggerTurn during Mosaic preflight starts first; while streaming it queues with no signal (85.246925ms)
 ✔ N21: a losing settle after the receipt settled finished is O2; the receipt stays finished (16.090188ms)
 ✔ N22: an agent-level custom message is dropped by the clear with no signal; evidence names the seal (14.423581ms)
 ✔ N23: a nextTurn message survives clear and abort and attaches to the next prompt, with no signal (14.347618ms)
 ✔ N24: the seal is an allow-list: --extension, a missing --no-* flag, a second --mode or --session, a session or output flag, or a stray word refuses unsealed-engine; no engine starts (44.072894ms)
 ℹ tests 152
 ℹ suites 0
 ℹ pass 149
 ℹ fail 3
 ℹ cancelled 0
 ℹ skipped 0
 ℹ todo 0
 ℹ duration_ms 32310.893453
 ✖ failing tests:
 test at packages/conversation/tests/cohort.test.mjs:139:1
 ✖ K1: force stop kills a tool child that called setsid; stopped with a verified proof (2653.677511ms)
  AssertionError [ERR_ASSERTION]: the escaped child is a listed member
      at TestContext.<anonymous> (file:///home/jwoltje/filbert-scratch/r45/base/packages/conversation/tests/cohort.test.mjs:151:12)
      at async Test.run (node:internal/test_runner/test:1409:7)
      at async startSubtestAfterBootstrap (node:internal/test_runner/harness:387:3) {
    generatedMessage: false,
    code: 'ERR_ASSERTION',
    actual: false,
    expected: true,
    operator: '==',
    diff: 'simple'
  }
 test at packages/conversation/tests/cohort.test.mjs:176:1
 ✖ K3: SIGTERM acknowledged while a member lives: stopping until the kill phase, never stopped from TERM (2165.26786ms)
  AssertionError [ERR_ASSERTION]: the member ignored TERM
      at TestContext.<anonymous> (file:///home/jwoltje/filbert-scratch/r45/base/packages/conversation/tests/cohort.test.mjs:190:12)
      at async Test.run (node:internal/test_runner/test:1409:7)
      at async Test.processPendingSubtests (node:internal/test_runner/test:974:7) {
    generatedMessage: false,
    code: 'ERR_ASSERTION',
    actual: false,
    expected: true,
    operator: '==',
    diff: 'simple'
  }
 test at packages/conversation/tests/cohort.test.mjs:426:3
 ✖ K10: controller killed between the TERM and kill phases: restart checks the invocation ID and re-runs from TERM for the same stop (180.585304ms)
  AssertionError [ERR_ASSERTION]: the member is alive across the crash
      at TestContext.<anonymous> (file:///home/jwoltje/filbert-scratch/r45/base/packages/conversation/tests/cohort.test.mjs:431:12)
      at process.processTicksAndRejections (node:internal/process/task_queues:104:5)
      at async Test.run (node:internal/test_runner/test:1409:7)
      at async Test.processPendingSubtests (node:internal/test_runner/test:974:7) {
    generatedMessage: false,
    code: 'ERR_ASSERTION',
    actual: false,
    expected: true,
    operator: '==',
    diff: 'simple'
  }
@@ -0,0 +1,186 @@
 ✔ approvals: a request is validated before anything is posted; the rendering shows names and never ids (3.201101ms)
 ✔ approvals: the ledger is appended and folded into open requests with bind and approval states (1.660034ms)
 ✔ approvals: a reply approves only when it points at a request, says exactly approve, and comes from a listed approver once (0.580701ms)
 ✔ approvals: a button approves only on its own request message with the matching custom id (0.486235ms)
 ✔ approvals flow: a turn that opened a request posts the message with the button, records it, binds it, and both approvers approve (21.333687ms)
 ✔ approvals flow: a non-approver, a repeat, a wrong custom id and a service refusal each get their fixed line and a drop entry (7.393055ms)
 ✔ approvals flow: an invalid request from the model, a refused post, and no api client are recorded and post nothing (9.050164ms)
 ✔ approvals flow: start retries a bind and an approval left as unknown, under their original keys (2.631665ms)
 ✔ authorize: open channel, listed user (1.952524ms)
 ✔ authorize: wrong guild (0.194831ms)
 ✔ authorize: no guild (DM) (0.147706ms)
 ✔ authorize: unlisted channel (0.285655ms)
 ✔ authorize: unknown channel, no info (0.196753ms)
 ✔ authorize: thread of listed parent (0.435598ms)
 ✔ authorize: thread of unlisted parent (0.164288ms)
 ✔ authorize: text channel that is not a thread and not listed (0.135153ms)
 ✔ authorize: unlisted user (0.160797ms)
 ✔ authorize: no author (0.24555ms)
 ✔ authorize: bot author (listed id, bot flag) (1.405614ms)
 ✔ authorize: system author (0.126607ms)
 ✔ authorize: the bot itself (0.082559ms)
 ✔ authorize: webhook (0.078451ms)
 ✔ authorize: mention channel without mention (0.129292ms)
 ✔ authorize: mention channel with bot mention (0.130115ms)
 ✔ authorize: mention channel with @everyone only (0.138676ms)
 ✔ authorize: mention channel mentioning someone else (0.08371ms)
 ✔ authorize: mention channel, content says @bot but mentions empty (0.097637ms)
 ✔ authorize: private thread under mention channel, mentioned (0.089673ms)
 ✔ authorize: private thread under mention channel, not mentioned (0.077673ms)
 ✔ authorize: thread in another guild per channel info (0.128817ms)
 ✔ authorize: not an object (0.062853ms)
 ✔ authorize: no id (0.161612ms)
 ✔ authorize: oversize content is accepted and flagged (0.070759ms)
 ✔ authorize: exactly the limit is not oversize (0.058304ms)
 ✔ authorize: a user's channel allowlist drops them outside it, threads count as the parent, others are unaffected (0.476843ms)
 ✔ authorize: order puts wrong guild before user, and user before channel (no channel lookup for strangers) (0.126935ms)
 ✔ binding: a complete binding validates and is frozen (4.063052ms)
 ✔ binding: unknown key, missing field, wrong type refuse with exit 2 (1.558439ms)
 ✔ binding: empty allowlists refuse (0.781921ms)
 ✔ binding: a user's channel allowlist must be non-empty, listed and unique; absent means every listed channel (1.743709ms)
 ✔ reloadDiff: reloadable keys are summarised by id; every fixed key refuses with exit 2 (2.132428ms)
 ✔ binding: dmRecipient is optional, a snowflake, one of the listed users, and a fixed key (0.614055ms)
 ✔ binding: file must be 0600, regular, not a symlink (2.040162ms)
 ✔ binding: token file mode, symlink, emptiness and shape are checked; token never appears in errors (1.347954ms)
 ✔ cli: check refuses a non-0600 token file with exit 2 before any network use (84.997781ms)
 ✔ context files: absolute paths, traversal, symlinks and out-of-repo targets refuse; in-repo files resolve (1.854618ms)
 ✔ cli: check refuses a missing context file and a missing binding with exit 2; usage is exit 4 (312.162463ms)
 ✔ cli: reload validates the file first (exit 2), then needs a live owner (exit 1); usage is exit 4 (200.198135ms)
 ✔ cli: run refuses when STOP is present, before any network use (116.785106ms)
 ✔ binding: tools is optional, validated strictly, a fixed key for reload, and its roots are resolved against the data root (1.681435ms)
 ✔ binding: a git key is validated at load and reaches the extension whole, and only on a writable root (1.164195ms)
 ✔ delivery: an accepted message is in the inbox before the turn, the reply is chunked with one nonce per chunk, and the turn record is write-once (25.061237ms)
 ✔ delivery: refused and unknown outcomes are journaled; a later chunk is not sent after a failure (23.476294ms)
 ✔ delivery: restart with an unknown entry re-sends the same nonce once and reconciles before accepting traffic (1.371221ms)
 ✔ delivery: an unknown entry older than the dedupe window is marked refused, not re-sent; a still-unknown one refuses start (1.19478ms)
 ✔ delivery: repeated unknown reconciliations never refresh the dedupe window; the original intent time decides (1.298255ms)
 ✔ turn: a failed engine turn posts the fixed line, never model output, and writes a failed record (2.87817ms)
 ✔ turn: a second message during a turn is held by the engine, both get their own reply and record (32.869668ms)
 ✔ turn: a thread under a listed channel is answered in the thread; an unknown thread is looked up once (2.761047ms)
 ✔ drop: an unlisted user gets silence and one drop line; no inbox entry, no REST call, no engine call (0.75218ms)
 ✔ drop: an oversize message is accepted into the inbox, answered with the fixed line and journaled as a drop (0.78081ms)
 ✔ restart: an inbox with three ids and a replay of the same three produces zero turns (40.645485ms)
 ✔ stop: STOP present refuses start; STOP written while running refuses new turns and the current one finishes (31.614341ms)
 ✔ ceiling: the ceiling plus one is refused and journaled; one fixed line per UTC day; a new day accepts again (4.192219ms)
 ✔ ceiling: a burst arriving while turns are still running cannot queue past the ceiling (3.866001ms)
 ✔ ceiling: a turn interrupted by a crash still counts after restart; admissions are durable (2.168526ms)
 ✔ ceiling: the daily notice survives a same-day restart; one delivery attempt in total, even when the first attempt crashed mid-flight (4.138629ms)
 ✔ duplicate: the same event delivered twice while the thread lookup is held yields one prompt, one admission and one reply (1.999622ms)
 ✔ journal: no token-shaped string and no model output on the drop path reaches disk (0.428104ms)
 ✔ receipt: an admitted message gets one eyes reaction on the inbound message; drops and refusals get none; a failed reaction is recorded and does not fail the turn (2.076407ms)
 ✔ receipt: Discord refusing the reaction leaves the turn intact and records ok false (1.717986ms)
 ✔ reload: a new user is silent before and answered after; a removed channel goes silent; a lower ceiling applies at once (3.409683ms)
 ✔ reload: a fixed key refuses with exit 2 and the old binding stays in force (2.044094ms)
 ✔ tools: with a tools binding the turn record lists every read and its outcome; without one the field is null (3.595509ms)
 ✔ context: the Discord block names the server, channels and modes, and states the rules from Q15 and Q16 (2.590484ms)
 ✔ context: with tools the block names the roots, keeps file content as data, and says to state refusals plainly (0.907284ms)
 ✔ context: a writable root adds the write rules and says a write is real only once Jason commits (2.336321ms)
 ✔ context: the envelope is one bracketed line then the text; names cannot break the line (1.24491ms)
 ✔ context: a git root swaps the terminal-commit line for the git verbs, and a vault root adds the id protocol (1.4446ms)
 ✔ context: assembleContext concatenates files in launcher format and appends the block; sha256 is stable (1.222964ms)
 ✔ context: splitReply keeps paragraphs together under the limit and splits long ones at lines, spaces, then hard (0.652689ms)
 ✔ engine: buildPiArgs carries the fixed flags, engine settings, session dir and prompt file (2.596485ms)
 ✔ engine: with tools, buildPiArgs turns pi's own tools off, loads the extension explicitly and allowlists exactly our three (2.135157ms)
 ✔ engine: a run with tool turns settles once, on the answer, with every tool call in the result (60.544838ms)
 ✔ engine: a run that ends on a tool-only turn fails the prompt as empty; a retried run settles on the real end (35.320946ms)
 ✔ engine: one prompt, one turn, text and usage come back (29.839246ms)
 ✔ engine: a prompt while streaming is held until pi settles, then sent as its own run, and answered in order (333.239021ms)
 ✔ engine: a held prompt that times out before pi settles fails on its own and is never sent (229.917302ms)
 ✔ engine: timeout sends abort and fails only that turn; the process stays (104.025224ms)
 ✔ engine: tool events from a run that outlived its timeout never land in the next prompt's record (228.179554ms)
 ✔ engine: a prompt after a turn that timed out before its agent_start waits for pi to settle instead of being refused (132.927031ms)
 ✔ engine: when pi has not started a timed-out turn by the end of the abort grace, the engine stops pi and fails held prompts (213.653029ms)
 ✔ engine: a timed-out turn pi starts only after the grace never answers a later prompt (614.160788ms)
 ✔ engine: late events of a run past its grace, before pi exits, answer nothing and nothing more is sent (early prompt response) (1.627204ms)
 ✔ engine: late events of a run past its grace, before pi exits, answer nothing and nothing more is sent (late prompt response) (0.520418ms)
 ✔ engine: a timed-out run pi did start outlives the grace; the next prompt goes out when it ends (429.886242ms)
 ✔ engine: a malformed JSONL line fails the turn, not the process (27.00359ms)
 ✔ engine: a turn that ends in error rejects with the error code; process exit fails pending turns (46.705536ms)
 ✔ gateway: hello -> identify with intents, ready, heartbeat with jitter, ack (2.638535ms)
 ✔ gateway: missed ack closes the socket and resumes with the last sequence (1.923849ms)
 ✔ gateway: op 7 reconnect resumes; op 9 non-resumable re-identifies (0.632194ms)
 ✔ gateway: op 9 resumable resumes (0.429861ms)
 ✔ gateway: close 4014 is fatal, reports the missing intent, never reconnects (0.788995ms)
 ✔ gateway: 4004 and 4013 are fatal too; 1006 reconnects with identify when no session (1.041148ms)
 ✔ gateway: close() is final and unparseable frames are ignored (0.397092ms)
 ✔ git: config validation is strict, needs write: true, a work tree and a private token file (71.308392ms)
 ✔ git: the child environment drops every host git config, names one helper, and carries the token path only for origin (37.288053ms)
 ✔ git: status reports the branch, ahead/behind and changed paths, and refuses off the named branch or mid-merge (65.469921ms)
 ✔ git: parseStatus reads porcelain v2 including renames and conflicts (0.257703ms)
 ✔ git: a commit stages exactly the named files, carries the seat author and the requester trailer, and pushes at once (79.837685ms)
 ✔ git: commit refusals: message, paths, requester, nothing to commit, and an index that already holds other work (85.969098ms)
 ✔ git: a commit whose push fails is still a commit, says so, and the next commit's push carries both (D6) (96.306132ms)
 ✔ git: pull is fast-forward only; a diverged origin or dirty local files refuse with nothing merged (199.121223ms)
 ✔ git: push pushes the named branch only and reports up to date (61.956937ms)
 ✔ git: no token value or token path ever reaches a git argument list; outputs are masked and capped (86.647333ms)
 ✔ git: the credential helper answers get over https from a private file and nothing else (210.101951ms)
 ✔ git: the vault protocol validates before a commit, honours another owner's lock, reserves ids, and locks around writes (749.416922ms)
 ✔ lock: the claim is exclusive; a second start against a live owner refuses (5.298493ms)
 ✔ lock: a stale lock (dead owner, reused pid, or record without start) refuses run and is never signaled; only unlock clears it (5.429571ms)
 ✔ lock: an incomplete claim (directory without owner record) is busy and refuses run; unlock clears it (1.017847ms)
 ✔ lock: an owner record that exists but cannot be read is invalid: never signaled, never removed, never claimed over (2.535219ms)
 ✔ lock: legacy upgrade; a live connector holding a {pid, start} record is unknown, unlock refuses and nothing changes; after it exits, unlock clears it (53.217434ms)
 ✔ lock: a live pid whose record carries a malformed or noncanonical start or boot string is unknown, not a mismatch; nothing signals, removes, or claims over it (292.904169ms)
 ✔ lock: identity syntax; only canonical unsigned decimal start ticks and lowercase boot uuids are identities (0.386939ms)
 ✔ lock: a process whose start marker or boot id cannot be read refuses to claim (0.427908ms)
 ✔ lock: a live pid whose identity cannot be read right now is unknown: never signaled, never removed, never claimed over (0.851219ms)
 ✔ lock: four processes racing for the same binding; exactly one claims it and the others refuse (39.514684ms)
 ✔ lock: stale handoff; concurrent starts over a stale lock all refuse, nothing reclaims, one unlock then exactly one live owner (113.625484ms)
 ✔ lock: four-party schedule; claims landing inside an unlock's gap never survive, one unlock leaves no owner and no residue (65.633357ms)
 ✔ notices: a kind is recorded per UTC day and found again (0.464313ms)
 ✔ notify: the DM channel opens once, every send carries the nonce, and only the message id comes back (5.44218ms)
 ✔ notify: refusals and unknowns surface as RestOutcome without ids; a refused send reopens the channel next time (2.120122ms)
 ✔ notify: no dmRecipient, a non-0600 binding or token, or a missing binding refuse before any network use (2.201331ms)
 ✔ recover: nothing to do is clean; a lock whose owner is gone or that has no record is cleared and STOP ends up absent (65.287043ms)
 ✔ recover: an operator STOP refuses with exit 3 and is never removed, whatever the lock says (43.180652ms)
 ✔ recover: a brake written during the unlock wins; STOP stays with both lines and the start is refused (26.948772ms)
 ✔ recover: a held binding refuses with exit 3 and writes no STOP: live owner, alive pid without verifiable identity, unreadable record (64.561167ms)
 ✔ cli: recover exits 0 when ready, 3 behind a brake or a held binding, and run's own STOP refusal is 3 (658.809302ms)
 ✔ rest: createMessage sends nonce, enforce_nonce, empty allowed_mentions and a soft reply reference (2.434973ms)
 ✔ rest: 429 waits retry_after and retries; 4xx is refused; 5xx and socket errors are unknown (3.239627ms)
 ✔ rest: content and nonce limits are enforced locally; typing never throws (0.599442ms)
 ✔ rest: react PUTs the encoded emoji on the inbound message for @me; 2xx is true, anything else is false and never throws (0.695726ms)
 ✔ rest: createDm posts the recipient and resolves the channel id; 4xx refused, 5xx unknown, bad id never sent (1.531794ms)
 ✔ setspark config: a bare https or loopback origin, a private key file, a principal (5.320663ms)
 ✔ setspark config: reaches the tools config and the binding as a fixed key (2.450232ms)
 ✔ setspark config: the binding's key survives resolveToolRoots and the engine's JSON hand-off to the extension (1.459636ms)
 ✔ setspark config: approvers come from the binding's users, never from the binding's setspark key (1.677375ms)
 ✔ setspark verbs: required_approvers go out as discord ids from names and come back as names (22.100773ms)
 ✔ setspark verbs: no Discord user id reaches tool text, whatever shape the service returns it in (8.043384ms)
 ✔ setspark contract: a decision made with names opens a request the connector accepts; names stored by an old record still refuse (6.883186ms)
 ✔ setspark keys: read per call, one printable token per file, rotation without a restart (4.004438ms)
 ✔ setspark idempotency keys: principal, turn id, call index; connector keys name a step (0.603316ms)
 ✔ setspark http core: json in and out, bearer header, idempotency header, fixed user agent, no key anywhere else (2.745332ms)
 ✔ setspark http core: error bodies become fixed refusals with code and the 409 fields; server text is data, cut (1012.566197ms)
 ✔ setspark verbs: a setspark key enables the eight verbs and no counters (0.527999ms)
 ✔ setspark verbs: writes carry the turn's key and the asserted requester, reads carry no key, and the api key never appears in text or details (6.51347ms)
 ✔ setspark verbs: no turn refuses every write before any request; bad arguments refuse before any request; reads still work (1.238868ms)
 ✔ setspark verbs: renderRecord caps long output and hides the accepted snapshot (0.210266ms)
 ✔ setspark api: bind, add_approval (button and reply) and get use integer request ids and the connector's keys (2.395749ms)
 ✔ tools: config refuses a missing, symlinked, dotted, non-directory or duplicate root and bad limits (4.409164ms)
 ✔ tools: every escape is refused with a fixed reason and nothing outside the root is read (5.544056ms)
 ✔ tools: happy paths list, read a window, and search case-insensitively; dotfiles and symlinks never appear (5.663439ms)
 ✔ tools: the tool set renders text for the model, records details for the journal, and enforces the per-run budget (4.490327ms)
 ✔ tools: listing and search caps hold (9.712986ms)
 ✔ tools: credential shapes are caught; ordinary prose and ids are not (0.91641ms)
 ✔ tools: the read uses the checked file itself; a symlink, a swapped file, a FIFO, a grown file or a hard link at read time is refused (5.396493ms)
 ✔ tools: an unreadable file under the root is skipped by search and refused by read (1.118591ms)
 ✔ tools: config accepts write: true only as a boolean, and enables the write tools only then (1.239234ms)
 ✔ tools: every write outside the fence is refused before any byte lands, and no temp file remains (5.472032ms)
 ✔ tools: write_file leaves the exact bytes, edit_file replaces one exact match, and the set renders the change as uncommitted (2.128317ms)
 ✔ tools: a target that changed between the check and the rename is refused and the temp file is removed (1.204303ms)
 ✔ web: config takes an https or loopback-http SearXNG base url and a bounded fetch cap (4.119455ms)
 ✔ web: address rules refuse every private, loopback, link-local, mapped and multicast form (3.260389ms)
 ✔ web: web_fetch refuses bad urls, private hosts, rebinding names, non-https redirects, too many hops, error status, non-text bodies, and times out (1026.63004ms)
 ✔ web: web_fetch returns html as text with the title, follows an https redirect, keeps plain text and json, and cuts at the cap (4.242048ms)
 ✔ web: html to text drops scripts, styles and comments, decodes entities and keeps block breaks (0.258013ms)
 ✔ web: web_search asks the instance for json, returns at most ten clean results, and refuses a bad query, a down instance or an unusable answer (2.624952ms)
 ✔ web: the tool set enables the web tools only with a web key, counts them in the budget, and records url, status and hits (2.303252ms)
 ℹ tests 178
 ℹ suites 0
 ℹ pass 178
 ℹ fail 0
 ℹ cancelled 0
 ℹ skipped 0
 ℹ todo 0
 ℹ duration_ms 2634.058891
@@ -0,0 +1,86 @@
 ✔ a raw token file, with or without one trailing newline, reaches curl only through the config stream (138.671771ms)
 ✔ the raw path accepts nothing else, and refuses before curl runs (423.976474ms)
 ✔ the file checks still apply on the raw path: mode, symlink, missing, directory (306.653279ms)
 ✔ the raw path base URL has no override (68.146647ms)
 ✔ the JSON path is unchanged, and JSON never falls through to the raw path (268.204784ms)
 ✔ a file that changes between the two reads refuses before curl runs, with or without a body (796.716083ms)
 ✔ the token reaches no child environment, even with an inherited CFG or SHELLOPTS=allexport (747.472393ms)
 ✔ real helper GET HTTP 200 preserves exit 0 without credentials (18.039884ms)
 ✔ real helper POST HTTP 201 preserves exit 0 without credentials (12.248504ms)
 ✔ real helper GET HTTP 403 preserves exit 1 without credentials (11.836651ms)
 ✔ fixture git subjects only, follow-ups and three session kinds (119.607682ms)
 ✔ text and JSON carry same numbers, open and truncated title (186.370317ms)
 ✔ missing credentials exit 2, no-issues never calls API and shows unknown (169.209561ms)
 ✔ empty range gives no rows and zero totals (109.226221ms)
 ✔ inclusive UTC dates, first-line preamble only, role and seat boundaries (113.45241ms)
 ✔ close-only issue included, even median, missing metadata stays unknown (113.56368ms)
 ✔ unique commits but per-issue links count multiple tags once each (112.873951ms)
 ✔ page cap refuses rather than silently undercounting (101.322696ms)
 ✔ bad API payload not JSON refuses (96.714416ms)
 ✔ bad API payload {} refuses (94.745159ms)
 ✔ bad API payload [{"number":1}] refuses (89.245405ms)
 ✔ partial or malformed session log refuses with location, not content (101.876415ms)
 ✔ a U+2028 or U+2029 inside a session string is one line, not a malformed record (132.282475ms)
 ✔ no sessions is an empty table; symlink source refuses (192.239786ms)
 ✔ reads only refactor even when another branch is checked out (105.435584ms)
 ✔ invalid dates, reverse dates and duplicate options refuse (70.488615ms)
 ✔ T3 agent assignments do not count as human in Table 2 (98.640321ms)
 ✔ preamble parsing and issue number boundaries (0.473611ms)
 ✔ T3 header: agent, or board from control-board; anything short of the full header is human (0.160889ms)
 ✔ no closed issues with human messages means undefined ratio, not invented zero (0.230338ms)
 ✔ T3: seat, archived, unmapped and Researcher threads count; imported, deleted and other-project threads do not (196.682254ms)
 ✔ T3: the default path is read from HOME and prints no path line; --no-t3 says so (472.742418ms)
 ✔ T3: a HOME with no database exits 1 and names --no-t3 (88.149974ms)
 ✔ T3: a file that is not a database exits 1 and names --no-t3 (94.805533ms)
 ✔ T3: a seat thread renamed to another seat exits 1 naming thread, title and roles (125.975358ms)
 ✔ T3: an unmapped thread addressed as a seat exits 1 (121.305514ms)
 ✔ T3: a header to another thread id is not cross-checked (123.101002ms)
 ✔ T3: no project, or two, for this root exits 1 (307.420146ms)
 ✔ T3: a removed column exits 1 and names it (103.087633ms)
 ✔ T3: a missing table exits 1 and names it (112.352744ms)
 ✔ T3: a counted row with an unknown role exits 1 without its text (134.640199ms)
 ✔ T3: a counted row with non-text content exits 1 without its text (134.30338ms)
 ✔ T3: a counted row with an unparseable created_at exits 1 without its text (124.285473ms)
 ✔ T3: a missing orchestration_events makes the diagnostic unknown and keeps the counts (228.991249ms)
 ✔ T3: a human message with no event counts in humanWithoutEvent (127.281084ms)
 ✔ T3: an unparseable event makes the diagnostic unknown and keeps the counts (224.491946ms)
 ✔ T3: an event with no string messageId makes the diagnostic unknown and keeps the counts (240.564015ms)
 ✔ T3: an error that is not from SQLite is rethrown, not reported as a database failure (58.530531ms)
 ✔ T3: a symlink at ~/.t3 exits 1 (90.708515ms)
 ✔ T3: a symlink at ~/.t3/userdata exits 1 (85.405863ms)
 ✔ T3: a symlink at ~/.t3/userdata/state.sqlite exits 1 (87.869959ms)
 ✔ T3: with --t3-db, a symlinked file or directory exits 1 (172.781915ms)
 ✔ T3 WAL: the newest message only in -wal, writer attached, is counted (134.769444ms)
 ✔ T3 WAL: stopped cleanly, counts are correct and the main file is unchanged (123.840167ms)
 ✔ T3 WAL: -wal without -shm in a writable directory is read (141.073277ms)
 ✔ T3 WAL: -wal without -shm in a read-only directory exits 1 (134.623712ms)
 ✔ T3 WAL: stopped cleanly in a read-only directory exits 1 (116.480721ms)
 ✔ T3: a lock held past the 5 s busy timeout exits 1 and names --no-t3 (5123.515152ms)
 ✔ a done row whose closing issue is open is a violation; a row that is not done is not (2.262245ms)
 ✔ an issue several rows close is expected closed only once all of them are done (0.636824ms)
 ✔ closure needs positive evidence: unknown is undecided, and so is a skipped or short issue check (0.465932ms)
 ✔ each owner of an in-progress or in-review row gets one liveness class (8.945669ms)
 ✔ a required row not done after 14 days is a violation; a legacy row uses genesis as its lower bound (0.42539ms)
 ✔ an ISO requiredSince, as `set required` writes it, ages from its UTC day; one that does not parse is a violation (0.251321ms)
 ✔ the text section always ends in a count and a result, and never prints a full pass (0.389998ms)
 ✔ pidAlive: a running pid is present, an exited one is gone, and EPERM still means present (30.271711ms)
 ✔ issue states: open list first, then the metric page, then at most 10 lookups (318.077146ms)
 ✔ a full open list: lookups settle what it leaves out, and only an unsettled issue keeps it undecided (419.682997ms)
 ✔ the open list refuses on a failed call or a bad record, and never echoes the helper (186.450359ms)
 ✔ a helper call past the deadline is killed with its child, and the call reports it (2007.63492ms)
 ✔ readQueue loads queue.json through the queue validator and refuses anything else (83.063941ms)
 ✔ protected changes list every in-range entry that changes a required or parked row (260.831405ms)
 ✔ the CLI prints the queue section above the weekly table and under a queue key in --json (522.539179ms)
 ✔ a queue with nothing wrong prints 0 violations and a reduced pass, never a full pass (222.158594ms)
 ✔ --no-issues makes no call and leaves the issue checks undecided; --no-queue skips the section (282.600961ms)
 ✔ the CLI refuses a bad queue before any call, and a failed open list with exit 2 (253.421151ms)
 ✔ --unsupported-runtime repeats once per seat and takes a seat name (327.815403ms)
 ✔ an unreadable config makes every owner invalid instead of passing them (132.587734ms)
 ℹ tests 78
 ℹ suites 0
 ℹ pass 78
 ℹ fail 0
 ℹ cancelled 0
 ℹ skipped 0
 ℹ todo 0
 ℹ duration_ms 11297.728644
@@ -0,0 +1,77 @@
 ✔ pure resolution selects current default or explicit enrolled account (2.079609ms)
 ✔ scope is explicit, bounded and never inferred (2.230143ms)
 ✔ fork pin is preserved against default change, override, missing account and revocation (0.849241ms)
 ✔ unenrolled account/provider, missing harness, model expansion and native model ceiling refuse (0.911311ms)
 ✔ only explicit synthetic credential forms and internal fixture stores admitted (6.641375ms)
 ✔ two concurrent workspaces of the same agent publish distinct complete private generations (61.909614ms)
 ✔ same execution ID is exclusively claimed and cannot overwrite a published generation (34.701654ms)
 ✔ failed generation after-auth preserves prior files, records failure and refuses blind same-ID retry (47.614211ms)
 ✔ failed generation before-publish preserves prior files, records failure and refuses blind same-ID retry (61.009733ms)
 ✔ credential lock contention refuses without duplicate side effects (12.128114ms)
 ✔ symlinked pre-existing final target is refused and never followed (27.71374ms)
 ✔ invalid registry cannot resolve; no fallback to supplied partial entries (0.296627ms)
 ✔ post-publication failure records uncertainty, preserves complete generation and prevents replay (32.444858ms)
 ✔ expired credentials refresh under transaction and subsequent generation reuses rotation (88.493926ms)
 ✔ refresh failure retains prior generation and store state (70.906183ms)
 ✔ refresh timeout retains prior generation and store state (150.598599ms)
 ✔ refresh malformed retains prior generation and store state (67.369629ms)
 ✔ concurrent refresh on same account refuses contention while unrelated account proceeds (217.751194ms)
 ✔ invalid refresh options refuse before burning claim (36.896252ms)
 ✔ fixed fake process rotates both OAuth fields without mutating caller input (35.708505ms)
 ✔ concurrent isolated processes preserve separate provider credentials (28.15465ms)
 ✔ fake failure is refused with fixed diagnostics (25.606812ms)
 ✔ fake malformed is refused with fixed diagnostics (24.03472ms)
 ✔ fake timeout is refused with fixed diagnostics (104.074695ms)
 ✔ fake unchanged is refused with fixed diagnostics (30.105534ms)
 ✔ caller executable/environment injection is rejected before spawning (0.380258ms)
 ✔ valid fixture tree validates and lists without secrets (78.529984ms)
 ✔ unknown-field refuses (0.383423ms)
 ✔ invalid-id refuses uppercase and traversal shapes (0.285878ms)
 ✔ plain-http baseUrl requires allowInsecureTransport (0.304645ms)
 ✔ native provider rejects allowInsecureTransport (0.134402ms)
 ✔ unsupported credential type and kind refuse (0.155521ms)
 ✔ account provider-path mismatch refuses (0.1092ms)
 ✔ profile account refs must be provider/account shaped (0.247616ms)
 ✔ seat selection accepts fork pin field, validates account refs (0.252071ms)
 ✔ harness manifest id must equal executable (gate 1) (0.217346ms)
 ✔ CLI validate: duplicate provider id across files refuses (30.167855ms)
 ✔ CLI validate: missing referenced provider/account refuse (34.544812ms)
 ✔ CLI validate: broken JSON refuses without secret echo (29.453698ms)
 ✔ CLI usage errors exit 2 (52.764154ms)
 ✔ credential.json sibling presence does not break validation and is never read (74.009646ms)
 ✔ D1 missing, empty and structurally empty roots refuse, no list projection (195.778149ms)
 ✔ D1 required directory auth cannot be absent (64.656927ms)
 ✔ D1 required directory auth/providers cannot be absent (67.184602ms)
 ✔ D1 required directory auth/accounts cannot be absent (65.462742ms)
 ✔ D1 required directory auth/settings cannot be absent (61.524711ms)
 ✔ D1 required directory harnesses cannot be absent (64.933926ms)
 ✔ D1 root file and unreadable metadata refuse (126.575629ms)
 ✔ D2 no symlink traversal at auth/providers/openai-codex.json (61.71521ms)
 ✔ D2 no symlink traversal at auth/accounts/openai-codex/homelab-openai (67.898272ms)
 ✔ D2 no symlink traversal at auth/providers (52.901333ms)
 ✔ D2 no symlink traversal at auth (57.489447ms)
 ✔ D2 root and ancestor symlinks and lexical traversal refuse (186.814266ms)
 ✔ private filesystem modes enforced for root (62.871372ms)
 ✔ private filesystem modes enforced for auth (55.863789ms)
 ✔ private filesystem modes enforced for auth/providers/openai-codex.json (56.804104ms)
 ✔ private filesystem modes enforced for auth/accounts/openai-codex/homelab-openai/account.json (63.768965ms)
 ✔ D3 numeric version 1 only across all record kinds (1.309762ms)
 ✔ D4 nested unknown keys and missing per-kind required fields refuse (65.515115ms)
 ✔ D5 unenrolled default refuses even when account exists (64.050411ms)
 ✔ D6 provider/account credential type must match (64.344785ms)
 ✔ D7 every harness endpoint enforces HTTP opt-in and shape (0.459451ms)
 ✔ D8 URLs reject embedded credentials and unsupported protocols without echo (165.076039ms)
 ✔ D9 malformed JSON diagnostics contain no content excerpt (64.218719ms)
 ✔ D10 missing metadata is missing-path, not invalid-json (62.667729ms)
 ✔ D10 library returns no partial entries on any invalid record (72.329981ms)
 ✔ null/scalar/array metadata refuses without stack or echo (254.035605ms)
 ✔ credential sibling is never opened, even when an unreadable symlink (35.816416ms)
 ✔ oversized metadata refuses before parsing (59.942545ms)
 ℹ tests 69
 ℹ suites 0
 ℹ pass 69
 ℹ fail 0
 ℹ cancelled 0
 ℹ skipped 0
 ℹ todo 0
 ℹ duration_ms 2279.776148
@@ -0,0 +1,158 @@
 ✔ F1: an ordinary commit after update-ref is refused until step 8; then it commits and the queue stays at C (1293.447642ms)
 ✔ F1: a plain `commit -e` whose guard ran before update-ref fails at its own HEAD update (1101.083362ms)
 ℹ git commit -e: index.lock free during the editor
 ✔ F1: a `commit -e -- path` whose guard ran before update-ref fails at its own HEAD update (1074.456052ms)
 ℹ git commit -e -- src.txt: index.lock held during the editor
 ✔ F1: step 8 with index.lock held exits 3, and ordinary commits stay refused until the printed command runs (1122.393406ms)
 ✔ F1: HEAD moving after the step-7 guard check and before update-ref: refused, nothing published (1503.485391ms)
 ✔ F1: H is recorded before the canary, so HEAD moving during the step-1 canary is refused at step 7 (1079.455381ms)
 ✔ F1: a queue commit landing after H is recorded: step 1 says HEAD moved, not the guard (1569.18684ms)
 ✔ F1: a queue commit landing between the HEAD check and the canary: the failed clean run is reported as HEAD moved (1158.207634ms)
 ✔ F1: a shared-index change during the procedure is not committed (1264.74922ms)
 ✔ F1: a queue path staged after update-ref: step 8 stops and touches nothing (1315.032514ms)
 ✔ F1: a missing or a different hook refuses (834.923829ms)
 ✔ F1: same bytes without the exec bit, a symlinked hook, and core.hooksPath in the local or global scope each refuse before update-ref (1402.70655ms)
 ✔ F1: the canary refuses a hook that git would not run (785.295278ms)
 ✔ F1: the guard deactivated after step 1 is refused at the step-7 recheck (1293.939835ms)
 ✔ bootstrap: implementation-only HEAD, the guard, genesis, the --genesis commit, then an extending commit (1068.985862ms)
 ✔ bootstrap: --genesis with a base present, no base without --genesis, an op before the first commit, a changed map, another branch (1328.258272ms)
 ✔ bootstrap: the archived tests and validator run outside any repository (1203.511963ms)
 ✔ general: an unrelated staged file stays staged, and the committed blobs are the snapshot bytes (1164.216663ms)
 ✔ general: a queue write after the snapshot is not committed (1266.8914ms)
 ✔ general: a snapshot whose log does not extend the base refuses (975.109162ms)
 ✔ general: install-hook privilege, repair of a same-bytes hook, and its refusals (207.935884ms)
 ✔ general: environment overrides, a linked worktree and usage (612.446006ms)
 ✔ general: a queue path staged before the run refuses at step 1 (711.225406ms)
 ✔ general: HEAD's queue tests failing in the archive refuse (959.330303ms)
 ✔ genesis document serializes deterministically and replays (4.317024ms)
 ✔ a hand edit that stays valid JSON fails replay; a formatting-only edit fails re-serialization (2.796549ms)
 ✔ a tampered result, receipt or viewSha fails replay (2.483399ms)
 ✔ op ids: 8 to 72 characters for callers, 80 in the log for .outcome entries (0.194516ms)
 ✔ add: defaults for an ordinary seat, privileged extras, refusals (4.962542ms)
 ✔ matrix: queued→briefed privileged; briefed→in-progress owner with after satisfied (6.010632ms)
 ✔ matrix: release, review round, changes requested and waiting-on-jason (12.885071ms)
 ✔ matrix J5: in-review→done by the gate owner with evidence naming the current round (8.050185ms)
 ✔ review issue, lead decision 23: none refuses, one is used, several need --issue, later rounds keep it (19.180382ms)
 ✔ the row schema refuses a round with a null issue, and the A1 review shape (P2) (2.426903ms)
 ✔ matrix R1: every state × target × actor class matches 8.7, gate owner jason or not, required or not (1537.958744ms)
 ✔ matrix: blocked keeps the claim and returns only to previousState (4.423046ms)
 ✔ matrix J4: parking is Jason's, refused while required; unpark returns to queued (2.809439ms)
 ✔ field edits: who may change what (5.358298ms)
 ✔ set issues keeps a logged narrowing of closes (N10) (2.937073ms)
 ✔ text the table shows refuses \ and <, everywhere it enters (N8) (2.297068ms)
 ✔ every accepted text renders to nine cells on every row (N8) (25.303514ms)
 ✔ genesis: the map refuses an owner among its row's reviewers; replay doesn't (2026-10-04) (0.775011ms)
 ✔ times and dates must be calendar values, not just the shape (2026-10-04) (1.766389ms)
 ✔ replay holds every op id to the caller's rule (N11) (5.241836ms)
 ✔ note: owner, listed reviewer or privileged; empty clears (1.262486ms)
 ✔ assign moves the claim with the owner; done clears it (2.793087ms)
 ✔ render is byte-stable and escapes pipes (0.549068ms)
 ✔ view classification: current, genuine stale, edited stale marker, changed current body, markers (0.642409ms)
 ✔ next: resume, then review, then start, then wait, then nothing; lowest id first (27.744196ms)
 ✔ canonical args make a retry's identity independent of list order (0.414978ms)
 ✔ manifests, headings and blob ids (0.632552ms)
 ✔ the migration map: one queue-map block, exact keys (0.804261ms)
 ✔ every call but `queue` reaches the seat CLI exactly as before A2 (731.541362ms)
 ✔ `queue` reaches the queue CLI with the rest of the arguments (238.216191ms)
 ✔ the pre-A2 fixture is the script A2 changed (0.287818ms)
 ✔ acquire publishes the record by link; release removes only its own lock (5.720988ms)
 ✔ a kill between the temp write and the link leaves no lock (45.276098ms)
 ✔ a short or failed temp write refuses and leaves no lock and no temp (2.721826ms)
 ✔ a link error other than EEXIST refuses (1.484864ms)
 ✔ an error after the link releases the lock: unreadable gate, failing temp stat (3.390013ms)
 ✔ a release that fails on a gate path is reported, never a stack trace (P1) (4.685306ms)
 ✔ a paused holder: another writer waits 10 s, then refuses naming it live (10092.593192ms)
 ✔ two concurrent unlockers: the second refuses on the gate (31.012372ms)
 ✔ a writer publishing during an unlock, lock first: unlock sees it live and refuses (2.190358ms)
 ✔ a writer publishing during an unlock, gate first: the writer releases and refuses (2.065014ms)
 ✔ a gate swapped while held is left in place and reported, on success and on refusal (N1) (2.789643ms)
 ✔ a reused pid within one boot is mismatch; unlock removes the lock and never signals the process (5.874995ms)
 ✔ the same pid and start on a different boot is mismatch (1.057776ms)
 ✔ a foreign host is unknown whatever the local pid says; unlock refuses (92.308672ms)
 ✔ unreadable /proc: classification is unknown and acquire refuses (1.295867ms)
 ✔ invalid records: empty, unparsable, wrong keys, bad start or boot (0.354373ms)
 ✔ a stale gate blocks writers; --check-gate says mismatch for a reused pid (6.94049ms)
 ✔ a delayed release by a dead owner, after unlock and a new owner: the inode check keeps the new lock (3.76974ms)
 ✔ release checks the inode too: a byte-identical lock file with a new inode is left in place (1.280141ms)
 ✔ unlock refuses a live, unknown or invalid lock, and does nothing without one (4.753956ms)
 ✔ the migration map validates and renders the golden genesis table (3.904784ms)
 ✔ the marked QUEUE.md holds every row and parked item between its markers (1.388114ms)
 ✔ map-check reports each kind of drift (5.03177ms)
 ✔ a request posts once as the requester; a retry sends nothing (644.508495ms)
 ✔ each transport answer maps to posted, failed or uncertain (8.9 step 3) (3439.609121ms)
 ✔ the pre-send checks: GET user must name the requester, under the deadline (1220.867052ms)
 ✔ the lead's request refuses a token for login sage (723.708575ms)
 ✔ the credential file: the seat's own, 0600, no symlink, never the shared default (493.172025ms)
 ✔ an unresolved request blocks a new request, a new round, waiting-on-jason and done (1765.486675ms)
 ✔ a same-op retry after a kill sends nothing, even with a stale view (2829.629069ms)
 ✔ a held lock at the outcome exits 3 and names what the transport said (628.916615ms)
 ✔ late outcomes: after an abandon, and after a resolve with the same or another id (1916.002633ms)
 ✔ resolve checks the comment: issue, markers, round, candidate and author (1924.486952ms)
 ✔ the lead resolves a seat's request: the comment's author is the requester, fetched with the lead's token (727.313694ms)
 ✔ validateRow checks a request round's shape, which every replayed entry must keep (507.664198ms)
 ✔ request, changes, a new candidate, approval: every round pinned; no review files (1729.129969ms)
 ✔ a row with no reviewers opens a round that posts nothing (1395.516036ms)
 ✔ verify-commit: a prospective tree must hold exactly the candidate's paths (1543.262043ms)
 ✔ semantics: v1 entries replay as before; review entries need v2 (363.36264ms)
 ✔ set reviewers refuses the row's owner (2026-09-28) (269.605903ms)
 ✔ the owner records no verdict, even as a listed reviewer (369.885082ms)
 ✔ a request comment over the length limit is not sent (375.006737ms)
 ✔ a late POST on a closed row leaves a conflict nothing can resolve, and resolve asks nothing (407.509135ms)
 ✔ a Jason-gated row reaches waiting-on-jason only on every reviewer's approval (1969.588199ms)
 ✔ genesis: refusals before anything is written (506.562641ms)
 ✔ genesis: the map must be committed, well formed, with committed briefs and seats (654.406808ms)
 ✔ genesis: markers, a stray witness, once only; a retry returns the receipt (550.263442ms)
 ✔ genesis: a file holding genesis alone with no witness is confirmed by sync or a retry (326.533418ms)
 ✔ canonical checks: worktree, second clone, detached HEAD, wrong branch, GIT_DIR, foreign code; a symlink works (716.581482ms)
 ✔ --by that differs from MOSAIC_AGENT_NAME warns on stderr and logs nothing more (N12) (596.068448ms)
 ✔ op ids: missing, too long, reserved; a retry answers; another payload refuses (655.337554ms)
 ✔ a retried add returns the id it first allocated, after reassignment and after done (725.218464ms)
 ✔ Rocko's S4 schedule: a lost result, another writer, then the retry opens no second round (693.687288ms)
 ✔ the review issue and the evidence round through the CLI (lead decision 23, 8.7) (1056.228626ms)
 ✔ claims and add defaults through the CLI; candidates are manifests or reachable commits (642.166551ms)
 ✔ add, set reviewers and assign refuse the row's owner as a reviewer (486.174853ms)
 ✔ the working-brief check: a changed working copy refuses the start and flags next (644.915802ms)
 ✔ next: resume first, then nothing for an idle seat; needs a seat (346.860331ms)
 ✔ view stale: new ops and verify refuse naming the unshown op; retries answer; reads warn; render fixes (705.897481ms)
 ✔ view unknown: a hand edit, an old marker over an edited body, missing or duplicate markers (1322.822023ms)
 ✔ a hand edit to queue.json refuses every verb, reads included (655.13831ms)
 ✔ verify and render --check leave bytes and mtimes unchanged (503.99518ms)
 ✔ render is byte-stable across runs and repositories (364.651361ms)
 ✔ snapshot and verify --snapshot (843.406358ms)
 ✔ usage errors exit 4 (649.156839ms)
 ✔ a short write, ENOSPC or a file fsync failure: nothing visible, temp removed (178.455612ms)
 ✔ a rename failure: nothing visible, temp removed (149.66643ms)
 ✔ a directory fsync failure: uncertain, exit 3, no receipt; the tail refuses new ops; a retry confirms (191.396309ms)
 ✔ a directory fsync failure, then sync names the op (303.146379ms)
 ✔ a witness write failure: uncertain, durable, exit 3; the view is untouched (163.931607ms)
 ✔ the .git fsync after the witness rename fails: uncertain, exit 3, the witness says so (130.884674ms)
 ✔ confirming a tail fsyncs queue.json and docs/plans before the witness; either failure changes nothing (162.482988ms)
 ✔ the docs/plans fsync after the view rename fails: the op stands, the view is written, a warning says so (133.645043ms)
 ✔ a lock swapped while held is left in place and reported, on a receipt and on a refusal (140.043364ms)
 ✔ a lock that cannot be released after an op is reported, on a receipt and on a refusal (154.777597ms)
 ✔ unlock prints a swapped gate's warning on stderr, the result on stdout (124.738055ms)
 ✔ a view write that fails keeps the op and reports a stale view (125.171976ms)
 ✔ SIGKILL before the rename: nothing recorded; the retry removes the leftover temp (586.757792ms)
 ✔ SIGKILL after the rename, before the witness: the tail refuses new ops and sync names the op (651.543058ms)
 ✔ SIGKILL after the witness, before the view: the stale refusal names the op (597.367324ms)
 ✔ SIGKILL after the view, before the receipt: the retry returns the receipt (585.535612ms)
 ✔ git checkout between steps 1 and 7: step 7 refuses and nothing is written (173.622174ms)
 ✔ git stash restoring an older valid pair: history lost; accept-history needs privilege, a reason and --yes (1210.247576ms)
 ✔ a deleted witness: refused after the locked recheck; accept-history records it absent (221.84185ms)
 ✔ a header edit during a write: the op stands, the view write is skipped with a warning (153.28568ms)
 ✔ a reader paused between the witness and the file while a writer finishes: no lost-history report (132.389949ms)
 ✔ file-then-witness order forced by a hook: the locked recheck prevents a false report (130.677054ms)
 ✔ a writer paused before and after the witness rename: readers see a tail, then a match (147.365771ms)
 ✔ a true rollback is reported only after the locked recheck; a held lock names its holder instead (498.356751ms)
 ✔ an accept-history in progress: an unlocked reader waits on the lock and never reports lost history (469.221844ms)
 ✔ the platform check refuses other filesystems (130.864218ms)
 ✔ tmpfs passes only a test layer that allows it (N5) (158.150412ms)
 ✔ unlock keeps a multi-line lock record on stdout (P3) (209.279676ms)
 ℹ tests 148
 ℹ suites 0
 ℹ pass 148
 ℹ fail 0
 ℹ cancelled 0
 ℹ skipped 0
 ℹ todo 0
 ℹ duration_ms 26358.293639
@@ -0,0 +1,27 @@
 ✔ resolveSeat: by name under --repo resolves the repo layout (1.481206ms)
 ✔ resolveSeat: by path resolves the fleet layout (0.362472ms)
 ✔ resolveSeat: refusals for missing dir, missing launch.sh, non-executable launch.sh, invalid name, and unknown layout (0.856131ms)
 ✔ tmuxContext: outside tmux, default socket, custom socket, and exec failure (0.828242ms)
 ✔ makeRegistration produces a record that validates; each shape violation throws SeatError (0.904445ms)
 ✔ writeRegistration/readRegistration: round trip, permissions, absence, and malformed records (1.554862ms)
 ✔ updateTask: changes task, taskSetBy and updatedAt only, and refuses appropriately (1.289573ms)
 ✔ CLI launch: registers, execs the fake launch script, and passes args through (32.787236ms)
 ✔ CLI launch: --harness lands in the record (32.828675ms)
 ✔ CLI launch: the launch script's own exit code passes through (28.31954ms)
 ✔ CLI launch: relaunching a seat rewrites the one registration record (64.395537ms)
 ✔ CLI launch: omitting --task records an empty string, not null (29.227222ms)
 ✔ CLI seat task: updates only the task after a launch, and refuses on an unlaunched seat (81.319556ms)
 ✔ CLI refusals: no args, unknown flag, missing config, already-registered env, and exec failure (127.788375ms)
 ✔ samePath: equal paths, symlinked dirs, distinct dirs, and non-strings (0.395671ms)
 ✔ resolveSetBy: explicit --by wins over the environment; absent or empty environment gives unknown; invalid explicit or environment values refuse with exit 4 (0.553605ms)
 ✔ validateRegistration/readRegistration: taskSetBy is optional; a record without it (written before #1511) still loads unchanged; an invalid one is refused; the version does not change (0.726541ms)
 ✔ updateTask: records setBy, preserves startedAt and every unrelated field, upgrades an old record in place only when the task is set, and replaces a previous attribution (10.854969ms)
 ✔ CLI seat task: --by beats MOSAIC_AGENT_NAME, the environment beats nothing, empty environment is unknown, invalid --by or environment refuses with exit 4 and leaves the record byte for byte (379.051082ms)
 ℹ tests 19
 ℹ suites 0
 ℹ pass 19
 ℹ fail 0
 ℹ cancelled 0
 ℹ skipped 0
 ℹ todo 0
 ℹ duration_ms 867.37898
@@ -0,0 +1,59 @@
 ✔ the boot config is checked before anything starts (25.484517ms)
 ✔ a business with no tracker entry refuses task verbs (18.675299ms)
 ✔ credential.expiring and .expired are recorded once per instance (44.712914ms)
 ✔ a token file that changes on disk records credential.changed (21.046939ms)
 ✔ autostart polls, reconciles and retries a startup the tracker was down for (41.67541ms)
 ✔ a refusal a restart must clear is not retried by the poll (12.489203ms)
 ✔ a poll that fires while two are queued is dropped (17.657755ms)
 ✔ close waits for a running verb and refuses one that has not started (45.913148ms)
 ✔ the bundled Vikunja is the pinned upstream image the runbook names (1.509112ms)
 ✔ every published port is on 127.0.0.1, and no secret is in the file (0.505503ms)
 ✔ the fake answers each route with the statuses and shapes Vikunja v2.7.0 sent (136.056128ms)
 ✔ the recorded task bodies pass the checks S3 applies to every read (0.687076ms)
 ✔ the client works against the fake over real HTTP with the platform fetch (111.070926ms)
 ✔ a correct install starts, and the first reconcile records tasks that already exist (39.319619ms)
 ✔ verbs refuse while a business is starting and after startup refused it (22.51381ms)
 ✔ startup refuses a token that can do more than its role needs (52.113844ms)
 ✔ startup refuses an unsupported version and flags an untested one (29.381676ms)
 ✔ startup refuses a board that the runbook did not install (42.307367ms)
 ✔ startup refuses a project the sync bot cannot read (7.346948ms)
 ✔ startup refuses a configured label the pm bot cannot see (9.464093ms)
 ✔ startup refuses an expired credential and a missing sync credential (18.02678ms)
 ✔ an unreachable tracker refuses with tracker-unavailable (8.588022ms)
 ✔ an edit in the UI is recorded once, with the fields that changed (98.406805ms)
 ✔ a move between open buckets is seen on the board, though updated does not change (45.2568ms)
 ✔ a person's comment is counted and a bot's is not (78.116956ms)
 ✔ the hourly reconcile catches a comment through comment_count (81.889781ms)
 ✔ a task closed in the UI leaves the open view with its done bucket (129.610061ms)
 ✔ a task that leaves the board is recorded as deleted, moved or out of reach (82.047323ms)
 ✔ a poll that read before a verb wrote does not overwrite the verb (41.528925ms)
 ✔ a tracker fault during a tick is reported and the next tick catches up (37.749804ms)
 ✔ a malformed answer refuses the tick with tracker-shape (15.198025ms)
 ✔ no token value reaches the database, the log or a refusal (47.096688ms)
 ✔ the first look at a task counts only comments inside the window (36.09698ms)
 ✔ task.create needs a recorded human request and a requirement id (60.738266ms)
 ✔ only labels named in the business file can be written (36.345449ms)
 ✔ task.schedule sets and clears a due date and relations (43.647102ms)
 ✔ assign and reassign move the role bots and record task.assigned (62.506333ms)
 ✔ task.update.assigned is for the assignee and records task.state (54.68988ms)
 ✔ a wrong expected digest records task.conflict and writes nothing (20.478626ms)
 ✔ a cross-role verb needs a resolved decision, used once (32.256943ms)
 ✔ task.close needs a verdict; after it every verb refuses with task-done (33.220223ms)
 ✔ a lost answer is settled by a re-read and never retried (38.272731ms)
 ✔ a create whose answer is lost is reported uncertain, and the poll finds the task (44.567161ms)
 ✔ a task the sync bot cannot read refuses and records nothing (27.215388ms)
 ✔ verbs and polls for one business run one at a time (99.301469ms)
 ✔ a due date with milliseconds is written to the second (68.906673ms)
 ✔ every write landed and the final read failed: the verb succeeds and records what it wrote (38.970276ms)
 ✔ some writes landed and the final read failed: write-uncertain, and nothing is recorded (14.496191ms)
 ✔ a create whose final read fails succeeds and records task.created (34.358337ms)
 ✔ an edit between the last write and the final read shows as external on the next poll (39.988391ms)
 ✔ task.created is recorded when a later label write fails (16.768111ms)
 ℹ tests 51
 ℹ suites 0
 ℹ pass 51
 ℹ fail 0
 ℹ cancelled 0
 ℹ skipped 0
 ℹ todo 0
 ℹ duration_ms 943.604698
@@ -0,0 +1,23 @@
 ✔ browser edge states: loading, empty, malformed, stale, hostile/long values, in-flight reply and appearance fallback (3375.0441ms)
 Rendered contrast: {"failures":[],"count":330,"lowest":4.504658476260286}
 ✔ served Console browser: real board fixtures, keyboard, drafts, receipts, themes, 320px and failures (3792.760217ms)
 ✔ conversation view: full history, collapsed tools, hidden thinking, inert hostile content, malformed and reconcile markers (3283.267961ms)
 ✔ conversation view: a fork keeps the open branch, says so, and opens the new one on request (1639.235309ms)
 ✔ conversation view: a newer session with no readable history keeps the marker (919.340475ms)
 ✔ conversation view: seats without history say so and offer no reply (542.697818ms)
 ✔ Discord row through real board/WebUI: independent brake/liveness, no Reply, literal content (2826.548118ms)
 ✔ return flow through the conversation view: send, tool call, delayed result, peer message, exact long answers, relaunch (54743.791523ms)
 ✔ both presentations replace old activity with relaunch notice, label retained history, then resume after new activity (2827.586217ms)
 ✔ reported return flow and relative Age: reply sent from the inspector, then the new answer appears there without manual refresh (22597.160796ms)
 ✔ loopback host and board origin fail closed (7.216403ms)
 ✔ real board fixture passes through WebUI; assets and isolated seen/reply work (94.784523ms)
 ✔ proxy preserves exact request bytes, status and receipt, rejects forms and malformed JSON, never follows redirect (66.672652ms)
 ✔ unreachable board reports URL; CLI rejects unsupported options (597.716602ms)
 ℹ tests 14
 ℹ suites 0
 ℹ pass 14
 ℹ fail 0
 ℹ cancelled 0
 ℹ skipped 0
 ℹ todo 0
 ℹ duration_ms 55083.850563
@@ -0,0 +1,17 @@
 OK    status with missing harness credential exits 3 and still lists accounts
 OK    status reports harness credential (read-only) + mosaic accounts
 OK    api key material never reaches output
 OK    oauth token material never reaches output
 OK    unparseable credential file exits 2
 OK    symlinked credential file exits 4
 OK    env-side credential names reported
 OK    env var values never reach output
 OK    accounts without an accounts dir reports none and creates nothing
 OK    accounts lists files and marks the active one
 OK    loose account perms flagged in listing
 OK    agent --auth with missing account file refuses (exit 4)
 OK    agent --auth with non-0600 account file refuses
 OK    agent --auth with invalid account name refuses
 OK    auth.sh without valid config refuses
 selftest: 15 passed, 0 failed
@@ -0,0 +1,55 @@
 Note: switching to '521597bbe0cf51a0a1aa3b7a16f13fc7b59c98c6'.
 You are in 'detached HEAD' state. You can look around, make experimental
 changes and commit them, and you can discard any commits you make in this
 state without impacting any branches by switching back to a branch.
 If you want to create a new branch to retain commits you create, you may
 do so (now or later) by using -c with the switch command. Example:
  git switch -c <new-branch-name>
 Or undo this operation with:
  git switch -
 Turn off this advice by setting config variable advice.detachedHead to false
 Not currently on any branch.
 nothing to commit, working tree clean
 Note: switching to '521597bbe0cf51a0a1aa3b7a16f13fc7b59c98c6'.
 You are in 'detached HEAD' state. You can look around, make experimental
 changes and commit them, and you can discard any commits you make in this
 state without impacting any branches by switching back to a branch.
 If you want to create a new branch to retain commits you create, you may
 do so (now or later) by using -c with the switch command. Example:
  git switch -c <new-branch-name>
 Or undo this operation with:
  git switch -
 Turn off this advice by setting config variable advice.detachedHead to false
 OK    dry-run: allowed change, exit 0, nothing committed (exit 0)
 OK    dry-run committed nothing
 OK    apply: allowed change exits 0 (exit 0)
 OK    apply: attribution in commit subject
 OK    apply: target tree clean after commit
 OK    disallowed path refused (exit 1)
 OK    disallowed path: target untouched
 OK    syntax gate refused broken .mjs (exit 1)
 OK    syntax gate: target untouched
 OK    suite failure refused (exit 1)
 OK    suite failure: target reverted to clean
 OK    disabled policy refused (exit 2)
 OK    disabled policy: target untouched
 OK    failed run refused (exit 1)
 OK    failed run: target untouched
 OK    missing run exits 4 (exit 4)
 OK    invalid policy exits 2 (exit 2)
 selftest: 17 passed, 0 failed
@@ -0,0 +1,26 @@
 OK    absent adapter defaults to pi
 OK    adapter mock validates (exit 0)
 OK    unsupported adapter exits 2 (exit 2)
 OK    env exports adapter
 OK    bootstrap creates default when absent (exit 0)
 OK    bootstrap wrote config file
 OK    bootstrap is idempotent on existing config (exit 0)
 OK    bootstrap did not rewrite existing config
 OK    validate missing config exits 3 (exit 3)
 OK    malformed JSON exits 2 (exit 2)
 OK    unsupported configVersion exits 2 (exit 2)
 OK    unknown top-level key exits 2 (exit 2)
 OK    unknown execution key exits 2 (exit 2)
 OK    unsupported backend exits 2 (exit 2)
 OK    unsupported environment exits 2 (exit 2)
 OK    relative dataRoot exits 2 (exit 2)
 OK    non-canonical dataRoot exits 2 (exit 2)
 OK    filesystem root dataRoot exits 2 (exit 2)
 OK    home directory dataRoot exits 2 (exit 2)
 OK    dataRoot containing config dir exits 2 (exit 2)
 OK    control character in provider exits 2 (exit 2)
 OK    symlinked config file exits 2 (exit 2)
 OK    env exports resolve correctly
 OK    failed validation modified nothing
 selftest: 24 passed, 0 failed
@@ -0,0 +1,70 @@
 toolchain: node v26.8.1
 OK    syntax: packages/discord/src/approvals.mjs
 OK    syntax: packages/discord/src/authorize.mjs
 OK    syntax: packages/discord/src/binding.mjs
 OK    syntax: packages/discord/src/cli.mjs
 OK    syntax: packages/discord/src/connector.mjs
 OK    syntax: packages/discord/src/context.mjs
 OK    syntax: packages/discord/src/engine-pi.mjs
 OK    syntax: packages/discord/src/errors.mjs
 OK    syntax: packages/discord/src/gateway.mjs
 OK    syntax: packages/discord/src/git.mjs
 OK    syntax: packages/discord/src/journal.mjs
 OK    syntax: packages/discord/src/notify.mjs
 OK    syntax: packages/discord/src/rest.mjs
 OK    syntax: packages/discord/src/setspark.mjs
 OK    syntax: packages/discord/src/tools.mjs
 OK    syntax: packages/discord/src/web.mjs
 OK    syntax: packages/discord/bin/git-credential.mjs
 OK    syntax: packages/discord/extension/tools.mjs
 OK    syntax: packages/discord/tests/approvals.test.mjs
 OK    syntax: packages/discord/tests/authorize.test.mjs
 OK    syntax: packages/discord/tests/binding.test.mjs
 OK    syntax: packages/discord/tests/connector.test.mjs
 OK    syntax: packages/discord/tests/context.test.mjs
 OK    syntax: packages/discord/tests/engine.test.mjs
 OK    syntax: packages/discord/tests/fake-pi.mjs
 OK    syntax: packages/discord/tests/gateway.test.mjs
 OK    syntax: packages/discord/tests/git.test.mjs
 OK    syntax: packages/discord/tests/helpers.mjs
 OK    syntax: packages/discord/tests/journal.test.mjs
 OK    syntax: packages/discord/tests/notify.test.mjs
 OK    syntax: packages/discord/tests/recover.test.mjs
 OK    syntax: packages/discord/tests/rest.test.mjs
 OK    syntax: packages/discord/tests/setspark.test.mjs
 OK    syntax: packages/discord/tests/tools.test.mjs
 OK    syntax: packages/discord/tests/web.test.mjs
 OK    syntax: packages/discord/fixtures/claim-worker.mjs
 OK    syntax: packages/discord/fixtures/legacy-owner-worker.mjs
 OK    syntax: scripts/discord.sh
 OK    syntax: scripts/discord-service.sh
 OK    packages/discord declares no dependencies
 OK    no bot-token-shaped string in packages/discord
 OK    fixture binding uses placeholder ids only
 OK    fixture binding validates
 OK    real pi with the extension exposes exactly list_dir, read_file, search and no built-in tool
 OK    real pi with a writable root exposes exactly the three reads plus write_file and edit_file, and writes nothing at start
 OK    real pi with a web key exposes the three reads plus web_fetch and web_search, and no write tool without a writable root
 OK    real pi with a git root exposes the reads, writes and the four git verbs, commits nothing at start, and never shows the token
 OK    real pi with protocol vault adds reserve_id to the git verbs
 OK    real pi with a setspark key exposes the reads and the eight record verbs, no counters, and never shows the key
 OK    real pi refuses a git key on a read-only root (fail closed)
 OK    real pi with the pilot flags (--no-tools) exposes no tool at all
 OK    real pi exits non-zero without MOSAIC_DISCORD_TOOLS: no session, no tools (fail closed)
 OK    a failing nested test fails the run under a parent runner's NODE_TEST_CONTEXT
 OK    node --test packages/discord/tests/ (ℹ pass 178)
 OK    scripts/discord.sh --help exits 0
 OK    scripts/discord.sh check without a binding exits 4
 OK    scripts/discord.sh recover without a binding exits 4
 OK    scripts/discord.sh reload without a binding exits 4
 OK    scripts/discord-service.sh without a command exits 4
 OK    service unit renders with the repository path, a supervised run as the main process, exit 3 never retried, and reload as SIGHUP
 OK    service install writes the rendered unit (0644) and leaves no temp file
 OK    service install a second time reports unchanged
 OK    systemd-analyze verify accepts the rendered unit
 OK    service uninstall removes the unit file
 OK    service install with an unknown flag exits 4
 OK    service install with USER unset finishes and names the account for lingering
 discord suite: 66 passed, 0 failed
@@ -0,0 +1,21 @@
 OK    initial ordinary-file install
 OK    installed tree matches canonical source
 OK    installed tree has no symlinks
 OK    check detects installation drift
 OK    sync refuses to overwrite installation drift
 OK    check detects an extra destination file
 OK    check detects an extra destination directory
 OK    check rejects a destination symlink
 OK    sync accepts a canonical source update
 OK    updated installation matches canonical source
 scripts/test-extension-package.sh: line 14: 38232 Killed                     "$@" > /dev/null 2>&1
 OK    forced interruption kills the replacing process
 OK    next invocation recovers old consistent installation
 OK    interrupted replacement rolled back
 OK    sync succeeds after interruption recovery
 OK    unlocked stale lock file does not block
 OK    active lock refuses a concurrent sync
 OK    source symlink fails closed
 OK    nested second entrypoint fails closed
 extension package selftest: 18 passed, 0 failed
@@ -0,0 +1,53 @@
 toolchain: node v26.8.1, python 3.12.8, jsonschema 4.26.0
 OK    syntax: scripts/foundation-inspect.mjs
 OK    syntax: scripts/foundation/strict-json.mjs
 OK    syntax: scripts/foundation/canonical.mjs
 OK    syntax: scripts/foundation/resolve.mjs
 OK    syntax: scripts/foundation/validate-record.mjs
 OK    syntax: scripts/foundation/fixtures/build-fixtures.mjs
 OK    syntax: scripts/foundation/canonical.test.mjs
 OK    syntax: scripts/foundation/cli.test.mjs
 OK    syntax: scripts/foundation/fixtures.test.mjs
 OK    syntax: scripts/foundation/resolve.test.mjs
 OK    syntax: scripts/foundation/strict-json.test.mjs
 OK    syntax: scripts/foundation/verify-schema.py (ast only; no bytecode written)
 OK    fixture generator runs
 OK    checked-in fixtures/bundles equal a fresh generation
 OK    checked-in fixtures/raw equal a fresh generation
 OK    checked-in fixtures/index.json equal a fresh generation
 OK    checked-in demo bundles equal a fresh generation
 OK    a failing nested test fails the run under a parent runner's NODE_TEST_CONTEXT
 OK    node --test scripts/foundation/ (ℹ pass 80)
 OK    differential schema oracle: PASS: differential schema oracle (finite corpus; compatibility evidence, not equivalence proof)
      platform witness: strftime('%Y') for year 999 -> '999' (pinned checker refuses years 0001..0999)
      node v26.8.1; corpus 1568 records (38 pinned fixtures, 478 unique bundle records, 1052 typeCase/mutation/lexical cases)
      schema column: agree-valid 540, agree-invalid 991, DISAGREEMENTS 0; strict-only (parser-bound) cases: 27; unsupported-kind records not schema-assessed by the inspector: 10
      profile column (schema-valid records only): profile-valid 510, profile-invalid 30
      profile refusals asserted: 30 schema-agreed-valid records refused only by the strict typed-string profile (rule profile-pattern-mismatch), 12 declared by name; 73 named probes verified against declared schema/profile columns
 OK    oracle: zero schema-column disagreements with the pinned checker
 OK    oracle: strict-only profile refusals are counted and asserted
 OK    demo: permitted read preview exits 0 (exit 0)
 OK    demo: permitted file.change preview exits 0 (exit 0)
 OK    demo: assignment.change proposal is unresolved (exit 3) (exit 3)
 OK    demo: revoked registration is refused (exit 3) (exit 3)
 OK    demo: message is not authority (exit 3) (exit 3)
 OK    usage: no arguments exits 2 (exit 2)
 OK    io: missing file exits 4 (exit 4)
 OK    io: directory exits 4 (exit 4)
 OK    io: symlink exits 4 (O_NOFOLLOW) (exit 4)
 OK    bound: oversize fixture exits 2 (exit 2)
 OK    profile: one final LF in a typed selection id is refused before admission (exit 2) (exit 2)
 OK    profile: two final LFs fail the schema pattern itself (exit 2) (exit 2)
 OK    profile: escaped newlines in free-form text stay allowed (exit 0) (exit 0)
 OK    profile refusal is invalid-request/profile-pattern-mismatch with selection and operation withheld, value not echoed
 OK    text output starts with the disclaimer
 OK    json output is valid JSON with result allowed and exactly the charter §7 fields
 OK    json golden matches byte-for-byte
 OK    sandboxed bundle run (env -i, PATH=/nonexistent) produced the unresolved proposal
 OK    sandbox inventory (path/type/size/mode/uid/gid/inode/mtime/sha256) unchanged by runs
 OK    canary never printed (bundle run and credential-file run)
 OK    a non-bundle JSON file is refused at the shape gate, not read into output
 OK    no field of the non-bundle file is echoed
 selftest: 44 passed, 0 failed
@@ -0,0 +1,35 @@
 toolchain: node v26.8.1, git version 2.55.0
 OK    syntax: packages/queue/src/cli.mjs
 OK    syntax: packages/queue/src/errors.mjs
 OK    syntax: packages/queue/src/io.mjs
 OK    syntax: packages/queue/src/lock.mjs
 OK    syntax: packages/queue/src/queue.mjs
 OK    syntax: packages/queue/src/review.mjs
 OK    syntax: packages/queue/src/store.mjs
 OK    syntax: packages/queue/tests/commit.test.mjs
 OK    syntax: packages/queue/tests/data.test.mjs
 OK    syntax: packages/queue/tests/dispatch.test.mjs
 OK    syntax: packages/queue/tests/helpers.mjs
 OK    syntax: packages/queue/tests/lock.test.mjs
 OK    syntax: packages/queue/tests/migration.test.mjs
 OK    syntax: packages/queue/tests/review.test.mjs
 OK    syntax: packages/queue/tests/store.test.mjs
 OK    syntax: packages/queue/tests/write.test.mjs
 OK    syntax: packages/queue/tests/fixtures/fake-gitea.mjs
 OK    syntax: packages/queue/tests/fixtures/kill-at.mjs
 OK    syntax: packages/queue/tests/fixtures/lock-child.mjs
 OK    syntax: packages/queue/tests/fixtures/mosaic-pre-a2.sh
 OK    syntax: scripts/queue-commit.sh
 OK    syntax: scripts/git-hooks/pre-commit
 OK    syntax: scripts/mosaic
 OK    queue-commit.sh, the guard and scripts/mosaic are executable
 OK    packages/queue declares no dependencies
 ℹ tests 148
 ℹ pass 148
 ℹ fail 0
 OK    node --test packages/queue/tests/
 OK    scripts/mosaic queue help
 skip  queue verify and render --check: this checkout (/home/jwoltje/filbert-scratch/r45/base) is not the queue's canonical root (/mnt/storage/src/mosaic-stack)
 queue suite: 27 passed, 0 failed
@@ -0,0 +1,7 @@
 OK    valid RELEASE resolves (exit 0)
 OK    invalid RELEASE exits 1 (exit 1)
 OK    missing RELEASE exits 1 (exit 1)
 OK    valid RELEASE leaves image tag consistent with version
 skip state-machine cases (docker daemon unavailable)
 selftest: 4 passed, 0 failed
@@ -0,0 +1,33 @@
 OK    valid task validates (exit 0)
 OK    unknown task key exits 2 (exit 2)
 OK    unsupported taskVersion exits 2 (exit 2)
 OK    invalid task id exits 2 (exit 2)
 OK    empty prompt exits 2 (exit 2)
 OK    NUL in expectExact exits 2 (exit 2)
 OK    out-of-range timeout exits 2 (exit 2)
 OK    missing mission file exits 4 (exit 4)
 OK    task with valid mission validates (exit 0)
 OK    invalid mission exits 2 (exit 2)
 OK    validate missing task exits 4 (exit 4)
 OK    validation does not modify the task file
 OK    prune dry-run exits 0 (exit 0)
 OK    dry-run deleted nothing
 OK    prune --keep=2 --yes removes oldest (exit 0)
 OK    kept exactly 2 newest runs
 OK    newest run kept, oldest pruned
 OK    append-only receipt written (3 entries)
 OK    sessions/workspaces untouched by prune
 OK    prune with invalid keep exits 4 (exit 4)
 skip adapter seam cases (docker daemon unavailable)
 skip workspace/capability cases (docker daemon unavailable)
 skip live task cases (docker unavailable)
 OK    onboard without name exits 4 (non-interactive) (exit 4)
 OK    onboard --name renders profile (exit 0)
 OK    profile written
 OK    canon structure: required filled, optional placeholdered
 OK    canon sections present
 FAIL user recall run succeeds (exit 1)
 FAIL recalled user name (response: )
 OK    no agent identity on headless run
 selftest: 26 passed, 2 failed
@@ -0,0 +1,75 @@
 ✔ launch identity is stamped, payload identity is refused and stale holder cannot send (22.380445ms)
 ✔ decision classes route from policy; gated resolution is human-only, choice and target must match (26.652343ms)
 ✔ claim exclusion, holder release, gated revoke and rerouting to a new holder are atomic (26.817512ms)
 ✔ launch events require a human CLI capability; generic emit cannot forge authority events (16.912593ms)
 ✔ within-role decisions close atomically and invalid options or blocking omissions refuse (17.484361ms)
 ✔ observer capabilities read human inbox but cannot mutate or forge launch identity (14.390004ms)
 ✔ task action subjects and linked decision trail are complete and ordered (22.844639ms)
 ✔ launch binding is durable and reconnecting requires the identical trusted record (8.901475ms)
 ✔ business isolation includes inherited object names and cross-business message references (14.697023ms)
 ✔ authority never transfers between action, run, target, unresolved or replaced role holder (18.867203ms)
 ✔ task projection uses schema current view, skipping earlier and equal-start polls (16.611621ms)
 ✔ revocation permanently bars the old run from reclaiming first, including after broker restart (17.131669ms)
 ✔ empty message references refuse before storage; refusal-evidence failure stays a typed error (15.160447ms)
 ✔ both arbiters require human resolution when their cross-role route is themselves (21.001418ms)
 ✔ S1 adapter takes resolved limits and refs, rejects mismatched instance, never mutates input (2.412021ms)
 ✔ only validated broker references load; returned data and exceptions cannot expose a known token (4.936616ms)
 ✔ bad file modes, symlinks, repository/data paths, malformed tokens and missing dates refuse (2.671535ms)
 ✔ expiry refuses use and env references never become client data (0.864599ms)
 ✔ S1 parsed service refs work, service mismatch refuses, Gitea rotation due is a warning state (1.892674ms)
 ✔ opaque tokens shorter than 16 characters refuse before use (0.544106ms)
 ✔ human proof binds CLI entry, process start and nonce; agents and incomplete ancestry refuse (2.134541ms)
 ✔ process reader gets own kernel identity without exposing environment values (1.32011ms)
 ✔ EACCES ancestor environments skip only markers; commands and registered launches still refuse (1.011074ms)
 ✔ real pid 1 remains inspectable when its environment is protected (0.390425ms)
 ✔ within-role sends cite an open gated launch decision without spending it or naming it in grants (28.53858ms)
 ✔ missing and foreign-business citations refuse and roll back message and grant (18.196485ms)
 ✔ cross-role sends still need a matching resolved decision and consume it once (25.135814ms)
 ✔ broker process binds trusted launches, offers reader capabilities, refuses human mutation, closes cleanly (79.108955ms)
 ✔ startup token refusal returns safe code without value or partial listening broker (48.829337ms)
 ✔ loaded fixture token is absent from socket replies and SQLite, including refusal evidence (60.908712ms)
 ✔ killed broker leaves an explicit stale lock; another process cannot silently reclaim it (107.873418ms)
 ✔ trusted host registers later launches; socket clients never have a registration verb (56.981582ms)
 ✔ runtime excludes declared project roots even when host supplies no repoRoots (42.470637ms)
 ✔ a refused launch binding leaves the broker and existing capabilities alive; bad protocol stops it (60.65868ms)
 ✔ v3b prototype refusals, views and append-only mutations (80.57291ms)
 ✔ gated approval authorizes once, survives store reopen, and fresh approval works (28.953415ms)
 ✔ another run cannot consume an approval; a failed check leaves it usable (21.466123ms)
 ✔ two scheduled callers have exactly one grant and one consumed refusal (17.097423ms)
 ✔ failed commit rolls consumption back; cross-role consumes and within-role stays reusable (31.814084ms)
 ✔ class drift gated to cross-role refuses before consumption (20.59826ms)
 ✔ class drift cross-role to gated refuses before consumption (18.052394ms)
 ✔ class drift gated to within-role refuses before consumption (15.486208ms)
 ✔ class drift cross-role to within-role refuses before consumption (18.124738ms)
 ✔ class drift within-role to gated refuses before consumption (13.419772ms)
 ✔ class drift within-role to cross-role refuses before consumption (20.491545ms)
 ✔ message.send consumes approval and prevents a later send or authorize (18.817052ms)
 ✔ role.revoke consumes approval and prevents a later revoke or authorize (25.541538ms)
 ✔ creates private WAL store and excludes a second writer until explicit close (18.268778ms)
 ✔ rollback is atomic and schema metadata is checked against trusted DDL, not just itself (17.236638ms)
 ✔ existing empty database and symlink runtime directory refuse, never initialize over damage (15.924833ms)
 ✔ crash during a transaction recovers no partial event after explicit fixture-only lock removal (58.344101ms)
 ✔ writer refuses mixed at/read_at forms atomically, even through trusted SQL helpers (9.566566ms)
 ✔ async transactions refuse before invoking their function (7.56334ms)
 ✔ recordTask keeps sync reads and a role write apart (23.135531ms)
 ✔ read_at must be one canonical UTC format, so the projection compares strings safely (10.143766ms)
 ✔ a bad entry refuses the whole record (10.400234ms)
 ✔ taskView reads the projection for one business (14.293611ms)
 ✔ requestTask hands only a holder and a task verb to the handler, and records refusals (24.359955ms)
 ✔ the server sends task verbs to the adapter with its own timeout; other verbs stay synchronous (275.793239ms)
 ✔ without an adapter the server refuses every task verb (16.577329ms)
 ✔ the runtime refuses an invalid adapter and closes a valid one (17.478752ms)
 ✔ the process loads the S3 adapter from plain-data trackers (115.152403ms)
 ✔ socket capability stamps launch identity; shared views use wire, no SQL client (28.115562ms)
 ✔ two wire claims serialize; a lost reply never automatically retries (19.883232ms)
 ✔ malformed, oversized and identity-forging envelopes refuse without echoing input (12.093017ms)
 ✔ client preserves UTF-8 when a response divides a multibyte character (12.273179ms)
 ✔ committed mutation followed by dropped reply reports unknown and is never retried (18.390612ms)
 ℹ tests 67
 ℹ suites 0
 ℹ pass 67
 ℹ fail 0
 ℹ cancelled 0
 ℹ skipped 0
 ℹ todo 0
 ℹ duration_ms 596.109045
@@ -0,0 +1,68 @@
 ✔ config directory and file path follow MOSAIC_CONFIG (2.091087ms)
 ✔ the fixture business validates and comes back frozen (5.321521ms)
 ✔ two instances may share a definition (2.037789ms)
 ✔ top-level refusals (6.351726ms)
 ✔ arbiters and projects (8.005976ms)
 ✔ role instances (3.893541ms)
 ✔ Vikunja bots (8.360818ms)
 ✔ a role without Vikunja takes no tracker block (3.148697ms)
 ✔ credential references match the definition's services (3.774244ms)
 ✔ launch (9.004305ms)
 ✔ loadBusiness: file checks (1.763435ms)
 ✔ loadBusiness: not a regular file (39.742587ms)
 ✔ loading writes nothing (1.395759ms)
 ✔ names that are Object.prototype properties don't count as declared (2.537876ms)
 ✔ the shipped example refuses as written and validates once filled in (0.542455ms)
 ✔ usage errors exit 4 (305.032324ms)
 ✔ validate: a good business exits 0 and prints instance digests (67.820644ms)
 ✔ validate: project files (326.103036ms)
 ✔ validate: missing files and a broken system config (241.613221ms)
 ✔ validate: credential reference problems exit 2 and name each one (65.283695ms)
 ✔ validate: a token file inside the repository is refused (65.265828ms)
 ✔ validate: role definitions come from MOSAIC_ROLES_DIR (200.706934ms)
 ✔ resolve: prints one instance's record (200.999245ms)
 ✔ resolve: refusals (402.197715ms)
 ✔ parse: exactly one of file or env, plus the service's date (3.793344ms)
 ✔ check: a good file has no problems (1.913368ms)
 ✔ check never opens the file: a write-only token passes (0.714324ms)
 ✔ check: file problems (1.363191ms)
 ✔ check: token files can't live in the repository or dataRoot, even through a linked directory (1.906015ms)
 ✔ check: dates and environment references (0.655015ms)
 ✔ path and load (3.164846ms)
 ✔ refusals (1.988671ms)
 ✔ systemVars flattens the validated config (2.92044ms)
 ✔ precedence: system, business, project, project role, agent (7.214308ms)
 ✔ limits narrow the definition and never widen it (3.430914ms)
 ✔ role.launch stays within-role only for the instance the launch block names (6.183548ms)
 ✔ limits.authority without role.launch leaves the launcher with no launch block (1.860412ms)
 ✔ limits.authority narrows cross-role actions too (1.009894ms)
 ✔ classify (1.115549ms)
 ✔ the record carries what the broker and launcher need (0.857858ms)
 ✔ digest: key order doesn't matter, any value change does (6.307397ms)
 ✔ refusals (2.303527ms)
 ✔ the four shipped version 2 roles load (3.823656ms)
 ✔ shipped role scopes match addendum B section 2 and the SR runbook (2.018995ms)
 ✔ shipped authority follows the note's table (1.408237ms)
 ✔ version 1 files keep loading with no authority (1.257812ms)
 ✔ the conductor policy isn't a role (0.384142ms)
 ✔ a missing role file is exit 4, a symbolic link too (0.568308ms)
 ✔ version 2 refusals (1.861931ms)
 ✔ authority: closed vocabulary, no gated-only action, no overlap (3.158576ms)
 ✔ credentials: Gitea scopes (1.219846ms)
 ✔ credentials: Vikunja scopes are a group-to-verbs map from the grantable list (1.561772ms)
 ✔ credentials: services (0.559046ms)
 ✔ contract: a non-empty regular Markdown file beside the role file (0.63259ms)
 ✔ every key names known layers and a merge rule (1.216219ms)
 ✔ unknown keys and wrong layers refuse (1.149061ms)
 ✔ types (2.600103ms)
 ✔ merge: defaults, then the most specific layer wins (0.416272ms)
 ✔ merge: limits only narrow, and provenance lists each source (0.478389ms)
 ✔ merge doesn't change its inputs (0.197768ms)
 ℹ tests 60
 ℹ suites 0
 ℹ pass 60
 ℹ fail 0
 ℹ cancelled 0
 ℹ skipped 0
 ℹ todo 0
 ℹ duration_ms 1954.930287
@@ -0,0 +1,70 @@
 ✔ inbox lists only decisions routed to the human, with what approving authorizes and how to decide (20.134056ms)
 ✔ decide resolves by id prefix with --yes, and the trail shows it in broker order with the task to follow (26.241572ms)
 ✔ decide refuses without a terminal or --yes, on an unknown option and on a short reference (15.711639ms)
 ✔ decide prints a declining choice as declining (13.380758ms)
 ✔ an unknown outcome is reported once and never resent (9.236045ms)
 ✔ a decision closed before the answer arrives exits 2 and points at its trail (9.554273ms)
 ✔ a prefix that matches two open decisions exits 2 and resolves neither (10.621801ms)
 ✔ without --business a command uses the live host's business, and a stale host.json is not a host (11.366682ms)
 ✔ every human command refuses inside an agent run before it touches the bus (12.082916ms)
 ✔ usage errors exit 4; no business and no host is a usage error (11.128705ms)
 ✔ agents and tasks print through the broker (10.449874ms)
 ✔ notify.json: missing, loose, malformed and extra keys refuse; a binding or null passes (1.909287ms)
 ✔ bootConfig builds the broker's boot message for one business, with no trackers key when no project names one (48.708686ms)
 ✔ trackers come from the tracker.* variables of the one project that names a tracker project (39.925021ms)
 ✔ with two projects, the one that sets tracker.project supplies the tracker and the other is no conflict (35.254166ms)
 ✔ two projects that each name a tracker project refuse, since the boot shape holds one (33.529767ms)
 ✔ a business without tracker.baseUrl gets no trackers entry (41.164155ms)
 ✔ an unknown business and a broken system config refuse with exit 3 (68.437879ms)
 ✔ empty views say so (0.935675ms)
 ✔ the trail keeps the broker's order and names a decision's task without its rows (0.990128ms)
 ✔ tasks print the tracker fields the snapshot carries (0.159053ms)
 ✔ the host boots the broker, binds a launch in process, and the notifier DMs a blocking decision exactly once (881.972396ms)
 ✔ a notifier that dies takes the host down with exit 1, so the unit restarts the pair (135.708239ms)
 ✔ a second host for the same data root refuses with exit 3 while the first runs (73.349518ms)
 ✔ a notifier that refuses stops the broker and the host refuses with exit 3 (124.893626ms)
 ✔ a notifier that refuses after the broker died still refuses with exit 3, without a send to the dead broker (134.813304ms)
 ✔ a notifier that dies before it replies, after the broker died, still refuses, without a send to the dead broker (87.76665ms)
 ✔ watchChildren reports a child that died before it was called, and one that dies later (36.929762ms)
 ✔ bus stop refuses to signal a live pid that is not a bus host (203.219855ms)
 ✔ bus start refuses with exit 3 and the code when the broker refuses to boot; bus status names the lock (276.768265ms)
 ✔ bus start refuses with exit 3 without a notifier config (96.336866ms)
 ✔ bus start runs until bus stop; status reports it while it runs (651.277703ms)
 ✔ bus-service.sh renders the unit and installs it into a given directory (41.913796ms)
 ✔ zoned uses the IANA zone across DST (25.237436ms)
 ✔ each open blocking decision is DM'd once, across polls and a restart; non-blocking ones are not (19.12349ms)
 ✔ two blocking decisions get two DMs with different nonces (15.04801ms)
 ✔ the digest nonce differs per business and per day and fits Discord's 25 characters (0.284039ms)
 ✔ a failed DM is journaled, backs off, and is retried until it lands (15.246058ms)
 ✔ five definite refusals stop a DM: one gave-up line, one log line, and a restart keeps the count (18.300879ms)
 ✔ 429s, 5xx-style unknowns and refusals without a status never count toward the limit (19.283766ms)
 ✔ a crash between the fifth refusal and its gave-up line: the next poll appends it and sends nothing (12.623787ms)
 ✔ the digest goes at 08:00 Chicago once a day, with blocking ones marked as DM'd (16.108573ms)
 ✔ a late start with no digest for the day sends one at once; an empty inbox gets one line (10.506593ms)
 ✔ an inbox read failure is logged and the next poll retries (0.812213ms)
 ✔ no Discord id reaches the journal or the log (12.42784ms)
 ✔ the journal: a torn tail is copied out and truncated, so an append after it reopens cleanly (2.848665ms)
 ✔ the journal: a crash between the copy and the truncate leaves a tail the next open repairs (2.496016ms)
 ✔ the journal: a whole file that is one torn line truncates to empty (1.549448ms)
 ✔ the journal: a malformed complete line refuses and leaves the file and any torn tail alone (0.82019ms)
 ✔ the journal: a loose file mode, a loose directory or a symlinked journal refuses (0.901995ms)
 ✔ the journal: a line with a wrong type refuses with exit 3 and names the field (3.155995ms)
 ✔ the journal: a symlinked directory refuses and says it is a link (0.639875ms)
 ✔ the journal: an append after the file was swapped for a symlink refuses and writes nothing through it (0.65835ms)
 ✔ the journal: a directory it cannot write or create refuses with exit 3 and names the path (0.631409ms)
 ✔ digest content stays within Discord's 2000 characters (0.442711ms)
 ✔ runLoop never overlaps ticks and stops after the one in flight (110.965551ms)
 task.close {} answered: invalid-request; fake saw 18 requests, first GET /info 200, GET /projects/1 200, GET /projects/1/views 200
 task.close on a missing task answered: task-not-found; it made GET /tasks/999 404
 ✔ bootConfig trackers reach the S3 adapter in the real broker child, which goes ready against a fake Vikunja (275.770061ms)
 ✔ the transport writes {business, verb, args} to the child and reads its JSON (42.708073ms)
 ✔ a bus code on stderr becomes the exit code; garbage and timeouts are outcome problems (2171.791227ms)
 ✔ busExit and refuseInsideAgent (0.512192ms)
 ℹ tests 60
 ℹ suites 0
 ℹ pass 60
 ℹ fail 0
 ℹ cancelled 0
 ℹ skipped 0
 ℹ todo 0
 ℹ duration_ms 2858.176494
@@ -0,0 +1,132 @@
 ✔ explicit request, Seen, ordinary completion and a new request have distinct attention states (6.804459ms)
 ✔ attention convention ignores reasoning/quoted examples and permits leading blank lines (0.419003ms)
 ✔ completed smoke replies and ordinary questions are idle, not human blockers (1.05788ms)
 ✔ only an explicit first-line input request makes a finished reply waiting (0.306714ms)
 ✔ tool activity, user text, errors and unfinished turns override attention text (0.191863ms)
 ✔ STOP access failure is unknown, not absence, under a non-root identity (37.705417ms)
 ✔ connector Task never inherits Discord routing envelopes; ordinary Task still uses user text (3.159431ms)
 ✔ connector discovery keeps only safe identity; rejects modes, mismatches, links and traversal (1.494166ms)
 ✔ canonical owner identity and STOP are independent; no tmux fallback or forged registration (6.396242ms)
 ✔ connector reply refusal precedes forged live tmux registration; ordinary agent still sends (0.5883ms)
 ✔ server rescans connector discovery and refuses HTTP reply without transport (37.825985ms)
 ✔ connector session links and linked directories are not read (1.123235ms)
 ✔ newer live matching launch marks old activity, preserves history/attention/attribution, then clears on new activity (4.059555ms)
 ✔ CLI print uses the relaunch notice instead of old current preview (69.0968ms)
 ✔ connector owner and fixed task never inherit a native relaunch notice (2.309453ms)
 ✔ equality, stale/unknown/offline, mismatched registration and unknown activity do not assert relaunch (1.447186ms)
 ✔ loadConfig: missing file throws ConfigError (1.74368ms)
 ✔ loadConfig: invalid JSON throws ConfigError (0.381839ms)
 ✔ loadConfig: missing dataRoot throws ConfigError (0.285709ms)
 ✔ loadConfig: relative dataRoot throws ConfigError (0.314598ms)
 ✔ loadConfig: valid config returns dataRoot (1.0148ms)
 ✔ findNewestSession: picks the newest by mtime among two files (0.633765ms)
 ✔ findNewestSession: finds files in nested subdirectories (0.368338ms)
 ✔ findNewestSession: returns null for a missing dir (0.150221ms)
 ✔ readSession: extracts fields, collapses/truncates text, counts a truncated final line (0.912704ms)
 ✔ readSession: model and provider follow the latest model_change entry or assistant turn; null when the log names neither; scanAgent carries them (2.004588ms)
 ✔ readSession: lastError carries the assistant errorMessage only when the last assistant turn errored (0.831458ms)
 ✔ findNewestSession/scan: never read sibling auth or secrets next to a sessions dir (1.688843ms)
 ✔ deriveState: full state table (0.267492ms)
 ✔ rule: newest entry is an assistant message with a tool call, after a question-looking text, is working (0.485035ms)
 ✔ rule: newest entry is a tool result with no assistant text after it is working (0.417277ms)
 ✔ rule: a finished ordinary turn is idle, even if it says your move (0.419062ms)
 ✔ task: the first user message of the session, from text blocks (0.387843ms)
 ✔ task: a plain-string user content is accepted, whitespace collapsed and long text capped (0.42006ms)
 ✔ task: no user message in the log means null (shown as unknown), never a guess (0.451738ms)
 ✔ workspace: the live tmux pane path wins; the session cwd is the fallback; neither means null (0.73409ms)
 ✔ activeProject: basename of the nearest .git directory or .git file above the workspace; none means null (1.270857ms)
 ✔ scan: the written record carries task, workspace and activeProject (0.772335ms)
 ✔ registration: overrides task, project and workspace; every source says registration; registered carries the launch fields; the grouping column is untouched (0.921072ms)
 ✔ registration: empty task and null project/workspace leave the derived values in place; registered is still non-null (0.610015ms)
 ✔ registration: a record whose pid is gone is stale; derived values win, sources say derived, registered stays with alive false; a pid the probe cannot decide is not stale; pidAlive itself (1.50323ms)
 ✔ registration: no registration leaves the Gate A fields exactly as before, and registered is null (0.680322ms)
 ✔ loadRegistrations: a missing seatsDir gives empty lists (0.248283ms)
 ✔ loadRegistrations: one good record, one malformed JSON, one with an unknown field; a stray file under seatsDir is ignored (1.064486ms)
 ✔ matchRegistration: matches by sessionsDir, and by realpath through a symlink; sessionsDir null never matches; same seat name with a different sessionsDir does not match (fleet vs repo darkwing) (0.379473ms)
 ✔ scan: writes the registration override to disk; index.json carries registered and registrationErrors (0.881035ms)
 ✔ scan: a relative seatsDir throws ConfigError; an omitted seatsDir behaves as before (0.775753ms)
 ✔ scanAgent: waitingOnYou is true for waiting/error and false otherwise (0.529713ms)
 ✔ scanAgent: ageSeconds is computed from the injected now (0.268697ms)
 ✔ scanAgent: sessionFile null and state idle when sessions dir is empty but alive (0.153852ms)
 ✔ discoverRepoAgents: finds agents with a sessions dir, skips those without, sorted by name (0.351278ms)
 ✔ discoverFleetAgents: finds agents with a sessions dir, sorted by name, fleet tmux fields (0.495575ms)
 ✔ scan: writes per-agent files and index.json, rerun overwrites, no leftover tmp files (1.087513ms)
 ✔ scan: relative boardDir throws ConfigError (0.101032ms)
 ✔ CLI: scan with assume-alive liveness exits 0, prints board summary, writes board files (74.511371ms)
 ✔ CLI: missing config exits 2 with a refused: message (61.342575ms)
 ✔ CLI: unknown command exits 2 (77.494963ms)
 ✔ CLI: unknown --liveness value exits 2 (67.918675ms)
 ✔ panesRunPi: true when any trimmed line equals 'pi' (0.256634ms)
 ✔ panesRunPi: false for bash-only, claude, empty, or node-pi-style lines (0.082198ms)
 ✔ tmuxIsAlive: a pane running pi is alive (0.226101ms)
 ✔ tmuxIsAlive: session exists but pi has exited is not alive (0.089543ms)
 ✔ tmuxIsAlive: no such tmux session is not alive (0.077396ms)
 ✔ tmuxIsAlive: tmux could not be run at all is unknown (null), never assumed alive (0.087494ms)
 ✔ tmuxIsAlive: passes -L <socket> only when a socket is given (0.107476ms)
 ✔ parsePanes: one pane per line, command and optional tab-separated path (0.091544ms)
 ✔ tmuxInspect: reports the path of the pane running pi, not of a shell pane (0.079394ms)
 ✔ tmuxInspect: no pi pane, no session, or no tmux gives no workspace and the matching liveness (0.112923ms)
 ✔ loadSeen: missing file returns {} (0.176853ms)
 ✔ loadSeen: invalid JSON throws ConfigError (0.215421ms)
 ✔ loadSeen: a JSON array throws ConfigError (0.270344ms)
 ✔ loadSeen: a non-string value throws ConfigError (0.199671ms)
 ✔ markSeen: seen true adds the key and writes seen.json mode 0600, no leftover tmp files (0.401563ms)
 ✔ markSeen: seen false deletes the key (0.297031ms)
 ✔ markSeen: missing, empty, or non-string fields throw ConfigError (0.248251ms)
 ✔ markSeen: project containing '/' throws ConfigError (0.152326ms)
 ✔ markSeen: non-boolean seen throws ConfigError (0.128344ms)
 ✔ scanAgent: a seen mark matching the waiting session's lastTimestamp clears waitingOnYou (0.341063ms)
 ✔ scanAgent: a stale mark (agent wrote something newer) is not seen and waitingOnYou is true (0.251943ms)
 ✔ scanAgent: a working session with a matching mark is not seen (marks only apply to waiting/error) (0.252277ms)
 ✔ scanAgent: an error-state session with a matching mark is seen (0.294401ms)
 ✔ scan: index.seen and waitingOnYou reflect seen.json, which scan never rewrites or deletes (0.565515ms)
 ✔ scan: a corrupt seen.json makes scan throw ConfigError (fail closed) (0.231652ms)
 ✔ taskSetBy: a registered task carries the record's setter; a record without the field (pre-#1511) reads unknown; the value is not copied into registered (0.765658ms)
 ✔ taskSetBy: null whenever the task shown is not the registered one: no registration, an empty registered task, a stale registration; the field is always present (0.684286ms)
 ✔ taskSetBy: scan() reads the field from disk through the seat package (bounded there), writes it to the per-agent record and index, and an invalid on-disk value is a registrationError, never a row value (1.186219ms)
 ✔ isLoopbackHost: recognizes loopback hosts (1.494333ms)
 ✔ isLoopbackHost: rejects non-loopback hosts (5.060193ms)
 ✔ startServer: refuses a non-loopback host with ConfigError, never opens a socket (3.605504ms)
 ✔ startServer: serves page, healthz, and a rescanning /api/board (38.661997ms)
 ✔ startServer: a seatsDir registration overrides the row and index.registered reflects it (7.651859ms)
 ✔ startServer: /api/board returns 500 JSON with an error field when scan throws (2.727ms)
 ✔ CLI: serve refuses a non-loopback host with exit 2 and a refused: message (63.239491ms)
 ✔ CLI: serve rejects a non-numeric --port with exit 2 (69.154042ms)
 ✔ CLI: scan still works after the async cli refactor (64.98425ms)
 ✔ CLI: live serve prints its URL and answers /healthz (69.66211ms)
 ✔ page.html: esc() escapes every HTML-significant character (0.655725ms)
 ✔ POST /api/seen marks a row; GET /api/board still shows it seen; seen:false clears it (8.302311ms)
 ✔ POST /api/seen without a JSON content-type returns 400 and does not write a mark (1.985436ms)
 ✔ POST /api/seen with invalid JSON returns 400 (2.886182ms)
 ✔ POST /api/seen with a body over 4096 bytes returns 400 (or resets the connection) and writes no mark (2.189468ms)
 ✔ POST /api/seen with a missing agent returns 400 (1.39406ms)
 ✔ POST /api/board returns 405; PUT /api/seen returns 405 (1.75052ms)
 ✔ CLI: scan --print marks a seen row with 's' and the summary line ends with 'N seen)' (57.024677ms)
 ✔ page.html: seenControl() escapes rec.project/agent/lastActivity, and the POST uses a JSON content-type (0.326067ms)
 ✔ page.html: has a collapsed Seen section that lists seen rows with the shared row builder (0.262888ms)
 ✔ page.html: each project has a Hide seen checkbox (default on) beside Hide offline, with a hidden-count note (0.163235ms)
 ✔ page.html: a project header reads "N of N" only while a checkbox hides rows (0.124708ms)
 ✔ page.html: every row shows Task and Active project, derived or the word unknown, with the workspace in the detail (0.322965ms)
 ✔ page.html: task and active project cells show their source via sourceTag(); the detail has a Registered row via registeredText(); SOURCE_LABEL maps registration to registered; every dynamic value in sourceTag/fromSource/registeredText is escaped (0.570669ms)
 ✔ POST /api/reply: runs agent-send.sh with -s from the registration, -S <host>:control-board, -m text plus the fixed trailer, no -L on the default socket, MOSAIC_TMUX_SOCKET stripped; answers delivered with the exit code and both streams (30.544502ms)
 ✔ POST /api/reply: a registration with a tmux socket adds -L <socket> (30.97077ms)
 ✔ POST /api/reply: a non-zero tool exit is a 200 with delivered false, the exit code and the stderr verbatim (25.810815ms)
 ✔ POST /api/reply: refusals before the tool runs: empty or blank or long text 400, unknown row 404, no registration 409, stale registration 409, no tmux session 409, bad JSON 400; the tool is never called (14.085827ms)
 ✔ POST /api/reply: a missing agent-send.sh is a 500 with the path in the error, not a crash (5.444484ms)
 ✔ replyToRow: DEFAULT_AGENT_SEND is the repository's tools/tmux/agent-send.sh and it is executable (0.187269ms)
 ✔ page.html: the reply box appears only where canReply() holds (live registration with a tmux session), the detail has a Reply row, the submit posts JSON to /api/reply, receipts and drafts survive a refresh, and every receipt value is escaped (0.588031ms)
 ✔ startServer: /api/board carries taskSetBy from a live registration and null for the derived rows (2.864235ms)
 ✔ page.html: the task cell and detail show who set a registered task via setByTag()/setByText(), both escaped, only from rec.taskSetBy; the reply gate does not read it (0.32637ms)
 ✔ Host/Origin guard: GET /api/board and POST /api/reply refuse a foreign Host, a wrong port and a cross-origin Origin with 403 JSON, before any scan or send, and never send CORS headers (8.426521ms)
 ✔ Host/Origin guard: loopback names on this port are accepted, with or without a same-origin Origin (31.234601ms)
 ✔ conversation routes (F16): a foreign Host, a wrong port and a cross-origin Origin get 403 before the reader runs, with no CORS headers (6.307879ms)
 ✔ every refusal code the reader can raise has an HTTP status (1.362529ms)
 ✔ conversation routes: catalogue, first page, next page and follow over HTTP; refusals map to 4xx with their code; nothing is written (58.278992ms)
 ℹ tests 124
 ℹ suites 0
 ℹ pass 124
 ℹ fail 0
 ℹ cancelled 0
 ℹ skipped 0
 ℹ todo 0
 ℹ duration_ms 739.046363
@@ -0,0 +1,205 @@
 ✔ W1: two processes acquire the same pair at once; exactly one claim (133.852204ms)
 ✔ W1: two writers publish the same revision at once: one wins, the other gets null, the winner's record stays (9.903435ms)
 ✔ W1: a revision name appears only after its bytes are synced; before that, only a temp file exists (4.896975ms)
 ✔ W2: acquire while a claim is reserved or active refuses already-active (135.640382ms)
 ✔ W3: acquire while stopping, uncertain, or stopped without proof refuses unsafe-replacement (206.391742ms)
 ✔ W4: same session with another seat tuple, and the reverse, both refuse; a loser on the seat key closes it no-unit (221.599112ms)
 ✔ W4: a hard link of one session under another seat is the same session: the second controller refuses already-active and launches nothing (23.384355ms)
 ✔ W4: a copy of one session under another seat is the same session: the second controller refuses already-active and launches nothing (24.450943ms)
 ✔ W4: a session header ID that changes after construction refuses target; nothing is claimed or launched (5.102513ms)
 ✔ W5: SIGKILL between every publication barrier of acquire and transition; restart never finds two holders or a lost claim (5568.837794ms)
 ✔ W5: SIGKILL between every publication barrier of release; restart finishes or holds the release (21997.530269ms)
 ✔ W6: controller killed mid-turn while the engine lives; restart is uncertain, no launch, prompts refuse (155.448481ms)
 ✔ W12: a live owner paused with SIGSTOP; a second controller refuses already-active and changes nothing (92.901015ms)
 ✔ W13: crash after the engine spawns, before active; restart finds the live unit: uncertain, no second spawn, force stop only (227.675785ms)
 ✔ W14: crash after reservation, before the spawn marker: stopped with a no-unit observation; the pair is free (210.868261ms)
 ✔ W20: crash after the spawn marker, scope collected; uncertain in both runs, the marker is copied, no launch until a boot proof (234.593289ms)
 ✔ W15: crash between the two keys during release; restart finishes it under the same claim ID (36.707665ms)
 ✔ W7: recorded boot ID differs on the same machine: stopped with a boot proof; open tool calls become uncertain (112.154849ms)
 ✔ W8: resume after a proven stop with the same pins: new claim ID, generation +1, same conversation, branch and leaf (36.795317ms)
 ✔ W9: resume with a changed binary, argv digest, branch or leaf is refused and the claim is unchanged (89.467083ms)
 ✔ W11: the controller writes no session file; only the fake engine's own appends appear (21.736151ms)
 ✔ W16: a highest revision that won't parse holds the pair uncertain; the older stopped revision is not reused (54.569487ms)
 ✔ W17: a claim root copied from another host refuses foreign-host and promotes nothing (59.350817ms)
 ✔ G1: a session path or claim root under .pi/state, ~/.claude, the data root or a registration refuses at construction (4.251176ms)
 ✔ G2: a symlink inside the fixture root to a live session file is refused by the real-path check (1.178681ms)
 ✔ G3: a fixture path swapped for a live path after construction is refused at bind (1.751493ms)
 ✖ K1: force stop kills a tool child that called setsid; stopped with a verified proof (2202.925091ms)
 ✔ K2: K1 on the process-group fallback ends uncertain, never stopped (131.632977ms)
 ✖ K3: SIGTERM acknowledged while a member lives: stopping until the kill phase, never stopped from TERM (2166.353377ms)
 ✔ K4: two engines; force stop one; the other survives by independent observation (4254.701406ms)
 ✔ K5: a stop during a tool call leaves the effect uncertain, and it is shown (2151.481499ms)
 ✔ K12: a member forking in a loop: the freeze stops it, enumeration is complete, populated 0 after cgroup.kill (2249.888496ms)
 ✔ K13: a member writing its pid into another cgroup is refused by the namespace; the kill is complete (2151.391984ms)
 ✔ K15: the shim gone, engine/cgroup.events unreadable, or the engine cgroup missing: evidence unavailable, not empty; uncertain (4379.099031ms)
 ✖ K10: controller killed between the TERM and kill phases: restart checks the invocation ID and re-runs from TERM for the same stop (209.431279ms)
 ✔ K11: controller killed after the confirmation is recorded, before TERM: restart checks the invocation ID and re-runs from TERM for the same stop (378.264182ms)
 ✔ K14: a unit with the recorded name but another invocation ID: evidence unavailable, no signals, uncertain (303.168036ms)
 ✔ K6: recover without proof, without confirmation, or with changed pins is refused (63.5942ms)
 ✔ K7: recover after proof, then launch: new claim and execution, generation +1, same leaf; the cancelled prompt is not replayed (31.907608ms)
 ✔ K8: an engine that loads another leaf on resume is refused before admission; it stays claimed until a proven stop (39.622245ms)
 ✔ K9: an interrupt that never settles stays uncertain; force stop stays available; takeover is refused while fenced (3026.332595ms)
 ✔ K16: a claim from another machine ID refuses foreign-host; no boot proof is issued (5.924081ms)
 ✔ K17: two launcher calls with one eligibility record: one launch, the other refuses, no second engine (31.602414ms)
 ✔ K18: the leaf changes after eligibility: launch refused; the reservation stays until released with proof (27.632187ms)
 ✔ S1: `/goal x`, with leading spaces or a tab, refuses text-policy at admission; zero engine bytes (41.162844ms)
 ✔ S2: every prefix pinned Pi interprets is refused, from the list the code uses; the rest reach the engine exactly (30.407461ms)
 ✔ S3: `/goal` on the second line is pinned from the source: Pi checks only index 0, so it is admitted and sent exactly (21.591465ms)
 ✔ S4: a `/` left in the composer is cleared when control transfers and returns; the next submit sends only the new text (38.967149ms)
 ✔ S5: an observer terminal gets a paste then Enter, as send-message.sh does: not admitted: controller, nothing sent (24.046296ms)
 ✔ S6: a mediated-shaped registration (no tmux) passed to the board's replyToRow: 409 no tmux session; exec never runs (0.517583ms)
 ✔ S7: ESC, bracketed-paste markers and U+2028/U+2029 travel as one JSON string; the engine receives the exact text in one record (20.4388ms)
 ✔ P3: a Pi confirm, select, input or editor dialog is shown disabled with a reason and never answered (126.035815ms)
 ✔ E1: send, ack, user, toolCall, toolResult, final answer: shown once, no refresh, draft and reading position kept (36.777384ms)
 ✔ E2: U+2028, U+2029 inside JSON strings and CRLF line ends each parse as one record, on the splitter and through the controller (18.087403ms)
 ✔ E3: a multipart final, two blocks, null request correlation and duplicate delivery (31.009894ms)
 ✔ E4: a page read after message_end but before its entry is persisted: marker at the seam, re-read after run-settled, each message once (39.751146ms)
 ✔ E4: a gap or a new epoch also reconciles; nothing is concatenated across a gap (11.347106ms)
 ✔ E5: an unknown native event gives no client event; evidence records its type and bytes; the terminal count goes up (30.498751ms)
 ✔ E6: a tool result delayed across a pause and a reconnect is reconciled without a manual refresh (50.879279ms)
 ✔ E7: the terminal renders the same stream as the library client, as observer and then as controller, and submits only as controller (43.41427ms)
 ✔ terminal: engine control characters are made visible; a lost connection refuses submit (29.777275ms)
 ✔ terminal: outcome unknown is shown as such, with no resend offer, and nothing is resent (1.192976ms)
 ✔ terminal: text after Enter in the same input chunk starts the next message; it never joins the one submitted (0.508887ms)
 ✔ terminal: a paste-start marker split right after its ESC still opens the paste; the Enter inside it never submits (0.635774ms)
 ✔ terminal: invisible and bidi characters are made visible; head, status and notice lines stay one line (0.203619ms)
 ✔ every record these fixtures produced is a valid CHAT-01 record (E5: no record fails the schema) (426.881843ms)
 ✔ H1: two takeovers with the same expected generation: one wins, +1; the other refuses generation (57.446613ms)
 ✔ H2: the old controller's prompt after a takeover commits is refused with zero engine bytes (82.431468ms)
 ✔ H3: a takeover while a prompt holds the dispatch lock: written under the old actor, or refused; never both (133.261666ms)
 ✔ H4: self-takeover is refused (16.211065ms)
 ✔ H9: Interrupt racing a prompt's dispatch: before the write, dispatch-refused and no-turn; after, §3 rules (97.939907ms)
 ✔ H10: Interrupt and force stop together: one stop chain, force stop supersedes (90.004566ms)
 ✔ H10: an overlap during the pause before the abort: no abort, the stop ends uncertain (22.175376ms)
 ✔ H10: a no-turn Interrupt lifts only its own fence; admission stays closed under force stop, overlap or revocation (101.29259ms)
 ✔ H11: the controller disconnects mid-turn: work continues, the claim is unchanged, control stays put (162.330347ms)
 ✔ H12: an exact retry after reconnecting to the same incarnation returns the same receipt; one dispatch (23.073584ms)
 ✔ H13: a retry with the same request ID and different text is refused (20.503171ms)
 ✔ H14: late stdout from the old engine after a replacement is dropped by incarnation, counted, never rendered (160.147974ms)
 ✔ H15: a revoked connection's command is refused; the revocation fence holds (70.472627ms)
 ✔ H16: a second controller for the same session refuses already-active; the first is untouched (20.528122ms)
 ✔ H10: a second force stop while the first escalation runs refuses fenced; one escalation, and the claim records only the first stop's phases (57.87053ms)
 ✔ H17: a confirmation reused, answered from another connection, or used after the stop changed is refused (60.931438ms)
 ✔ H18: two prompts before any native output: the second refuses busy; one engine write (13.031369ms)
 ✔ H19: the pipe fails mid-line under a large prompt: delivery-unknown transport-unknown, poisoned, no later write (120.535221ms)
 ✔ H19: the link itself never writes again after an unknown outcome, whoever calls it (0.769166ms)
 ✔ H19: the controller dies mid-write of a large line: after restart the outcome is unknown and nothing is resent (474.647371ms)
 ✔ H20: the line is written but the ack is lost when the controller dies: orphan, outcome unknown, nothing resent (402.296798ms)
 ✔ H21: a retry of the exact request with the old token after a crash is stale-incarnation; no second write (335.17129ms)
 ✔ H22: after H21 and a valid recovery, a new request with the new token is admitted (2389.207984ms)
 ✔ H23: requests pending at a restart are not resent; each shows outcome unknown (459.593295ms)
 ✔ a plain conversation: catalogue row, one page, CHAT-01 records (11.130499ms)
 ✔ native entries map to blocks: tools, thinking, bash, notices, ids that do not fit (2.570142ms)
 ✔ F1: a malformed line is an unavailable part at its position, and reading continues (2.07203ms)
 ✔ F1: a missing parent stops the history with a notice that names the unreadable lines (2.811213ms)
 ✔ F1: an unreadable fork is never merged into another branch's history (1.359937ms)
 ✔ F1: a follow stays on its branch when the next entry's parent is unreadable (2.4616ms)
 ✔ F1: a file whose entries are all unreadable shows a notice per line (1.05576ms)
 ✔ F2: a truncated trailing line marks the view incomplete, not an error (1.678293ms)
 ✔ pagination: 100 parts, then the rest; parts concatenate to the whole branch (5.10643ms)
 ✔ F3: a replaced file (new inode) refuses old cursors with reconcile (5.602312ms)
 ✔ F4: a same-inode rewrite of the prefix refuses old cursors with reconcile (7.735978ms)
 ✔ F5: growth between pages keeps the epoch and the page stops at the pinned length (6.262886ms)
 ✔ F6: unknown, foreign and expired cursors refuse and leave the cursor usable (11.55803ms)
 ✔ F7: a symlinked file and a symlinked directory component are refused, never opened (15.622698ms)
 ✔ F8: a file swapped for a symlink after the catalogue is refused (2.769422ms)
 ✔ F9: registrations never add or redirect a root (2.073453ms)
 ✔ F10: a header cwd naming another project is refused (5.419193ms)
 ✔ F11: parentSession renders with a marker and the parent is never opened (0.926192ms)
 ✔ F12: two leaves: the default leaf is shown and the other branch reads alone (7.514641ms)
 ✔ F12: a follow refuses when an appended duplicate id changes the branch's earlier parts (4.071265ms)
 ✔ F12: a second root (Pi's resetLeaf) starts its own branch (1.694144ms)
 ✔ F13: compaction is a marker in place, then the retained content (0.920202ms)
 ✔ F14: long strings split into fragments and parts, reassemble exactly, and pages respect the byte cap (754.032088ms)
 ✔ fragments never cut a surrogate pair and keep an empty string (6.844356ms)
 ✔ F15: a Claude seat is an unsupported-harness placeholder whose directory is never read (2.724415ms)
 ✔ unknown conversations, empty files and non-Pi files refuse (6.117025ms)
 ✔ an unreadable file or root inside the roots is refused per row, not a failed catalogue (3.380886ms)
 ✔ a seat directory without search permission refuses that root, not the catalogue (3.162776ms)
 ✔ every page and cursor is a valid CHAT-01 record (906.130106ms)
 ✔ the engine pin holds for the installed package (4.095079ms)
 ✔ pinned Pi, sealed and without credentials, answers the controller's commands with the shapes the fake models (329.078059ms)
 ✔ pinned Pi appends thinking_level_change at start when the branch lacks one, so the leaf moves (K8 then fails closed) (367.946117ms)
 ✔ N25: ordinary Interrupt reconciles; a non-empty queue_update in the window is O5 (86.798399ms)
 ✔ N1: an extension's follow-up queued after the fence is cleared before any abort; O5, Unknown (48.062807ms)
 ✔ N1: a follow-up queued before the fence is O5 at once; the Interrupt refuses fenced (26.288732ms)
 ✔ N2: with abort first, the fake runs the external item (the ordering guard has teeth) (22.777314ms)
 ✔ N3: the fence lands in preflight, preflight errors, no run: failed, No run, uncertain (32.219782ms)
 ✔ N4: the ack arrives after the first abort and a run starts: clear and abort again; Interrupted (23.411923ms)
 ✔ N5: an input handler takes the prompt: ack, no run, delivery-unknown handled-without-run (116.896584ms)
 ✔ N6: an extension queues between clear_queue and abort: O5 and O6, Unknown (43.955144ms)
 ✔ N7: clear_queue times out: no abort, nativeQueue unknown, force stop still ends it (1542.958269ms)
 ✔ N7: clear_queue answers an error: no abort, nativeQueue unknown, the link not poisoned (19.768859ms)
 ✔ N8: an extension prompt starts a run during Mosaic preflight; the losing settle is O3 (81.992353ms)
 ✔ N9: a run that started before the fence and ends aborted: failed interrupted, Interrupted (15.892432ms)
 ✔ N9: decision 34: a run that ends aborted with no stop in progress: aborted-without-stop, uncertain, outcome unknown (17.178428ms)
 ✔ N9: an aborted that lands after the fence but before any abort is written: aborted-without-stop, Unknown (33.514465ms)
 ✔ N10: fake conformance (32.242568ms)
 ✔ N11: the run fails before any user message_start: delivery-unknown ack-without-start, never failed (31.604688ms)
 ✔ N12: input that starts a run after the final empty clear is O1 and not part of the stop's proof (19.419789ms)
 ✔ N13: agent_start with no slot held is O1; a later prompt refuses with zero engine bytes (64.723847ms)
 ✔ N14: the run completes while clear_queue is in flight: finished, Completed first, uncertain (30.077104ms)
 ✔ N14: the run completes after the abort is written, before Pi applies it: finished, never relabelled (32.547228ms)
 ✔ N15: the fence lands in preflight, then an input handler takes it: handled-without-run, No run (23.737116ms)
 ✔ N16: Interrupt with no slot and no run refuses no-turn: no stop, no bytes, admission open (15.431539ms)
 ✔ N17: the run fails on its own during the exchange: failed, Failed on its own (30.152089ms)
 ✔ N18: no final assistant message_end, or a lost line: working stays working; before working, transport-unknown (113.257341ms)
 ✔ N19: a losing extension prompt settles inside the Mosaic run before its user message: O3, run-overlap (134.703827ms)
 ✔ N20: an extension triggerTurn during Mosaic preflight starts first; while streaming it queues with no signal (84.206586ms)
 ✔ N21: a losing settle after the receipt settled finished is O2; the receipt stays finished (17.434773ms)
 ✔ N22: an agent-level custom message is dropped by the clear with no signal; evidence names the seal (15.350181ms)
 ✔ N23: a nextTurn message survives clear and abort and attaches to the next prompt, with no signal (13.781605ms)
 ✔ N24: the seal is an allow-list: --extension, a missing --no-* flag, a second --mode or --session, a session or output flag, or a stray word refuses unsealed-engine; no engine starts (42.472646ms)
 ℹ tests 152
 ℹ suites 0
 ℹ pass 149
 ℹ fail 3
 ℹ cancelled 0
 ℹ skipped 0
 ℹ todo 0
 ℹ duration_ms 31680.952099
 ✖ failing tests:
 test at packages/conversation/tests/cohort.test.mjs:139:1
 ✖ K1: force stop kills a tool child that called setsid; stopped with a verified proof (2202.925091ms)
  AssertionError [ERR_ASSERTION]: the escaped child is a listed member
      at TestContext.<anonymous> (file:///home/jwoltje/filbert-scratch/r45/cand/packages/conversation/tests/cohort.test.mjs:151:12)
      at async Test.run (node:internal/test_runner/test:1409:7)
      at async startSubtestAfterBootstrap (node:internal/test_runner/harness:387:3) {
    generatedMessage: false,
    code: 'ERR_ASSERTION',
    actual: false,
    expected: true,
    operator: '==',
    diff: 'simple'
  }
 test at packages/conversation/tests/cohort.test.mjs:176:1
 ✖ K3: SIGTERM acknowledged while a member lives: stopping until the kill phase, never stopped from TERM (2166.353377ms)
  AssertionError [ERR_ASSERTION]: the member ignored TERM
      at TestContext.<anonymous> (file:///home/jwoltje/filbert-scratch/r45/cand/packages/conversation/tests/cohort.test.mjs:190:12)
      at async Test.run (node:internal/test_runner/test:1409:7)
      at async Test.processPendingSubtests (node:internal/test_runner/test:974:7) {
    generatedMessage: false,
    code: 'ERR_ASSERTION',
    actual: false,
    expected: true,
    operator: '==',
    diff: 'simple'
  }
 test at packages/conversation/tests/cohort.test.mjs:426:3
 ✖ K10: controller killed between the TERM and kill phases: restart checks the invocation ID and re-runs from TERM for the same stop (209.431279ms)
  AssertionError [ERR_ASSERTION]: the member is alive across the crash
      at TestContext.<anonymous> (file:///home/jwoltje/filbert-scratch/r45/cand/packages/conversation/tests/cohort.test.mjs:431:12)
      at process.processTicksAndRejections (node:internal/process/task_queues:104:5)
      at async Test.run (node:internal/test_runner/test:1409:7)
      at async Test.processPendingSubtests (node:internal/test_runner/test:974:7) {
    generatedMessage: false,
    code: 'ERR_ASSERTION',
    actual: false,
    expected: true,
    operator: '==',
    diff: 'simple'
  }
@@ -0,0 +1,186 @@
 ✔ approvals: a request is validated before anything is posted; the rendering shows names and never ids (3.652717ms)
 ✔ approvals: the ledger is appended and folded into open requests with bind and approval states (2.095993ms)
 ✔ approvals: a reply approves only when it points at a request, says exactly approve, and comes from a listed approver once (0.54469ms)
 ✔ approvals: a button approves only on its own request message with the matching custom id (0.452789ms)
 ✔ approvals flow: a turn that opened a request posts the message with the button, records it, binds it, and both approvers approve (23.493731ms)
 ✔ approvals flow: a non-approver, a repeat, a wrong custom id and a service refusal each get their fixed line and a drop entry (7.837264ms)
 ✔ approvals flow: an invalid request from the model, a refused post, and no api client are recorded and post nothing (7.33591ms)
 ✔ approvals flow: start retries a bind and an approval left as unknown, under their original keys (2.359428ms)
 ✔ authorize: open channel, listed user (2.063606ms)
 ✔ authorize: wrong guild (0.232728ms)
 ✔ authorize: no guild (DM) (0.315948ms)
 ✔ authorize: unlisted channel (0.197313ms)
 ✔ authorize: unknown channel, no info (1.556667ms)
 ✔ authorize: thread of listed parent (0.220056ms)
 ✔ authorize: thread of unlisted parent (0.194441ms)
 ✔ authorize: text channel that is not a thread and not listed (0.389674ms)
 ✔ authorize: unlisted user (1.236333ms)
 ✔ authorize: no author (0.268897ms)
 ✔ authorize: bot author (listed id, bot flag) (0.144111ms)
 ✔ authorize: system author (0.957145ms)
 ✔ authorize: the bot itself (0.097035ms)
 ✔ authorize: webhook (0.788496ms)
 ✔ authorize: mention channel without mention (0.216629ms)
 ✔ authorize: mention channel with bot mention (0.126082ms)
 ✔ authorize: mention channel with @everyone only (0.200495ms)
 ✔ authorize: mention channel mentioning someone else (0.080933ms)
 ✔ authorize: mention channel, content says @bot but mentions empty (0.118102ms)
 ✔ authorize: private thread under mention channel, mentioned (0.088496ms)
 ✔ authorize: private thread under mention channel, not mentioned (0.075425ms)
 ✔ authorize: thread in another guild per channel info (0.074497ms)
 ✔ authorize: not an object (0.071747ms)
 ✔ authorize: no id (0.080135ms)
 ✔ authorize: oversize content is accepted and flagged (0.080836ms)
 ✔ authorize: exactly the limit is not oversize (0.068839ms)
 ✔ authorize: a user's channel allowlist drops them outside it, threads count as the parent, others are unaffected (0.468829ms)
 ✔ authorize: order puts wrong guild before user, and user before channel (no channel lookup for strangers) (0.321456ms)
 ✔ binding: a complete binding validates and is frozen (2.628137ms)
 ✔ binding: unknown key, missing field, wrong type refuse with exit 2 (1.454888ms)
 ✔ binding: empty allowlists refuse (0.431805ms)
 ✔ binding: a user's channel allowlist must be non-empty, listed and unique; absent means every listed channel (1.282583ms)
 ✔ reloadDiff: reloadable keys are summarised by id; every fixed key refuses with exit 2 (1.965346ms)
 ✔ binding: dmRecipient is optional, a snowflake, one of the listed users, and a fixed key (1.284463ms)
 ✔ binding: file must be 0600, regular, not a symlink (2.055307ms)
 ✔ binding: token file mode, symlink, emptiness and shape are checked; token never appears in errors (1.46213ms)
 ✔ cli: check refuses a non-0600 token file with exit 2 before any network use (96.721294ms)
 ✔ context files: absolute paths, traversal, symlinks and out-of-repo targets refuse; in-repo files resolve (1.578284ms)
 ✔ cli: check refuses a missing context file and a missing binding with exit 2; usage is exit 4 (355.155821ms)
 ✔ cli: reload validates the file first (exit 2), then needs a live owner (exit 1); usage is exit 4 (234.212192ms)
 ✔ cli: run refuses when STOP is present, before any network use (147.786226ms)
 ✔ binding: tools is optional, validated strictly, a fixed key for reload, and its roots are resolved against the data root (1.908457ms)
 ✔ binding: a git key is validated at load and reaches the extension whole, and only on a writable root (1.315136ms)
 ✔ delivery: an accepted message is in the inbox before the turn, the reply is chunked with one nonce per chunk, and the turn record is write-once (23.053597ms)
 ✔ delivery: refused and unknown outcomes are journaled; a later chunk is not sent after a failure (17.105767ms)
 ✔ delivery: restart with an unknown entry re-sends the same nonce once and reconciles before accepting traffic (1.598351ms)
 ✔ delivery: an unknown entry older than the dedupe window is marked refused, not re-sent; a still-unknown one refuses start (1.109963ms)
 ✔ delivery: repeated unknown reconciliations never refresh the dedupe window; the original intent time decides (1.312252ms)
 ✔ turn: a failed engine turn posts the fixed line, never model output, and writes a failed record (2.646579ms)
 ✔ turn: a second message during a turn is held by the engine, both get their own reply and record (33.105429ms)
 ✔ turn: a thread under a listed channel is answered in the thread; an unknown thread is looked up once (3.300546ms)
 ✔ drop: an unlisted user gets silence and one drop line; no inbox entry, no REST call, no engine call (1.045885ms)
 ✔ drop: an oversize message is accepted into the inbox, answered with the fixed line and journaled as a drop (1.001041ms)
 ✔ restart: an inbox with three ids and a replay of the same three produces zero turns (43.015168ms)
 ✔ stop: STOP present refuses start; STOP written while running refuses new turns and the current one finishes (31.318708ms)
 ✔ ceiling: the ceiling plus one is refused and journaled; one fixed line per UTC day; a new day accepts again (5.996738ms)
 ✔ ceiling: a burst arriving while turns are still running cannot queue past the ceiling (3.01912ms)
 ✔ ceiling: a turn interrupted by a crash still counts after restart; admissions are durable (2.559289ms)
 ✔ ceiling: the daily notice survives a same-day restart; one delivery attempt in total, even when the first attempt crashed mid-flight (4.581434ms)
 ✔ duplicate: the same event delivered twice while the thread lookup is held yields one prompt, one admission and one reply (1.157529ms)
 ✔ journal: no token-shaped string and no model output on the drop path reaches disk (0.426272ms)
 ✔ receipt: an admitted message gets one eyes reaction on the inbound message; drops and refusals get none; a failed reaction is recorded and does not fail the turn (2.700785ms)
 ✔ receipt: Discord refusing the reaction leaves the turn intact and records ok false (2.075005ms)
 ✔ reload: a new user is silent before and answered after; a removed channel goes silent; a lower ceiling applies at once (3.891254ms)
 ✔ reload: a fixed key refuses with exit 2 and the old binding stays in force (2.128902ms)
 ✔ tools: with a tools binding the turn record lists every read and its outcome; without one the field is null (3.824301ms)
 ✔ context: the Discord block names the server, channels and modes, and states the rules from Q15 and Q16 (3.098697ms)
 ✔ context: with tools the block names the roots, keeps file content as data, and says to state refusals plainly (0.835755ms)
 ✔ context: a writable root adds the write rules and says a write is real only once Jason commits (2.21152ms)
 ✔ context: the envelope is one bracketed line then the text; names cannot break the line (2.958133ms)
 ✔ context: a git root swaps the terminal-commit line for the git verbs, and a vault root adds the id protocol (1.624172ms)
 ✔ context: assembleContext concatenates files in launcher format and appends the block; sha256 is stable (1.738849ms)
 ✔ context: splitReply keeps paragraphs together under the limit and splits long ones at lines, spaces, then hard (0.688051ms)
 ✔ engine: buildPiArgs carries the fixed flags, engine settings, session dir and prompt file (4.261033ms)
 ✔ engine: with tools, buildPiArgs turns pi's own tools off, loads the extension explicitly and allowlists exactly our three (0.451512ms)
 ✔ engine: a run with tool turns settles once, on the answer, with every tool call in the result (57.228583ms)
 ✔ engine: a run that ends on a tool-only turn fails the prompt as empty; a retried run settles on the real end (37.77451ms)
 ✔ engine: one prompt, one turn, text and usage come back (34.179736ms)
 ✔ engine: a prompt while streaming is held until pi settles, then sent as its own run, and answered in order (344.382562ms)
 ✔ engine: a held prompt that times out before pi settles fails on its own and is never sent (236.013539ms)
 ✔ engine: timeout sends abort and fails only that turn; the process stays (104.867427ms)
 ✔ engine: tool events from a run that outlived its timeout never land in the next prompt's record (232.82586ms)
 ✔ engine: a prompt after a turn that timed out before its agent_start waits for pi to settle instead of being refused (126.632884ms)
 ✔ engine: when pi has not started a timed-out turn by the end of the abort grace, the engine stops pi and fails held prompts (214.392388ms)
 ✔ engine: a timed-out turn pi starts only after the grace never answers a later prompt (614.922216ms)
 ✔ engine: late events of a run past its grace, before pi exits, answer nothing and nothing more is sent (early prompt response) (1.345609ms)
 ✔ engine: late events of a run past its grace, before pi exits, answer nothing and nothing more is sent (late prompt response) (0.426624ms)
 ✔ engine: a timed-out run pi did start outlives the grace; the next prompt goes out when it ends (428.485032ms)
 ✔ engine: a malformed JSONL line fails the turn, not the process (24.737054ms)
 ✔ engine: a turn that ends in error rejects with the error code; process exit fails pending turns (48.54618ms)
 ✔ gateway: hello -> identify with intents, ready, heartbeat with jitter, ack (3.303979ms)
 ✔ gateway: missed ack closes the socket and resumes with the last sequence (1.748646ms)
 ✔ gateway: op 7 reconnect resumes; op 9 non-resumable re-identifies (0.64364ms)
 ✔ gateway: op 9 resumable resumes (0.378703ms)
 ✔ gateway: close 4014 is fatal, reports the missing intent, never reconnects (0.846124ms)
 ✔ gateway: 4004 and 4013 are fatal too; 1006 reconnects with identify when no session (0.530317ms)
 ✔ gateway: close() is final and unparseable frames are ignored (0.4019ms)
 ✔ git: config validation is strict, needs write: true, a work tree and a private token file (67.17032ms)
 ✔ git: the child environment drops every host git config, names one helper, and carries the token path only for origin (43.499521ms)
 ✔ git: status reports the branch, ahead/behind and changed paths, and refuses off the named branch or mid-merge (78.79627ms)
 ✔ git: parseStatus reads porcelain v2 including renames and conflicts (0.291115ms)
 ✔ git: a commit stages exactly the named files, carries the seat author and the requester trailer, and pushes at once (82.141136ms)
 ✔ git: commit refusals: message, paths, requester, nothing to commit, and an index that already holds other work (122.090612ms)
 ✔ git: a commit whose push fails is still a commit, says so, and the next commit's push carries both (D6) (89.938936ms)
 ✔ git: pull is fast-forward only; a diverged origin or dirty local files refuse with nothing merged (263.800177ms)
 ✔ git: push pushes the named branch only and reports up to date (79.986947ms)
 ✔ git: no token value or token path ever reaches a git argument list; outputs are masked and capped (94.991393ms)
 ✔ git: the credential helper answers get over https from a private file and nothing else (198.30173ms)
 ✔ git: the vault protocol validates before a commit, honours another owner's lock, reserves ids, and locks around writes (798.143353ms)
 ✔ lock: the claim is exclusive; a second start against a live owner refuses (4.396422ms)
 ✔ lock: a stale lock (dead owner, reused pid, or record without start) refuses run and is never signaled; only unlock clears it (76.864189ms)
 ✔ lock: an incomplete claim (directory without owner record) is busy and refuses run; unlock clears it (0.799364ms)
 ✔ lock: an owner record that exists but cannot be read is invalid: never signaled, never removed, never claimed over (2.415724ms)
 ✔ lock: legacy upgrade; a live connector holding a {pid, start} record is unknown, unlock refuses and nothing changes; after it exits, unlock clears it (41.220051ms)
 ✔ lock: a live pid whose record carries a malformed or noncanonical start or boot string is unknown, not a mismatch; nothing signals, removes, or claims over it (343.762684ms)
 ✔ lock: identity syntax; only canonical unsigned decimal start ticks and lowercase boot uuids are identities (0.542473ms)
 ✔ lock: a process whose start marker or boot id cannot be read refuses to claim (25.632805ms)
 ✔ lock: a live pid whose identity cannot be read right now is unknown: never signaled, never removed, never claimed over (1.029429ms)
 ✔ lock: four processes racing for the same binding; exactly one claims it and the others refuse (43.572685ms)
 ✔ lock: stale handoff; concurrent starts over a stale lock all refuse, nothing reclaims, one unlock then exactly one live owner (183.79061ms)
 ✔ lock: four-party schedule; claims landing inside an unlock's gap never survive, one unlock leaves no owner and no residue (93.704603ms)
 ✔ notices: a kind is recorded per UTC day and found again (0.54971ms)
 ✔ notify: the DM channel opens once, every send carries the nonce, and only the message id comes back (5.184289ms)
 ✔ notify: refusals and unknowns surface as RestOutcome without ids; a refused send reopens the channel next time (2.440721ms)
 ✔ notify: no dmRecipient, a non-0600 binding or token, or a missing binding refuse before any network use (3.141812ms)
 ✔ recover: nothing to do is clean; a lock whose owner is gone or that has no record is cleared and STOP ends up absent (54.032714ms)
 ✔ recover: an operator STOP refuses with exit 3 and is never removed, whatever the lock says (45.592566ms)
 ✔ recover: a brake written during the unlock wins; STOP stays with both lines and the start is refused (28.321045ms)
 ✔ recover: a held binding refuses with exit 3 and writes no STOP: live owner, alive pid without verifiable identity, unreadable record (72.192985ms)
 ✔ cli: recover exits 0 when ready, 3 behind a brake or a held binding, and run's own STOP refusal is 3 (761.15839ms)
 ✔ rest: createMessage sends nonce, enforce_nonce, empty allowed_mentions and a soft reply reference (2.576273ms)
 ✔ rest: 429 waits retry_after and retries; 4xx is refused; 5xx and socket errors are unknown (3.288617ms)
 ✔ rest: content and nonce limits are enforced locally; typing never throws (0.766124ms)
 ✔ rest: react PUTs the encoded emoji on the inbound message for @me; 2xx is true, anything else is false and never throws (0.931996ms)
 ✔ rest: createDm posts the recipient and resolves the channel id; 4xx refused, 5xx unknown, bad id never sent (0.950264ms)
 ✔ setspark config: a bare https or loopback origin, a private key file, a principal (5.167307ms)
 ✔ setspark config: reaches the tools config and the binding as a fixed key (2.841201ms)
 ✔ setspark config: the binding's key survives resolveToolRoots and the engine's JSON hand-off to the extension (1.323203ms)
 ✔ setspark config: approvers come from the binding's users, never from the binding's setspark key (1.946563ms)
 ✔ setspark verbs: required_approvers go out as discord ids from names and come back as names (20.420568ms)
 ✔ setspark verbs: no Discord user id reaches tool text, whatever shape the service returns it in (8.293349ms)
 ✔ setspark contract: a decision made with names opens a request the connector accepts; names stored by an old record still refuse (5.469769ms)
 ✔ setspark keys: read per call, one printable token per file, rotation without a restart (6.703627ms)
 ✔ setspark idempotency keys: principal, turn id, call index; connector keys name a step (0.632777ms)
 ✔ setspark http core: json in and out, bearer header, idempotency header, fixed user agent, no key anywhere else (3.143337ms)
 ✔ setspark http core: error bodies become fixed refusals with code and the 409 fields; server text is data, cut (1012.996595ms)
 ✔ setspark verbs: a setspark key enables the eight verbs and no counters (0.440876ms)
 ✔ setspark verbs: writes carry the turn's key and the asserted requester, reads carry no key, and the api key never appears in text or details (6.550508ms)
 ✔ setspark verbs: no turn refuses every write before any request; bad arguments refuse before any request; reads still work (1.366425ms)
 ✔ setspark verbs: renderRecord caps long output and hides the accepted snapshot (0.222175ms)
 ✔ setspark api: bind, add_approval (button and reply) and get use integer request ids and the connector's keys (2.498032ms)
 ✔ tools: config refuses a missing, symlinked, dotted, non-directory or duplicate root and bad limits (3.692007ms)
 ✔ tools: every escape is refused with a fixed reason and nothing outside the root is read (3.770283ms)
 ✔ tools: happy paths list, read a window, and search case-insensitively; dotfiles and symlinks never appear (4.802707ms)
 ✔ tools: the tool set renders text for the model, records details for the journal, and enforces the per-run budget (3.318209ms)
 ✔ tools: listing and search caps hold (9.386252ms)
 ✔ tools: credential shapes are caught; ordinary prose and ids are not (0.806032ms)
 ✔ tools: the read uses the checked file itself; a symlink, a swapped file, a FIFO, a grown file or a hard link at read time is refused (4.720349ms)
 ✔ tools: an unreadable file under the root is skipped by search and refused by read (1.213734ms)
 ✔ tools: config accepts write: true only as a boolean, and enables the write tools only then (1.268516ms)
 ✔ tools: every write outside the fence is refused before any byte lands, and no temp file remains (4.495725ms)
 ✔ tools: write_file leaves the exact bytes, edit_file replaces one exact match, and the set renders the change as uncommitted (2.560525ms)
 ✔ tools: a target that changed between the check and the rename is refused and the temp file is removed (1.735625ms)
 ✔ web: config takes an https or loopback-http SearXNG base url and a bounded fetch cap (3.136088ms)
 ✔ web: address rules refuse every private, loopback, link-local, mapped and multicast form (1.907344ms)
 ✔ web: web_fetch refuses bad urls, private hosts, rebinding names, non-https redirects, too many hops, error status, non-text bodies, and times out (1024.755556ms)
 ✔ web: web_fetch returns html as text with the title, follows an https redirect, keeps plain text and json, and cuts at the cap (4.292763ms)
 ✔ web: html to text drops scripts, styles and comments, decodes entities and keeps block breaks (0.240366ms)
 ✔ web: web_search asks the instance for json, returns at most ten clean results, and refuses a bad query, a down instance or an unusable answer (2.695733ms)
 ✔ web: the tool set enables the web tools only with a web key, counts them in the budget, and records url, status and hits (2.408608ms)
 ℹ tests 178
 ℹ suites 0
 ℹ pass 178
 ℹ fail 0
 ℹ cancelled 0
 ℹ skipped 0
 ℹ todo 0
 ℹ duration_ms 2643.218847
@@ -0,0 +1,86 @@
 ✔ a raw token file, with or without one trailing newline, reaches curl only through the config stream (126.128293ms)
 ✔ the raw path accepts nothing else, and refuses before curl runs (400.542688ms)
 ✔ the file checks still apply on the raw path: mode, symlink, missing, directory (320.044861ms)
 ✔ the raw path base URL has no override (68.875532ms)
 ✔ the JSON path is unchanged, and JSON never falls through to the raw path (317.322975ms)
 ✔ a file that changes between the two reads refuses before curl runs, with or without a body (768.036835ms)
 ✔ the token reaches no child environment, even with an inherited CFG or SHELLOPTS=allexport (846.164709ms)
 ✔ real helper GET HTTP 200 preserves exit 0 without credentials (13.945172ms)
 ✔ real helper POST HTTP 201 preserves exit 0 without credentials (10.844397ms)
 ✔ real helper GET HTTP 403 preserves exit 1 without credentials (8.950885ms)
 ✔ fixture git subjects only, follow-ups and three session kinds (114.725705ms)
 ✔ text and JSON carry same numbers, open and truncated title (184.761853ms)
 ✔ missing credentials exit 2, no-issues never calls API and shows unknown (152.862609ms)
 ✔ empty range gives no rows and zero totals (104.702569ms)
 ✔ inclusive UTC dates, first-line preamble only, role and seat boundaries (113.97158ms)
 ✔ close-only issue included, even median, missing metadata stays unknown (118.811086ms)
 ✔ unique commits but per-issue links count multiple tags once each (129.964094ms)
 ✔ page cap refuses rather than silently undercounting (114.108135ms)
 ✔ bad API payload not JSON refuses (113.455091ms)
 ✔ bad API payload {} refuses (110.844375ms)
 ✔ bad API payload [{"number":1}] refuses (112.540073ms)
 ✔ partial or malformed session log refuses with location, not content (121.714506ms)
 ✔ a U+2028 or U+2029 inside a session string is one line, not a malformed record (107.395892ms)
 ✔ no sessions is an empty table; symlink source refuses (160.016344ms)
 ✔ reads only refactor even when another branch is checked out (105.155384ms)
 ✔ invalid dates, reverse dates and duplicate options refuse (60.13167ms)
 ✔ T3 agent assignments do not count as human in Table 2 (106.620619ms)
 ✔ preamble parsing and issue number boundaries (0.513199ms)
 ✔ T3 header: agent, or board from control-board; anything short of the full header is human (0.149218ms)
 ✔ no closed issues with human messages means undefined ratio, not invented zero (0.207988ms)
 ✔ T3: seat, archived, unmapped and Researcher threads count; imported, deleted and other-project threads do not (234.383673ms)
 ✔ T3: the default path is read from HOME and prints no path line; --no-t3 says so (521.013093ms)
 ✔ T3: a HOME with no database exits 1 and names --no-t3 (95.790127ms)
 ✔ T3: a file that is not a database exits 1 and names --no-t3 (101.911798ms)
 ✔ T3: a seat thread renamed to another seat exits 1 naming thread, title and roles (149.00837ms)
 ✔ T3: an unmapped thread addressed as a seat exits 1 (117.853131ms)
 ✔ T3: a header to another thread id is not cross-checked (123.003429ms)
 ✔ T3: no project, or two, for this root exits 1 (325.128967ms)
 ✔ T3: a removed column exits 1 and names it (130.524629ms)
 ✔ T3: a missing table exits 1 and names it (111.192816ms)
 ✔ T3: a counted row with an unknown role exits 1 without its text (121.609112ms)
 ✔ T3: a counted row with non-text content exits 1 without its text (119.865902ms)
 ✔ T3: a counted row with an unparseable created_at exits 1 without its text (117.132498ms)
 ✔ T3: a missing orchestration_events makes the diagnostic unknown and keeps the counts (219.876231ms)
 ✔ T3: a human message with no event counts in humanWithoutEvent (123.337704ms)
 ✔ T3: an unparseable event makes the diagnostic unknown and keeps the counts (217.196841ms)
 ✔ T3: an event with no string messageId makes the diagnostic unknown and keeps the counts (216.275372ms)
 ✔ T3: an error that is not from SQLite is rethrown, not reported as a database failure (60.752619ms)
 ✔ T3: a symlink at ~/.t3 exits 1 (87.471658ms)
 ✔ T3: a symlink at ~/.t3/userdata exits 1 (94.051514ms)
 ✔ T3: a symlink at ~/.t3/userdata/state.sqlite exits 1 (89.587247ms)
 ✔ T3: with --t3-db, a symlinked file or directory exits 1 (180.000997ms)
 ✔ T3 WAL: the newest message only in -wal, writer attached, is counted (117.32872ms)
 ✔ T3 WAL: stopped cleanly, counts are correct and the main file is unchanged (131.14448ms)
 ✔ T3 WAL: -wal without -shm in a writable directory is read (151.206765ms)
 ✔ T3 WAL: -wal without -shm in a read-only directory exits 1 (138.257245ms)
 ✔ T3 WAL: stopped cleanly in a read-only directory exits 1 (126.693454ms)
 ✔ T3: a lock held past the 5 s busy timeout exits 1 and names --no-t3 (5132.393612ms)
 ✔ a done row whose closing issue is open is a violation; a row that is not done is not (1.841977ms)
 ✔ an issue several rows close is expected closed only once all of them are done (0.491939ms)
 ✔ closure needs positive evidence: unknown is undecided, and so is a skipped or short issue check (0.38455ms)
 ✔ each owner of an in-progress or in-review row gets one liveness class (7.577627ms)
 ✔ a required row not done after 14 days is a violation; a legacy row uses genesis as its lower bound (0.408923ms)
 ✔ an ISO requiredSince, as `set required` writes it, ages from its UTC day; one that does not parse is a violation (0.219338ms)
 ✔ the text section always ends in a count and a result, and never prints a full pass (0.328283ms)
 ✔ pidAlive: a running pid is present, an exited one is gone, and EPERM still means present (23.340081ms)
 ✔ issue states: open list first, then the metric page, then at most 10 lookups (308.946454ms)
 ✔ a full open list: lookups settle what it leaves out, and only an unsettled issue keeps it undecided (395.879857ms)
 ✔ the open list refuses on a failed call or a bad record, and never echoes the helper (212.507314ms)
 ✔ a helper call past the deadline is killed with its child, and the call reports it (2009.325987ms)
 ✔ readQueue loads queue.json through the queue validator and refuses anything else (116.571991ms)
 ✔ protected changes list every in-range entry that changes a required or parked row (254.072185ms)
 ✔ the CLI prints the queue section above the weekly table and under a queue key in --json (549.337582ms)
 ✔ a queue with nothing wrong prints 0 violations and a reduced pass, never a full pass (213.387035ms)
 ✔ --no-issues makes no call and leaves the issue checks undecided; --no-queue skips the section (277.27238ms)
 ✔ the CLI refuses a bad queue before any call, and a failed open list with exit 2 (243.798875ms)
 ✔ --unsupported-runtime repeats once per seat and takes a seat name (309.141792ms)
 ✔ an unreadable config makes every owner invalid instead of passing them (129.721951ms)
 ℹ tests 78
 ℹ suites 0
 ℹ pass 78
 ℹ fail 0
 ℹ cancelled 0
 ℹ skipped 0
 ℹ todo 0
 ℹ duration_ms 11441.871862
@@ -0,0 +1,77 @@
 ✔ pure resolution selects current default or explicit enrolled account (1.872151ms)
 ✔ scope is explicit, bounded and never inferred (1.57115ms)
 ✔ fork pin is preserved against default change, override, missing account and revocation (0.704128ms)
 ✔ unenrolled account/provider, missing harness, model expansion and native model ceiling refuse (0.798909ms)
 ✔ only explicit synthetic credential forms and internal fixture stores admitted (5.36031ms)
 ✔ two concurrent workspaces of the same agent publish distinct complete private generations (161.818945ms)
 ✔ same execution ID is exclusively claimed and cannot overwrite a published generation (42.892165ms)
 ✔ failed generation after-auth preserves prior files, records failure and refuses blind same-ID retry (65.595802ms)
 ✔ failed generation before-publish preserves prior files, records failure and refuses blind same-ID retry (77.784406ms)
 ✔ credential lock contention refuses without duplicate side effects (14.592571ms)
 ✔ symlinked pre-existing final target is refused and never followed (114.334933ms)
 ✔ invalid registry cannot resolve; no fallback to supplied partial entries (0.27403ms)
 ✔ post-publication failure records uncertainty, preserves complete generation and prevents replay (35.053473ms)
 ✔ expired credentials refresh under transaction and subsequent generation reuses rotation (172.252181ms)
 ✔ refresh failure retains prior generation and store state (115.992743ms)
 ✔ refresh timeout retains prior generation and store state (175.78184ms)
 ✔ refresh malformed retains prior generation and store state (82.415683ms)
 ✔ concurrent refresh on same account refuses contention while unrelated account proceeds (217.852814ms)
 ✔ invalid refresh options refuse before burning claim (42.224182ms)
 ✔ fixed fake process rotates both OAuth fields without mutating caller input (30.575047ms)
 ✔ concurrent isolated processes preserve separate provider credentials (70.011095ms)
 ✔ fake failure is refused with fixed diagnostics (23.924708ms)
 ✔ fake malformed is refused with fixed diagnostics (46.619583ms)
 ✔ fake timeout is refused with fixed diagnostics (103.993257ms)
 ✔ fake unchanged is refused with fixed diagnostics (24.438804ms)
 ✔ caller executable/environment injection is rejected before spawning (0.319162ms)
 ✔ valid fixture tree validates and lists without secrets (69.435664ms)
 ✔ unknown-field refuses (0.368527ms)
 ✔ invalid-id refuses uppercase and traversal shapes (0.267159ms)
 ✔ plain-http baseUrl requires allowInsecureTransport (0.299317ms)
 ✔ native provider rejects allowInsecureTransport (0.241272ms)
 ✔ unsupported credential type and kind refuse (0.159833ms)
 ✔ account provider-path mismatch refuses (0.109991ms)
 ✔ profile account refs must be provider/account shaped (0.219682ms)
 ✔ seat selection accepts fork pin field, validates account refs (0.208864ms)
 ✔ harness manifest id must equal executable (gate 1) (0.20672ms)
 ✔ CLI validate: duplicate provider id across files refuses (32.824146ms)
 ✔ CLI validate: missing referenced provider/account refuse (33.993274ms)
 ✔ CLI validate: broken JSON refuses without secret echo (28.59761ms)
 ✔ CLI usage errors exit 2 (59.606687ms)
 ✔ credential.json sibling presence does not break validation and is never read (68.475171ms)
 ✔ D1 missing, empty and structurally empty roots refuse, no list projection (194.464649ms)
 ✔ D1 required directory auth cannot be absent (64.16133ms)
 ✔ D1 required directory auth/providers cannot be absent (63.723043ms)
 ✔ D1 required directory auth/accounts cannot be absent (63.719991ms)
 ✔ D1 required directory auth/settings cannot be absent (63.575918ms)
 ✔ D1 required directory harnesses cannot be absent (63.409486ms)
 ✔ D1 root file and unreadable metadata refuse (109.21742ms)
 ✔ D2 no symlink traversal at auth/providers/openai-codex.json (56.303987ms)
 ✔ D2 no symlink traversal at auth/accounts/openai-codex/homelab-openai (65.600399ms)
 ✔ D2 no symlink traversal at auth/providers (54.789867ms)
 ✔ D2 no symlink traversal at auth (59.71393ms)
 ✔ D2 root and ancestor symlinks and lexical traversal refuse (171.864478ms)
 ✔ private filesystem modes enforced for root (67.447829ms)
 ✔ private filesystem modes enforced for auth (68.390099ms)
 ✔ private filesystem modes enforced for auth/providers/openai-codex.json (64.223364ms)
 ✔ private filesystem modes enforced for auth/accounts/openai-codex/homelab-openai/account.json (82.608422ms)
 ✔ D3 numeric version 1 only across all record kinds (1.62563ms)
 ✔ D4 nested unknown keys and missing per-kind required fields refuse (75.984001ms)
 ✔ D5 unenrolled default refuses even when account exists (78.872984ms)
 ✔ D6 provider/account credential type must match (68.848194ms)
 ✔ D7 every harness endpoint enforces HTTP opt-in and shape (0.482101ms)
 ✔ D8 URLs reject embedded credentials and unsupported protocols without echo (178.618889ms)
 ✔ D9 malformed JSON diagnostics contain no content excerpt (73.23815ms)
 ✔ D10 missing metadata is missing-path, not invalid-json (67.801324ms)
 ✔ D10 library returns no partial entries on any invalid record (73.320294ms)
 ✔ null/scalar/array metadata refuses without stack or echo (264.773454ms)
 ✔ credential sibling is never opened, even when an unreadable symlink (36.871642ms)
 ✔ oversized metadata refuses before parsing (66.670211ms)
 ℹ tests 69
 ℹ suites 0
 ℹ pass 69
 ℹ fail 0
 ℹ cancelled 0
 ℹ skipped 0
 ℹ todo 0
 ℹ duration_ms 2358.596664
@@ -0,0 +1,158 @@
 ✔ F1: an ordinary commit after update-ref is refused until step 8; then it commits and the queue stays at C (1368.524395ms)
 ✔ F1: a plain `commit -e` whose guard ran before update-ref fails at its own HEAD update (1107.815614ms)
 ℹ git commit -e: index.lock free during the editor
 ✔ F1: a `commit -e -- path` whose guard ran before update-ref fails at its own HEAD update (1053.859957ms)
 ℹ git commit -e -- src.txt: index.lock held during the editor
 ✔ F1: step 8 with index.lock held exits 3, and ordinary commits stay refused until the printed command runs (1315.196413ms)
 ✔ F1: HEAD moving after the step-7 guard check and before update-ref: refused, nothing published (1160.527505ms)
 ✔ F1: H is recorded before the canary, so HEAD moving during the step-1 canary is refused at step 7 (1107.586369ms)
 ✔ F1: a queue commit landing after H is recorded: step 1 says HEAD moved, not the guard (1534.627974ms)
 ✔ F1: a queue commit landing between the HEAD check and the canary: the failed clean run is reported as HEAD moved (934.919964ms)
 ✔ F1: a shared-index change during the procedure is not committed (1106.384292ms)
 ✔ F1: a queue path staged after update-ref: step 8 stops and touches nothing (1034.784534ms)
 ✔ F1: a missing or a different hook refuses (881.067414ms)
 ✔ F1: same bytes without the exec bit, a symlinked hook, and core.hooksPath in the local or global scope each refuse before update-ref (1206.112179ms)
 ✔ F1: the canary refuses a hook that git would not run (700.887362ms)
 ✔ F1: the guard deactivated after step 1 is refused at the step-7 recheck (1025.963075ms)
 ✔ bootstrap: implementation-only HEAD, the guard, genesis, the --genesis commit, then an extending commit (974.769984ms)
 ✔ bootstrap: --genesis with a base present, no base without --genesis, an op before the first commit, a changed map, another branch (1049.631216ms)
 ✔ bootstrap: the archived tests and validator run outside any repository (965.953223ms)
 ✔ general: an unrelated staged file stays staged, and the committed blobs are the snapshot bytes (917.850172ms)
 ✔ general: a queue write after the snapshot is not committed (1337.162804ms)
 ✔ general: a snapshot whose log does not extend the base refuses (1279.192635ms)
 ✔ general: install-hook privilege, repair of a same-bytes hook, and its refusals (237.3212ms)
 ✔ general: environment overrides, a linked worktree and usage (596.349331ms)
 ✔ general: a queue path staged before the run refuses at step 1 (681.715215ms)
 ✔ general: HEAD's queue tests failing in the archive refuse (910.425138ms)
 ✔ genesis document serializes deterministically and replays (7.054109ms)
 ✔ a hand edit that stays valid JSON fails replay; a formatting-only edit fails re-serialization (3.014575ms)
 ✔ a tampered result, receipt or viewSha fails replay (4.082269ms)
 ✔ op ids: 8 to 72 characters for callers, 80 in the log for .outcome entries (0.269546ms)
 ✔ add: defaults for an ordinary seat, privileged extras, refusals (5.909142ms)
 ✔ matrix: queued→briefed privileged; briefed→in-progress owner with after satisfied (7.861542ms)
 ✔ matrix: release, review round, changes requested and waiting-on-jason (20.322188ms)
 ✔ matrix J5: in-review→done by the gate owner with evidence naming the current round (17.233235ms)
 ✔ review issue, lead decision 23: none refuses, one is used, several need --issue, later rounds keep it (36.694668ms)
 ✔ the row schema refuses a round with a null issue, and the A1 review shape (P2) (4.077834ms)
 ✔ matrix R1: every state × target × actor class matches 8.7, gate owner jason or not, required or not (1594.6255ms)
 ✔ matrix: blocked keeps the claim and returns only to previousState (4.122793ms)
 ✔ matrix J4: parking is Jason's, refused while required; unpark returns to queued (2.723419ms)
 ✔ field edits: who may change what (5.395692ms)
 ✔ set issues keeps a logged narrowing of closes (N10) (3.281972ms)
 ✔ text the table shows refuses \ and <, everywhere it enters (N8) (1.962337ms)
 ✔ every accepted text renders to nine cells on every row (N8) (24.65681ms)
 ✔ genesis: the map refuses an owner among its row's reviewers; replay doesn't (2026-10-04) (0.566795ms)
 ✔ times and dates must be calendar values, not just the shape (2026-10-04) (2.504801ms)
 ✔ replay holds every op id to the caller's rule (N11) (9.205636ms)
 ✔ note: owner, listed reviewer or privileged; empty clears (1.274341ms)
 ✔ assign moves the claim with the owner; done clears it (2.904442ms)
 ✔ render is byte-stable and escapes pipes (0.588507ms)
 ✔ view classification: current, genuine stale, edited stale marker, changed current body, markers (0.744893ms)
 ✔ next: resume, then review, then start, then wait, then nothing; lowest id first (25.079881ms)
 ✔ canonical args make a retry's identity independent of list order (0.330732ms)
 ✔ manifests, headings and blob ids (0.483773ms)
 ✔ the migration map: one queue-map block, exact keys (0.588646ms)
 ✔ every call but `queue` reaches the seat CLI exactly as before A2 (749.22435ms)
 ✔ `queue` reaches the queue CLI with the rest of the arguments (210.45014ms)
 ✔ the pre-A2 fixture is the script A2 changed (0.378694ms)
 ✔ acquire publishes the record by link; release removes only its own lock (8.169093ms)
 ✔ a kill between the temp write and the link leaves no lock (71.284163ms)
 ✔ a short or failed temp write refuses and leaves no lock and no temp (4.246816ms)
 ✔ a link error other than EEXIST refuses (2.629069ms)
 ✔ an error after the link releases the lock: unreadable gate, failing temp stat (5.709139ms)
 ✔ a release that fails on a gate path is reported, never a stack trace (P1) (7.955663ms)
 ✔ a paused holder: another writer waits 10 s, then refuses naming it live (10073.647915ms)
 ✔ two concurrent unlockers: the second refuses on the gate (27.810421ms)
 ✔ a writer publishing during an unlock, lock first: unlock sees it live and refuses (1.976767ms)
 ✔ a writer publishing during an unlock, gate first: the writer releases and refuses (1.948203ms)
 ✔ a gate swapped while held is left in place and reported, on success and on refusal (N1) (2.299562ms)
 ✔ a reused pid within one boot is mismatch; unlock removes the lock and never signals the process (3.735687ms)
 ✔ the same pid and start on a different boot is mismatch (0.45282ms)
 ✔ a foreign host is unknown whatever the local pid says; unlock refuses (58.060779ms)
 ✔ unreadable /proc: classification is unknown and acquire refuses (1.143527ms)
 ✔ invalid records: empty, unparsable, wrong keys, bad start or boot (0.280584ms)
 ✔ a stale gate blocks writers; --check-gate says mismatch for a reused pid (4.074518ms)
 ✔ a delayed release by a dead owner, after unlock and a new owner: the inode check keeps the new lock (2.189275ms)
 ✔ release checks the inode too: a byte-identical lock file with a new inode is left in place (1.390838ms)
 ✔ unlock refuses a live, unknown or invalid lock, and does nothing without one (3.773733ms)
 ✔ the migration map validates and renders the golden genesis table (5.562141ms)
 ✔ the marked QUEUE.md holds every row and parked item between its markers (1.878447ms)
 ✔ map-check reports each kind of drift (7.292576ms)
 ✔ a request posts once as the requester; a retry sends nothing (688.350161ms)
 ✔ each transport answer maps to posted, failed or uncertain (8.9 step 3) (3473.819954ms)
 ✔ the pre-send checks: GET user must name the requester, under the deadline (1226.019299ms)
 ✔ the lead's request refuses a token for login sage (572.249013ms)
 ✔ the credential file: the seat's own, 0600, no symlink, never the shared default (503.984503ms)
 ✔ an unresolved request blocks a new request, a new round, waiting-on-jason and done (1722.410185ms)
 ✔ a same-op retry after a kill sends nothing, even with a stale view (2421.424105ms)
 ✔ a held lock at the outcome exits 3 and names what the transport said (570.240504ms)
 ✔ late outcomes: after an abandon, and after a resolve with the same or another id (1743.338011ms)
 ✔ resolve checks the comment: issue, markers, round, candidate and author (1601.710919ms)
 ✔ the lead resolves a seat's request: the comment's author is the requester, fetched with the lead's token (576.148221ms)
 ✔ validateRow checks a request round's shape, which every replayed entry must keep (391.275684ms)
 ✔ request, changes, a new candidate, approval: every round pinned; no review files (1594.71266ms)
 ✔ a row with no reviewers opens a round that posts nothing (1017.17584ms)
 ✔ verify-commit: a prospective tree must hold exactly the candidate's paths (1232.03807ms)
 ✔ semantics: v1 entries replay as before; review entries need v2 (404.516761ms)
 ✔ set reviewers refuses the row's owner (2026-09-28) (294.923201ms)
 ✔ the owner records no verdict, even as a listed reviewer (377.194736ms)
 ✔ a request comment over the length limit is not sent (419.581132ms)
 ✔ a late POST on a closed row leaves a conflict nothing can resolve, and resolve asks nothing (508.094636ms)
 ✔ a Jason-gated row reaches waiting-on-jason only on every reviewer's approval (2142.907828ms)
 ✔ genesis: refusals before anything is written (512.662775ms)
 ✔ genesis: the map must be committed, well formed, with committed briefs and seats (661.798245ms)
 ✔ genesis: markers, a stray witness, once only; a retry returns the receipt (570.512249ms)
 ✔ genesis: a file holding genesis alone with no witness is confirmed by sync or a retry (326.662108ms)
 ✔ canonical checks: worktree, second clone, detached HEAD, wrong branch, GIT_DIR, foreign code; a symlink works (715.634667ms)
 ✔ --by that differs from MOSAIC_AGENT_NAME warns on stderr and logs nothing more (N12) (534.717688ms)
 ✔ op ids: missing, too long, reserved; a retry answers; another payload refuses (710.319686ms)
 ✔ a retried add returns the id it first allocated, after reassignment and after done (810.716943ms)
 ✔ Rocko's S4 schedule: a lost result, another writer, then the retry opens no second round (529.80669ms)
 ✔ the review issue and the evidence round through the CLI (lead decision 23, 8.7) (931.381579ms)
 ✔ claims and add defaults through the CLI; candidates are manifests or reachable commits (656.148184ms)
 ✔ add, set reviewers and assign refuse the row's owner as a reviewer (445.647197ms)
 ✔ the working-brief check: a changed working copy refuses the start and flags next (651.123258ms)
 ✔ next: resume first, then nothing for an idle seat; needs a seat (323.006358ms)
 ✔ view stale: new ops and verify refuse naming the unshown op; retries answer; reads warn; render fixes (630.609898ms)
 ✔ view unknown: a hand edit, an old marker over an edited body, missing or duplicate markers (1084.349585ms)
 ✔ a hand edit to queue.json refuses every verb, reads included (594.164641ms)
 ✔ verify and render --check leave bytes and mtimes unchanged (389.003599ms)
 ✔ render is byte-stable across runs and repositories (264.819899ms)
 ✔ snapshot and verify --snapshot (770.221731ms)
 ✔ usage errors exit 4 (681.438982ms)
 ✔ a short write, ENOSPC or a file fsync failure: nothing visible, temp removed (232.825599ms)
 ✔ a rename failure: nothing visible, temp removed (145.840545ms)
 ✔ a directory fsync failure: uncertain, exit 3, no receipt; the tail refuses new ops; a retry confirms (175.377624ms)
 ✔ a directory fsync failure, then sync names the op (334.200058ms)
 ✔ a witness write failure: uncertain, durable, exit 3; the view is untouched (146.849109ms)
 ✔ the .git fsync after the witness rename fails: uncertain, exit 3, the witness says so (154.073374ms)
 ✔ confirming a tail fsyncs queue.json and docs/plans before the witness; either failure changes nothing (159.117564ms)
 ✔ the docs/plans fsync after the view rename fails: the op stands, the view is written, a warning says so (152.332799ms)
 ✔ a lock swapped while held is left in place and reported, on a receipt and on a refusal (148.389128ms)
 ✔ a lock that cannot be released after an op is reported, on a receipt and on a refusal (151.546104ms)
 ✔ unlock prints a swapped gate's warning on stderr, the result on stdout (115.474484ms)
 ✔ a view write that fails keeps the op and reports a stale view (114.610379ms)
 ✔ SIGKILL before the rename: nothing recorded; the retry removes the leftover temp (615.523099ms)
 ✔ SIGKILL after the rename, before the witness: the tail refuses new ops and sync names the op (653.62181ms)
 ✔ SIGKILL after the witness, before the view: the stale refusal names the op (620.210512ms)
 ✔ SIGKILL after the view, before the receipt: the retry returns the receipt (665.935427ms)
 ✔ git checkout between steps 1 and 7: step 7 refuses and nothing is written (263.734749ms)
 ✔ git stash restoring an older valid pair: history lost; accept-history needs privilege, a reason and --yes (920.323427ms)
 ✔ a deleted witness: refused after the locked recheck; accept-history records it absent (208.428956ms)
 ✔ a header edit during a write: the op stands, the view write is skipped with a warning (133.27623ms)
 ✔ a reader paused between the witness and the file while a writer finishes: no lost-history report (142.279617ms)
 ✔ file-then-witness order forced by a hook: the locked recheck prevents a false report (138.591803ms)
 ✔ a writer paused before and after the witness rename: readers see a tail, then a match (139.175204ms)
 ✔ a true rollback is reported only after the locked recheck; a held lock names its holder instead (518.066437ms)
 ✔ an accept-history in progress: an unlocked reader waits on the lock and never reports lost history (465.309105ms)
 ✔ the platform check refuses other filesystems (130.878385ms)
 ✔ tmpfs passes only a test layer that allows it (N5) (156.961291ms)
 ✔ unlock keeps a multi-line lock record on stdout (P3) (222.786106ms)
 ℹ tests 148
 ℹ suites 0
 ℹ pass 148
 ℹ fail 0
 ℹ cancelled 0
 ℹ skipped 0
 ℹ todo 0
 ℹ duration_ms 24575.018457
@@ -0,0 +1,27 @@
 ✔ resolveSeat: by name under --repo resolves the repo layout (1.251101ms)
 ✔ resolveSeat: by path resolves the fleet layout (0.314635ms)
 ✔ resolveSeat: refusals for missing dir, missing launch.sh, non-executable launch.sh, invalid name, and unknown layout (0.710458ms)
 ✔ tmuxContext: outside tmux, default socket, custom socket, and exec failure (0.647878ms)
 ✔ makeRegistration produces a record that validates; each shape violation throws SeatError (0.790986ms)
 ✔ writeRegistration/readRegistration: round trip, permissions, absence, and malformed records (1.269425ms)
 ✔ updateTask: changes task, taskSetBy and updatedAt only, and refuses appropriately (1.156407ms)
 ✔ CLI launch: registers, execs the fake launch script, and passes args through (30.222001ms)
 ✔ CLI launch: --harness lands in the record (32.913646ms)
 ✔ CLI launch: the launch script's own exit code passes through (29.484011ms)
 ✔ CLI launch: relaunching a seat rewrites the one registration record (61.925976ms)
 ✔ CLI launch: omitting --task records an empty string, not null (27.792171ms)
 ✔ CLI seat task: updates only the task after a launch, and refuses on an unlaunched seat (82.182296ms)
 ✔ CLI refusals: no args, unknown flag, missing config, already-registered env, and exec failure (148.150447ms)
 ✔ samePath: equal paths, symlinked dirs, distinct dirs, and non-strings (0.786055ms)
 ✔ resolveSetBy: explicit --by wins over the environment; absent or empty environment gives unknown; invalid explicit or environment values refuse with exit 4 (0.674817ms)
 ✔ validateRegistration/readRegistration: taskSetBy is optional; a record without it (written before #1511) still loads unchanged; an invalid one is refused; the version does not change (0.868062ms)
 ✔ updateTask: records setBy, preserves startedAt and every unrelated field, upgrades an old record in place only when the task is set, and replaces a previous attribution (11.536829ms)
 ✔ CLI seat task: --by beats MOSAIC_AGENT_NAME, the environment beats nothing, empty environment is unknown, invalid --by or environment refuses with exit 4 and leaves the record byte for byte (311.700226ms)
 ℹ tests 19
 ℹ suites 0
 ℹ pass 19
 ℹ fail 0
 ℹ cancelled 0
 ℹ skipped 0
 ℹ todo 0
 ℹ duration_ms 811.678252
@@ -0,0 +1,59 @@
 ✔ the boot config is checked before anything starts (31.406315ms)
 ✔ a business with no tracker entry refuses task verbs (10.50476ms)
 ✔ credential.expiring and .expired are recorded once per instance (31.40099ms)
 ✔ a token file that changes on disk records credential.changed (12.76132ms)
 ✔ autostart polls, reconciles and retries a startup the tracker was down for (36.972972ms)
 ✔ a refusal a restart must clear is not retried by the poll (11.379556ms)
 ✔ a poll that fires while two are queued is dropped (15.039808ms)
 ✔ close waits for a running verb and refuses one that has not started (43.104982ms)
 ✔ the bundled Vikunja is the pinned upstream image the runbook names (0.83167ms)
 ✔ every published port is on 127.0.0.1, and no secret is in the file (0.311917ms)
 ✔ the fake answers each route with the statuses and shapes Vikunja v2.7.0 sent (106.64679ms)
 ✔ the recorded task bodies pass the checks S3 applies to every read (0.5836ms)
 ✔ the client works against the fake over real HTTP with the platform fetch (103.710016ms)
 ✔ a correct install starts, and the first reconcile records tasks that already exist (26.761043ms)
 ✔ verbs refuse while a business is starting and after startup refused it (17.920027ms)
 ✔ startup refuses a token that can do more than its role needs (38.784719ms)
 ✔ startup refuses an unsupported version and flags an untested one (29.454798ms)
 ✔ startup refuses a board that the runbook did not install (39.884452ms)
 ✔ startup refuses a project the sync bot cannot read (6.805487ms)
 ✔ startup refuses a configured label the pm bot cannot see (8.89736ms)
 ✔ startup refuses an expired credential and a missing sync credential (17.507495ms)
 ✔ an unreachable tracker refuses with tracker-unavailable (7.515047ms)
 ✔ an edit in the UI is recorded once, with the fields that changed (88.08522ms)
 ✔ a move between open buckets is seen on the board, though updated does not change (39.239909ms)
 ✔ a person's comment is counted and a bot's is not (70.85552ms)
 ✔ the hourly reconcile catches a comment through comment_count (79.111948ms)
 ✔ a task closed in the UI leaves the open view with its done bucket (112.694106ms)
 ✔ a task that leaves the board is recorded as deleted, moved or out of reach (66.256236ms)
 ✔ a poll that read before a verb wrote does not overwrite the verb (36.027847ms)
 ✔ a tracker fault during a tick is reported and the next tick catches up (36.364956ms)
 ✔ a malformed answer refuses the tick with tracker-shape (13.83003ms)
 ✔ no token value reaches the database, the log or a refusal (56.431652ms)
 ✔ the first look at a task counts only comments inside the window (35.769886ms)
 ✔ task.create needs a recorded human request and a requirement id (41.974426ms)
 ✔ only labels named in the business file can be written (26.40169ms)
 ✔ task.schedule sets and clears a due date and relations (34.940101ms)
 ✔ assign and reassign move the role bots and record task.assigned (59.327811ms)
 ✔ task.update.assigned is for the assignee and records task.state (53.98079ms)
 ✔ a wrong expected digest records task.conflict and writes nothing (17.784255ms)
 ✔ a cross-role verb needs a resolved decision, used once (22.522257ms)
 ✔ task.close needs a verdict; after it every verb refuses with task-done (23.350917ms)
 ✔ a lost answer is settled by a re-read and never retried (25.303349ms)
 ✔ a create whose answer is lost is reported uncertain, and the poll finds the task (35.162388ms)
 ✔ a task the sync bot cannot read refuses and records nothing (15.624289ms)
 ✔ verbs and polls for one business run one at a time (91.847876ms)
 ✔ a due date with milliseconds is written to the second (65.82116ms)
 ✔ every write landed and the final read failed: the verb succeeds and records what it wrote (44.19338ms)
 ✔ some writes landed and the final read failed: write-uncertain, and nothing is recorded (16.809815ms)
 ✔ a create whose final read fails succeeds and records task.created (35.830588ms)
 ✔ an edit between the last write and the final read shows as external on the next poll (38.316051ms)
 ✔ task.created is recorded when a later label write fails (15.830074ms)
 ℹ tests 51
 ℹ suites 0
 ℹ pass 51
 ℹ fail 0
 ℹ cancelled 0
 ℹ skipped 0
 ℹ todo 0
 ℹ duration_ms 776.554894
@@ -0,0 +1,23 @@
 ✔ browser edge states: loading, empty, malformed, stale, hostile/long values, in-flight reply and appearance fallback (2593.877023ms)
 Rendered contrast: {"failures":[],"count":330,"lowest":4.504658476260286}
 ✔ served Console browser: real board fixtures, keyboard, drafts, receipts, themes, 320px and failures (2819.101922ms)
 ✔ conversation view: full history, collapsed tools, hidden thinking, inert hostile content, malformed and reconcile markers (2487.022264ms)
 ✔ conversation view: a fork keeps the open branch, says so, and opens the new one on request (1409.838855ms)
 ✔ conversation view: a newer session with no readable history keeps the marker (1033.313886ms)
 ✔ conversation view: seats without history say so and offer no reply (510.394246ms)
 ✔ Discord row through real board/WebUI: independent brake/liveness, no Reply, literal content (2198.027385ms)
 ✔ return flow through the conversation view: send, tool call, delayed result, peer message, exact long answers, relaunch (53388.794719ms)
 ✔ both presentations replace old activity with relaunch notice, label retained history, then resume after new activity (2301.993318ms)
 ✔ reported return flow and relative Age: reply sent from the inspector, then the new answer appears there without manual refresh (21973.89591ms)
 ✔ loopback host and board origin fail closed (10.966556ms)
 ✔ real board fixture passes through WebUI; assets and isolated seen/reply work (111.519321ms)
 ✔ proxy preserves exact request bytes, status and receipt, rejects forms and malformed JSON, never follows redirect (60.651787ms)
 ✔ unreachable board reports URL; CLI rejects unsupported options (969.115867ms)
 ℹ tests 14
 ℹ suites 0
 ℹ pass 14
 ℹ fail 0
 ℹ cancelled 0
 ℹ skipped 0
 ℹ todo 0
 ℹ duration_ms 53754.387454
@@ -0,0 +1,17 @@
 OK    status with missing harness credential exits 3 and still lists accounts
 OK    status reports harness credential (read-only) + mosaic accounts
 OK    api key material never reaches output
 OK    oauth token material never reaches output
 OK    unparseable credential file exits 2
 OK    symlinked credential file exits 4
 OK    env-side credential names reported
 OK    env var values never reach output
 OK    accounts without an accounts dir reports none and creates nothing
 OK    accounts lists files and marks the active one
 OK    loose account perms flagged in listing
 OK    agent --auth with missing account file refuses (exit 4)
 OK    agent --auth with non-0600 account file refuses
 OK    agent --auth with invalid account name refuses
 OK    auth.sh without valid config refuses
 selftest: 15 passed, 0 failed
@@ -0,0 +1,55 @@
 Note: switching to '521597bbe0cf51a0a1aa3b7a16f13fc7b59c98c6'.
 You are in 'detached HEAD' state. You can look around, make experimental
 changes and commit them, and you can discard any commits you make in this
 state without impacting any branches by switching back to a branch.
 If you want to create a new branch to retain commits you create, you may
 do so (now or later) by using -c with the switch command. Example:
  git switch -c <new-branch-name>
 Or undo this operation with:
  git switch -
 Turn off this advice by setting config variable advice.detachedHead to false
 Not currently on any branch.
 nothing to commit, working tree clean
 Note: switching to '521597bbe0cf51a0a1aa3b7a16f13fc7b59c98c6'.
 You are in 'detached HEAD' state. You can look around, make experimental
 changes and commit them, and you can discard any commits you make in this
 state without impacting any branches by switching back to a branch.
 If you want to create a new branch to retain commits you create, you may
 do so (now or later) by using -c with the switch command. Example:
  git switch -c <new-branch-name>
 Or undo this operation with:
  git switch -
 Turn off this advice by setting config variable advice.detachedHead to false
 OK    dry-run: allowed change, exit 0, nothing committed (exit 0)
 OK    dry-run committed nothing
 OK    apply: allowed change exits 0 (exit 0)
 OK    apply: attribution in commit subject
 OK    apply: target tree clean after commit
 OK    disallowed path refused (exit 1)
 OK    disallowed path: target untouched
 OK    syntax gate refused broken .mjs (exit 1)
 OK    syntax gate: target untouched
 OK    suite failure refused (exit 1)
 OK    suite failure: target reverted to clean
 OK    disabled policy refused (exit 2)
 OK    disabled policy: target untouched
 OK    failed run refused (exit 1)
 OK    failed run: target untouched
 OK    missing run exits 4 (exit 4)
 OK    invalid policy exits 2 (exit 2)
 selftest: 17 passed, 0 failed
@@ -0,0 +1,26 @@
 OK    absent adapter defaults to pi
 OK    adapter mock validates (exit 0)
 OK    unsupported adapter exits 2 (exit 2)
 OK    env exports adapter
 OK    bootstrap creates default when absent (exit 0)
 OK    bootstrap wrote config file
 OK    bootstrap is idempotent on existing config (exit 0)
 OK    bootstrap did not rewrite existing config
 OK    validate missing config exits 3 (exit 3)
 OK    malformed JSON exits 2 (exit 2)
 OK    unsupported configVersion exits 2 (exit 2)
 OK    unknown top-level key exits 2 (exit 2)
 OK    unknown execution key exits 2 (exit 2)
 OK    unsupported backend exits 2 (exit 2)
 OK    unsupported environment exits 2 (exit 2)
 OK    relative dataRoot exits 2 (exit 2)
 OK    non-canonical dataRoot exits 2 (exit 2)
 OK    filesystem root dataRoot exits 2 (exit 2)
 OK    home directory dataRoot exits 2 (exit 2)
 OK    dataRoot containing config dir exits 2 (exit 2)
 OK    control character in provider exits 2 (exit 2)
 OK    symlinked config file exits 2 (exit 2)
 OK    env exports resolve correctly
 OK    failed validation modified nothing
 selftest: 24 passed, 0 failed
@@ -0,0 +1,70 @@
 toolchain: node v26.8.1
 OK    syntax: packages/discord/src/approvals.mjs
 OK    syntax: packages/discord/src/authorize.mjs
 OK    syntax: packages/discord/src/binding.mjs
 OK    syntax: packages/discord/src/cli.mjs
 OK    syntax: packages/discord/src/connector.mjs
 OK    syntax: packages/discord/src/context.mjs
 OK    syntax: packages/discord/src/engine-pi.mjs
 OK    syntax: packages/discord/src/errors.mjs
 OK    syntax: packages/discord/src/gateway.mjs
 OK    syntax: packages/discord/src/git.mjs
 OK    syntax: packages/discord/src/journal.mjs
 OK    syntax: packages/discord/src/notify.mjs
 OK    syntax: packages/discord/src/rest.mjs
 OK    syntax: packages/discord/src/setspark.mjs
 OK    syntax: packages/discord/src/tools.mjs
 OK    syntax: packages/discord/src/web.mjs
 OK    syntax: packages/discord/bin/git-credential.mjs
 OK    syntax: packages/discord/extension/tools.mjs
 OK    syntax: packages/discord/tests/approvals.test.mjs
 OK    syntax: packages/discord/tests/authorize.test.mjs
 OK    syntax: packages/discord/tests/binding.test.mjs
 OK    syntax: packages/discord/tests/connector.test.mjs
 OK    syntax: packages/discord/tests/context.test.mjs
 OK    syntax: packages/discord/tests/engine.test.mjs
 OK    syntax: packages/discord/tests/fake-pi.mjs
 OK    syntax: packages/discord/tests/gateway.test.mjs
 OK    syntax: packages/discord/tests/git.test.mjs
 OK    syntax: packages/discord/tests/helpers.mjs
 OK    syntax: packages/discord/tests/journal.test.mjs
 OK    syntax: packages/discord/tests/notify.test.mjs
 OK    syntax: packages/discord/tests/recover.test.mjs
 OK    syntax: packages/discord/tests/rest.test.mjs
 OK    syntax: packages/discord/tests/setspark.test.mjs
 OK    syntax: packages/discord/tests/tools.test.mjs
 OK    syntax: packages/discord/tests/web.test.mjs
 OK    syntax: packages/discord/fixtures/claim-worker.mjs
 OK    syntax: packages/discord/fixtures/legacy-owner-worker.mjs
 OK    syntax: scripts/discord.sh
 OK    syntax: scripts/discord-service.sh
 OK    packages/discord declares no dependencies
 OK    no bot-token-shaped string in packages/discord
 OK    fixture binding uses placeholder ids only
 OK    fixture binding validates
 OK    real pi with the extension exposes exactly list_dir, read_file, search and no built-in tool
 OK    real pi with a writable root exposes exactly the three reads plus write_file and edit_file, and writes nothing at start
 OK    real pi with a web key exposes the three reads plus web_fetch and web_search, and no write tool without a writable root
 OK    real pi with a git root exposes the reads, writes and the four git verbs, commits nothing at start, and never shows the token
 OK    real pi with protocol vault adds reserve_id to the git verbs
 OK    real pi with a setspark key exposes the reads and the eight record verbs, no counters, and never shows the key
 OK    real pi refuses a git key on a read-only root (fail closed)
 OK    real pi with the pilot flags (--no-tools) exposes no tool at all
 OK    real pi exits non-zero without MOSAIC_DISCORD_TOOLS: no session, no tools (fail closed)
 OK    a failing nested test fails the run under a parent runner's NODE_TEST_CONTEXT
 OK    node --test packages/discord/tests/ (ℹ pass 178)
 OK    scripts/discord.sh --help exits 0
 OK    scripts/discord.sh check without a binding exits 4
 OK    scripts/discord.sh recover without a binding exits 4
 OK    scripts/discord.sh reload without a binding exits 4
 OK    scripts/discord-service.sh without a command exits 4
 OK    service unit renders with the repository path, a supervised run as the main process, exit 3 never retried, and reload as SIGHUP
 OK    service install writes the rendered unit (0644) and leaves no temp file
 OK    service install a second time reports unchanged
 OK    systemd-analyze verify accepts the rendered unit
 OK    service uninstall removes the unit file
 OK    service install with an unknown flag exits 4
 OK    service install with USER unset finishes and names the account for lingering
 discord suite: 66 passed, 0 failed
@@ -0,0 +1,21 @@
 OK    initial ordinary-file install
 OK    installed tree matches canonical source
 OK    installed tree has no symlinks
 OK    check detects installation drift
 OK    sync refuses to overwrite installation drift
 OK    check detects an extra destination file
 OK    check detects an extra destination directory
 OK    check rejects a destination symlink
 OK    sync accepts a canonical source update
 OK    updated installation matches canonical source
 scripts/test-extension-package.sh: line 14: 4117866 Killed                     "$@" > /dev/null 2>&1
 OK    forced interruption kills the replacing process
 OK    next invocation recovers old consistent installation
 OK    interrupted replacement rolled back
 OK    sync succeeds after interruption recovery
 OK    unlocked stale lock file does not block
 OK    active lock refuses a concurrent sync
 OK    source symlink fails closed
 OK    nested second entrypoint fails closed
 extension package selftest: 18 passed, 0 failed
@@ -0,0 +1,53 @@
 toolchain: node v26.8.1, python 3.12.8, jsonschema 4.26.0
 OK    syntax: scripts/foundation-inspect.mjs
 OK    syntax: scripts/foundation/strict-json.mjs
 OK    syntax: scripts/foundation/canonical.mjs
 OK    syntax: scripts/foundation/resolve.mjs
 OK    syntax: scripts/foundation/validate-record.mjs
 OK    syntax: scripts/foundation/fixtures/build-fixtures.mjs
 OK    syntax: scripts/foundation/canonical.test.mjs
 OK    syntax: scripts/foundation/cli.test.mjs
 OK    syntax: scripts/foundation/fixtures.test.mjs
 OK    syntax: scripts/foundation/resolve.test.mjs
 OK    syntax: scripts/foundation/strict-json.test.mjs
 OK    syntax: scripts/foundation/verify-schema.py (ast only; no bytecode written)
 OK    fixture generator runs
 OK    checked-in fixtures/bundles equal a fresh generation
 OK    checked-in fixtures/raw equal a fresh generation
 OK    checked-in fixtures/index.json equal a fresh generation
 OK    checked-in demo bundles equal a fresh generation
 OK    a failing nested test fails the run under a parent runner's NODE_TEST_CONTEXT
 OK    node --test scripts/foundation/ (ℹ pass 80)
 OK    differential schema oracle: PASS: differential schema oracle (finite corpus; compatibility evidence, not equivalence proof)
      platform witness: strftime('%Y') for year 999 -> '999' (pinned checker refuses years 0001..0999)
      node v26.8.1; corpus 1568 records (38 pinned fixtures, 478 unique bundle records, 1052 typeCase/mutation/lexical cases)
      schema column: agree-valid 540, agree-invalid 991, DISAGREEMENTS 0; strict-only (parser-bound) cases: 27; unsupported-kind records not schema-assessed by the inspector: 10
      profile column (schema-valid records only): profile-valid 510, profile-invalid 30
      profile refusals asserted: 30 schema-agreed-valid records refused only by the strict typed-string profile (rule profile-pattern-mismatch), 12 declared by name; 73 named probes verified against declared schema/profile columns
 OK    oracle: zero schema-column disagreements with the pinned checker
 OK    oracle: strict-only profile refusals are counted and asserted
 OK    demo: permitted read preview exits 0 (exit 0)
 OK    demo: permitted file.change preview exits 0 (exit 0)
 OK    demo: assignment.change proposal is unresolved (exit 3) (exit 3)
 OK    demo: revoked registration is refused (exit 3) (exit 3)
 OK    demo: message is not authority (exit 3) (exit 3)
 OK    usage: no arguments exits 2 (exit 2)
 OK    io: missing file exits 4 (exit 4)
 OK    io: directory exits 4 (exit 4)
 OK    io: symlink exits 4 (O_NOFOLLOW) (exit 4)
 OK    bound: oversize fixture exits 2 (exit 2)
 OK    profile: one final LF in a typed selection id is refused before admission (exit 2) (exit 2)
 OK    profile: two final LFs fail the schema pattern itself (exit 2) (exit 2)
 OK    profile: escaped newlines in free-form text stay allowed (exit 0) (exit 0)
 OK    profile refusal is invalid-request/profile-pattern-mismatch with selection and operation withheld, value not echoed
 OK    text output starts with the disclaimer
 OK    json output is valid JSON with result allowed and exactly the charter §7 fields
 OK    json golden matches byte-for-byte
 OK    sandboxed bundle run (env -i, PATH=/nonexistent) produced the unresolved proposal
 OK    sandbox inventory (path/type/size/mode/uid/gid/inode/mtime/sha256) unchanged by runs
 OK    canary never printed (bundle run and credential-file run)
 OK    a non-bundle JSON file is refused at the shape gate, not read into output
 OK    no field of the non-bundle file is echoed
 selftest: 44 passed, 0 failed
@@ -0,0 +1,35 @@
 toolchain: node v26.8.1, git version 2.55.0
 OK    syntax: packages/queue/src/cli.mjs
 OK    syntax: packages/queue/src/errors.mjs
 OK    syntax: packages/queue/src/io.mjs
 OK    syntax: packages/queue/src/lock.mjs
 OK    syntax: packages/queue/src/queue.mjs
 OK    syntax: packages/queue/src/review.mjs
 OK    syntax: packages/queue/src/store.mjs
 OK    syntax: packages/queue/tests/commit.test.mjs
 OK    syntax: packages/queue/tests/data.test.mjs
 OK    syntax: packages/queue/tests/dispatch.test.mjs
 OK    syntax: packages/queue/tests/helpers.mjs
 OK    syntax: packages/queue/tests/lock.test.mjs
 OK    syntax: packages/queue/tests/migration.test.mjs
 OK    syntax: packages/queue/tests/review.test.mjs
 OK    syntax: packages/queue/tests/store.test.mjs
 OK    syntax: packages/queue/tests/write.test.mjs
 OK    syntax: packages/queue/tests/fixtures/fake-gitea.mjs
 OK    syntax: packages/queue/tests/fixtures/kill-at.mjs
 OK    syntax: packages/queue/tests/fixtures/lock-child.mjs
 OK    syntax: packages/queue/tests/fixtures/mosaic-pre-a2.sh
 OK    syntax: scripts/queue-commit.sh
 OK    syntax: scripts/git-hooks/pre-commit
 OK    syntax: scripts/mosaic
 OK    queue-commit.sh, the guard and scripts/mosaic are executable
 OK    packages/queue declares no dependencies
 ℹ tests 148
 ℹ pass 148
 ℹ fail 0
 OK    node --test packages/queue/tests/
 OK    scripts/mosaic queue help
 skip  queue verify and render --check: this checkout (/home/jwoltje/filbert-scratch/r45/cand) is not the queue's canonical root (/mnt/storage/src/mosaic-stack)
 queue suite: 27 passed, 0 failed
@@ -0,0 +1,7 @@
 OK    valid RELEASE resolves (exit 0)
 OK    invalid RELEASE exits 1 (exit 1)
 OK    missing RELEASE exits 1 (exit 1)
 OK    valid RELEASE leaves image tag consistent with version
 skip state-machine cases (docker daemon unavailable)
 selftest: 4 passed, 0 failed
@@ -0,0 +1,33 @@
 OK    valid task validates (exit 0)
 OK    unknown task key exits 2 (exit 2)
 OK    unsupported taskVersion exits 2 (exit 2)
 OK    invalid task id exits 2 (exit 2)
 OK    empty prompt exits 2 (exit 2)
 OK    NUL in expectExact exits 2 (exit 2)
 OK    out-of-range timeout exits 2 (exit 2)
 OK    missing mission file exits 4 (exit 4)
 OK    task with valid mission validates (exit 0)
 OK    invalid mission exits 2 (exit 2)
 OK    validate missing task exits 4 (exit 4)
 OK    validation does not modify the task file
 OK    prune dry-run exits 0 (exit 0)
 OK    dry-run deleted nothing
 OK    prune --keep=2 --yes removes oldest (exit 0)
 OK    kept exactly 2 newest runs
 OK    newest run kept, oldest pruned
 OK    append-only receipt written (3 entries)
 OK    sessions/workspaces untouched by prune
 OK    prune with invalid keep exits 4 (exit 4)
 skip adapter seam cases (docker daemon unavailable)
 skip workspace/capability cases (docker daemon unavailable)
 skip live task cases (docker unavailable)
 OK    onboard without name exits 4 (non-interactive) (exit 4)
 OK    onboard --name renders profile (exit 0)
 OK    profile written
 OK    canon structure: required filled, optional placeholdered
 OK    canon sections present
 FAIL user recall run succeeds (exit 1)
 FAIL recalled user name (response: )
 OK    no agent identity on headless run
 selftest: 26 passed, 2 failed
@@ -0,0 +1,42 @@
 r1-cand node-business exit 0 
 r1-cand node-bus exit 0 
 r1-cand node-cli exit 0 
 r1-cand node-control-board exit 0 
 r1-cand node-conversation exit 1 
 r1-cand node-discord exit 0 
 r1-cand node-ledger exit 0 
 r1-cand node-mosaic exit 0 
 r1-cand node-queue exit 0 
 r1-cand node-seat exit 0 
 r1-cand node-tasks exit 0 
 r1-cand node-webui exit 0 
 r1-cand suite-auth exit 0 load 5.12
 r1-cand suite-conductor exit 0 load 5.03
 r1-cand suite-config exit 0 load 4.63
 r1-cand suite-discord exit 0 load 4.58
 r1-cand suite-extension-package exit 0 load 4.58
 r1-cand suite-foundation exit 0 load 3.87
 r1-cand suite-queue exit 0 load 5.10
 r1-cand suite-release exit 0 load 5.10
 r1-cand suite-task exit 1 load 5.10
 r1-base node-business exit 0 
 r1-base node-bus exit 0 
 r1-base node-cli exit 0 
 r1-base node-control-board exit 0 
 r1-base node-conversation exit 1 
 r1-base node-discord exit 0 
 r1-base node-ledger exit 0 
 r1-base node-mosaic exit 0 
 r1-base node-queue exit 0 
 r1-base node-seat exit 0 
 r1-base node-tasks exit 0 
 r1-base node-webui exit 0 
 r1-base suite-auth exit 0 load 11.24
 r1-base suite-conductor exit 0 load 10.66
 r1-base suite-config exit 0 load 10.66
 r1-base suite-discord exit 0 load 11.32
 r1-base suite-extension-package exit 0 load 11.32
 r1-base suite-foundation exit 0 load 6.34
 r1-base suite-queue exit 0 load 5.93
 r1-base suite-release exit 0 load 5.54
 r1-base suite-task exit 1 load 5.54
@@ -0,0 +1,8 @@
 packages/cli/README.md: OK
 packages/cli/src/host.mjs: OK
 packages/cli/src/notifier.mjs: OK
 packages/cli/tests/host.test.mjs: OK
 packages/cli/tests/notifier.test.mjs: OK
 packages/cli/tests/trackers-boot.test.mjs: OK
 packages/discord/tests/journal.test.mjs: OK
 scripts/bus-service.sh: OK
@@ -0,0 +1,32 @@
 N2 killed (fail 1, cancelled 0)
 N4 killed (fail 1, cancelled 0)
 N5 killed (fail 1, cancelled 0)
 M28 killed (fail 1, cancelled 0)
 G150 killed (fail 1, cancelled 0)
 G144 killed (fail 1, cancelled 0)
 F2a killed (fail 1, cancelled 0)
 F2b killed (fail 1, cancelled 0)
 F2c killed (fail 2, cancelled 0)
 F2d killed (fail 1, cancelled 0)
 F2e killed (fail 1, cancelled 0)
 F2f killed (fail 1, cancelled 0)
 J4a killed (fail 1, cancelled 0)
 J4b killed (fail 1, cancelled 0)
 J4c killed (fail 1, cancelled 0)
 J4d killed (fail 1, cancelled 0)
 E1 killed (fail 1, cancelled 0)
 E2 killed (fail 1, cancelled 0)
 X1 killed (fail 1, cancelled 0)
 X2 SURVIVED
 X3 killed (fail 3, cancelled 0)
 X4 killed (fail 1, cancelled 0)
 X5 killed (fail 1, cancelled 0)
 X6 killed (fail 1, cancelled 0)
 X7 killed (fail 1, cancelled 0)
 X8 killed (fail 1, cancelled 0)
 X9 SURVIVED
 X10 killed (fail 1, cancelled 0)
 X11 killed (fail 1, cancelled 0)
 X12 killed (fail 3, cancelled 0)
 X13 killed (fail 2, cancelled 0)
 X14 SURVIVED
@@ -0,0 +1,249 @@
 v24.21.0
 ✔ inbox lists only decisions routed to the human, with what approving authorizes and how to decide (169.210077ms)
 ✔ decide resolves by id prefix with --yes, and the trail shows it in broker order with the task to follow (220.995357ms)
 ✔ decide refuses without a terminal or --yes, on an unknown option and on a short reference (221.951051ms)
 ✔ decide prints a declining choice as declining (167.525958ms)
 ✔ an unknown outcome is reported once and never resent (127.301829ms)
 ✔ a decision closed before the answer arrives exits 2 and points at its trail (161.285204ms)
 ✔ a prefix that matches two open decisions exits 2 and resolves neither (122.126437ms)
 ✔ without --business a command uses the live host's business, and a stale host.json is not a host (129.876238ms)
 ✔ every human command refuses inside an agent run before it touches the bus (119.673429ms)
 ✔ usage errors exit 4; no business and no host is a usage error (166.691645ms)
 ✔ agents and tasks print through the broker (186.413662ms)
 ✔ notify.json: missing, loose, malformed and extra keys refuse; a binding or null passes (13.333788ms)
 ✔ bootConfig builds the broker's boot message for one business, with no trackers key when no project names one (201.980972ms)
 ✔ trackers come from the tracker.* variables of the one project that names a tracker project (114.238503ms)
 ✔ with two projects, the one that sets tracker.project supplies the tracker and the other is no conflict (83.289259ms)
 ✔ two projects that each name a tracker project refuse, since the boot shape holds one (75.913435ms)
 ✔ a business without tracker.baseUrl gets no trackers entry (120.454836ms)
 ✔ an unknown business and a broken system config refuse with exit 3 (173.234133ms)
 ✔ empty views say so (2.235721ms)
 ✔ the trail keeps the broker's order and names a decision's task without its rows (2.711433ms)
 ✔ tasks print the tracker fields the snapshot carries (0.326825ms)
 ✔ the host boots the broker, binds a launch in process, and the notifier DMs a blocking decision exactly once (1420.380971ms)
 ✔ a notifier that dies takes the host down with exit 1, so the unit restarts the pair (526.895008ms)
 ✔ a second host for the same data root refuses with exit 3 while the first runs (205.432629ms)
 ✔ a notifier that refuses stops the broker and the host refuses with exit 3 (380.380756ms)
 ✔ a notifier that refuses after the broker died still refuses with exit 3, without a send to the dead broker (401.224717ms)
 ✔ a notifier that dies before it replies, after the broker died, still refuses, without a send to the dead broker (193.95973ms)
 ✔ watchChildren reports a child that died before it was called, and one that dies later (30.977871ms)
 ✔ bus stop refuses to signal a live pid that is not a bus host (206.813112ms)
 ✔ bus start refuses with exit 3 and the code when the broker refuses to boot; bus status names the lock (314.80728ms)
 ✔ bus start refuses with exit 3 without a notifier config (148.547702ms)
 ✔ bus start runs until bus stop; status reports it while it runs (917.572739ms)
 ✔ bus-service.sh renders the unit and installs it into a given directory (55.222104ms)
 ✔ zoned uses the IANA zone across DST (174.293637ms)
 ✔ each open blocking decision is DM'd once, across polls and a restart; non-blocking ones are not (189.689945ms)
 ✔ two blocking decisions get two DMs with different nonces (185.5709ms)
 ✔ the digest nonce differs per business and per day and fits Discord's 25 characters (0.474682ms)
 ✔ a failed DM is journaled, backs off, and is retried until it lands (228.801997ms)
 ✔ five definite refusals stop a DM: one gave-up line, one log line, and a restart keeps the count (185.603648ms)
 ✔ 429s, 5xx-style unknowns and refusals without a status never count toward the limit (160.573693ms)
 ✔ a crash between the fifth refusal and its gave-up line: the next poll appends it and sends nothing (128.228554ms)
 ✔ the digest goes at 08:00 Chicago once a day, with blocking ones marked as DM'd (155.668925ms)
 ✔ a late start with no digest for the day sends one at once; an empty inbox gets one line (167.201859ms)
 ✔ an inbox read failure is logged and the next poll retries (2.445857ms)
 ✔ no Discord id reaches the journal or the log (151.557087ms)
 ✔ the journal: a torn tail is copied out and truncated, so an append after it reopens cleanly (33.621183ms)
 ✔ the journal: a crash between the copy and the truncate leaves a tail the next open repairs (88.02178ms)
 ✔ the journal: a whole file that is one torn line truncates to empty (24.4463ms)
 ✔ the journal: a malformed complete line refuses and leaves the file and any torn tail alone (19.133951ms)
 ✔ the journal: a loose file mode, a loose directory or a symlinked journal refuses (5.649955ms)
 ✔ the journal: a line with a wrong type refuses with exit 3 and names the field (21.592611ms)
 ✔ the journal: a symlinked directory refuses and says it is a link (2.071451ms)
 ✔ the journal: an append after the file was swapped for a symlink refuses and writes nothing through it (3.932517ms)
 ✔ the journal: a directory it cannot write or create refuses with exit 3 and names the path (2.475167ms)
 ✔ digest content stays within Discord's 2000 characters (0.631769ms)
 ✔ runLoop never overlaps ticks and stops after the one in flight (124.297708ms)
 task.close {} answered: invalid-request; fake saw 18 requests, first GET /info 200, GET /projects/1 200, GET /projects/1/views 200
 task.close on a missing task answered: task-not-found; it made GET /tasks/999 404
 ✔ bootConfig trackers reach the S3 adapter in the real broker child, which goes ready against a fake Vikunja (976.590424ms)
 ✔ the transport writes {business, verb, args} to the child and reads its JSON (209.432574ms)
 ✔ a bus code on stderr becomes the exit code; garbage and timeouts are outcome problems (2384.766754ms)
 ✔ busExit and refuseInsideAgent (0.613505ms)
 ✔ approvals: a request is validated before anything is posted; the rendering shows names and never ids (12.852656ms)
 ✔ approvals: the ledger is appended and folded into open requests with bind and approval states (5.782138ms)
 ✔ approvals: a reply approves only when it points at a request, says exactly approve, and comes from a listed approver once (0.867481ms)
 ✔ approvals: a button approves only on its own request message with the matching custom id (0.677824ms)
 ✔ approvals flow: a turn that opened a request posts the message with the button, records it, binds it, and both approvers approve (44.404919ms)
 ✔ approvals flow: a non-approver, a repeat, a wrong custom id and a service refusal each get their fixed line and a drop entry (24.440255ms)
 ✔ approvals flow: an invalid request from the model, a refused post, and no api client are recorded and post nothing (15.489524ms)
 ✔ approvals flow: start retries a bind and an approval left as unknown, under their original keys (27.563545ms)
 ✔ authorize: open channel, listed user (16.150485ms)
 ✔ authorize: wrong guild (0.28161ms)
 ✔ authorize: no guild (DM) (0.225398ms)
 ✔ authorize: unlisted channel (1.378769ms)
 ✔ authorize: unknown channel, no info (0.846771ms)
 ✔ authorize: thread of listed parent (0.639398ms)
 ✔ authorize: thread of unlisted parent (0.491093ms)
 ✔ authorize: text channel that is not a thread and not listed (0.248551ms)
 ✔ authorize: unlisted user (0.249877ms)
 ✔ authorize: no author (0.556257ms)
 ✔ authorize: bot author (listed id, bot flag) (0.192138ms)
 ✔ authorize: system author (0.166337ms)
 ✔ authorize: the bot itself (0.116806ms)
 ✔ authorize: webhook (0.113884ms)
 ✔ authorize: mention channel without mention (0.188505ms)
 ✔ authorize: mention channel with bot mention (0.203378ms)
 ✔ authorize: mention channel with @everyone only (0.169766ms)
 ✔ authorize: mention channel mentioning someone else (0.112607ms)
 ✔ authorize: mention channel, content says @bot but mentions empty (0.139715ms)
 ✔ authorize: private thread under mention channel, mentioned (0.12079ms)
 ✔ authorize: private thread under mention channel, not mentioned (0.100549ms)
 ✔ authorize: thread in another guild per channel info (0.242092ms)
 ✔ authorize: not an object (1.111897ms)
 ✔ authorize: no id (0.1492ms)
 ✔ authorize: oversize content is accepted and flagged (0.110482ms)
 ✔ authorize: exactly the limit is not oversize (0.086483ms)
 ✔ authorize: a user's channel allowlist drops them outside it, threads count as the parent, others are unaffected (0.684236ms)
 ✔ authorize: order puts wrong guild before user, and user before channel (no channel lookup for strangers) (0.216255ms)
 ✔ binding: a complete binding validates and is frozen (11.883208ms)
 ✔ binding: unknown key, missing field, wrong type refuse with exit 2 (7.763507ms)
 ✔ binding: empty allowlists refuse (0.506707ms)
 ✔ binding: a user's channel allowlist must be non-empty, listed and unique; absent means every listed channel (1.64819ms)
 ✔ reloadDiff: reloadable keys are summarised by id; every fixed key refuses with exit 2 (5.565414ms)
 ✔ binding: dmRecipient is optional, a snowflake, one of the listed users, and a fixed key (7.762208ms)
 ✔ binding: file must be 0600, regular, not a symlink (5.293905ms)
 ✔ binding: token file mode, symlink, emptiness and shape are checked; token never appears in errors (2.823072ms)
 ✔ cli: check refuses a non-0600 token file with exit 2 before any network use (220.470979ms)
 ✔ context files: absolute paths, traversal, symlinks and out-of-repo targets refuse; in-repo files resolve (3.235281ms)
 ✔ cli: check refuses a missing context file and a missing binding with exit 2; usage is exit 4 (885.135928ms)
 ✔ cli: reload validates the file first (exit 2), then needs a live owner (exit 1); usage is exit 4 (415.829812ms)
 ✔ cli: run refuses when STOP is present, before any network use (277.976045ms)
 ✔ binding: tools is optional, validated strictly, a fixed key for reload, and its roots are resolved against the data root (3.601177ms)
 ✔ binding: a git key is validated at load and reaches the extension whole, and only on a writable root (2.952646ms)
 ✔ delivery: an accepted message is in the inbox before the turn, the reply is chunked with one nonce per chunk, and the turn record is write-once (37.062415ms)
 ✔ delivery: refused and unknown outcomes are journaled; a later chunk is not sent after a failure (48.925104ms)
 ✔ delivery: restart with an unknown entry re-sends the same nonce once and reconciles before accepting traffic (2.735505ms)
 ✔ delivery: an unknown entry older than the dedupe window is marked refused, not re-sent; a still-unknown one refuses start (4.826431ms)
 ✔ delivery: repeated unknown reconciliations never refresh the dedupe window; the original intent time decides (7.131981ms)
 ✔ turn: a failed engine turn posts the fixed line, never model output, and writes a failed record (7.448548ms)
 ✔ turn: a second message during a turn is held by the engine, both get their own reply and record (34.250691ms)
 ✔ turn: a thread under a listed channel is answered in the thread; an unknown thread is looked up once (6.149321ms)
 ✔ drop: an unlisted user gets silence and one drop line; no inbox entry, no REST call, no engine call (6.309747ms)
 ✔ drop: an oversize message is accepted into the inbox, answered with the fixed line and journaled as a drop (12.979224ms)
 ✔ restart: an inbox with three ids and a replay of the same three produces zero turns (48.344945ms)
 ✔ stop: STOP present refuses start; STOP written while running refuses new turns and the current one finishes (79.624664ms)
 ✔ ceiling: the ceiling plus one is refused and journaled; one fixed line per UTC day; a new day accepts again (12.828338ms)
 ✔ ceiling: a burst arriving while turns are still running cannot queue past the ceiling (12.537168ms)
 ✔ ceiling: a turn interrupted by a crash still counts after restart; admissions are durable (9.427092ms)
 ✔ ceiling: the daily notice survives a same-day restart; one delivery attempt in total, even when the first attempt crashed mid-flight (8.457426ms)
 ✔ duplicate: the same event delivered twice while the thread lookup is held yields one prompt, one admission and one reply (4.386104ms)
 ✔ journal: no token-shaped string and no model output on the drop path reaches disk (0.981586ms)
 ✔ receipt: an admitted message gets one eyes reaction on the inbound message; drops and refusals get none; a failed reaction is recorded and does not fail the turn (2.448548ms)
 ✔ receipt: Discord refusing the reaction leaves the turn intact and records ok false (5.770748ms)
 ✔ reload: a new user is silent before and answered after; a removed channel goes silent; a lower ceiling applies at once (16.068602ms)
 ✔ reload: a fixed key refuses with exit 2 and the old binding stays in force (2.733839ms)
 ✔ tools: with a tools binding the turn record lists every read and its outcome; without one the field is null (5.840293ms)
 ✔ context: the Discord block names the server, channels and modes, and states the rules from Q15 and Q16 (10.650128ms)
 ✔ context: with tools the block names the roots, keeps file content as data, and says to state refusals plainly (2.780979ms)
 ✔ context: a writable root adds the write rules and says a write is real only once Jason commits (4.29911ms)
 ✔ context: the envelope is one bracketed line then the text; names cannot break the line (1.58776ms)
 ✔ context: a git root swaps the terminal-commit line for the git verbs, and a vault root adds the id protocol (8.457708ms)
 ✔ context: assembleContext concatenates files in launcher format and appends the block; sha256 is stable (1.601151ms)
 ✔ context: splitReply keeps paragraphs together under the limit and splits long ones at lines, spaces, then hard (4.966235ms)
 ✔ engine: buildPiArgs carries the fixed flags, engine settings, session dir and prompt file (3.068728ms)
 ✔ engine: with tools, buildPiArgs turns pi's own tools off, loads the extension explicitly and allowlists exactly our three (0.97972ms)
 ✔ engine: a run with tool turns settles once, on the answer, with every tool call in the result (118.279655ms)
 ✔ engine: a run that ends on a tool-only turn fails the prompt as empty; a retried run settles on the real end (90.864625ms)
 ✔ engine: one prompt, one turn, text and usage come back (82.471638ms)
 ✔ engine: a prompt while streaming is held until pi settles, then sent as its own run, and answered in order (374.102779ms)
 ✔ engine: a held prompt that times out before pi settles fails on its own and is never sent (325.526012ms)
 ✔ engine: timeout sends abort and fails only that turn; the process stays (125.370647ms)
 ✔ engine: tool events from a run that outlived its timeout never land in the next prompt's record (266.818641ms)
 ✔ engine: a prompt after a turn that timed out before its agent_start waits for pi to settle instead of being refused (173.810103ms)
 ✔ engine: when pi has not started a timed-out turn by the end of the abort grace, the engine stops pi and fails held prompts (215.100324ms)
 ✔ engine: a timed-out turn pi starts only after the grace never answers a later prompt (622.159336ms)
 ✔ engine: late events of a run past its grace, before pi exits, answer nothing and nothing more is sent (early prompt response) (3.420612ms)
 ✔ engine: late events of a run past its grace, before pi exits, answer nothing and nothing more is sent (late prompt response) (1.316421ms)
 ✔ engine: a timed-out run pi did start outlives the grace; the next prompt goes out when it ends (466.085165ms)
 ✔ engine: a malformed JSONL line fails the turn, not the process (50.432037ms)
 ✔ engine: a turn that ends in error rejects with the error code; process exit fails pending turns (59.967268ms)
 ✔ gateway: hello -> identify with intents, ready, heartbeat with jitter, ack (3.410353ms)
 ✔ gateway: missed ack closes the socket and resumes with the last sequence (2.021625ms)
 ✔ gateway: op 7 reconnect resumes; op 9 non-resumable re-identifies (0.699336ms)
 ✔ gateway: op 9 resumable resumes (0.428487ms)
 ✔ gateway: close 4014 is fatal, reports the missing intent, never reconnects (1.052509ms)
 ✔ gateway: 4004 and 4013 are fatal too; 1006 reconnects with identify when no session (0.679693ms)
 ✔ gateway: close() is final and unparseable frames are ignored (0.502568ms)
 ✔ git: config validation is strict, needs write: true, a work tree and a private token file (285.019903ms)
 ✔ git: the child environment drops every host git config, names one helper, and carries the token path only for origin (136.347733ms)
 ✔ git: status reports the branch, ahead/behind and changed paths, and refuses off the named branch or mid-merge (272.428902ms)
 ✔ git: parseStatus reads porcelain v2 including renames and conflicts (1.350846ms)
 ✔ git: a commit stages exactly the named files, carries the seat author and the requester trailer, and pushes at once (397.671011ms)
 ✔ git: commit refusals: message, paths, requester, nothing to commit, and an index that already holds other work (284.970897ms)
 ✔ git: a commit whose push fails is still a commit, says so, and the next commit's push carries both (D6) (226.383507ms)
 ✔ git: pull is fast-forward only; a diverged origin or dirty local files refuse with nothing merged (405.146352ms)
 ✔ git: push pushes the named branch only and reports up to date (110.552937ms)
 ✔ git: no token value or token path ever reaches a git argument list; outputs are masked and capped (133.506908ms)
 ✔ git: the credential helper answers get over https from a private file and nothing else (356.776939ms)
 ✔ git: the vault protocol validates before a commit, honours another owner's lock, reserves ids, and locks around writes (608.652604ms)
 ✔ lock: the claim is exclusive; a second start against a live owner refuses (9.207667ms)
 ✔ lock: a stale lock (dead owner, reused pid, or record without start) refuses run and is never signaled; only unlock clears it (137.107672ms)
 ✔ lock: an incomplete claim (directory without owner record) is busy and refuses run; unlock clears it (1.449643ms)
 ✔ lock: an owner record that exists but cannot be read is invalid: never signaled, never removed, never claimed over (6.372436ms)
 ✔ lock: legacy upgrade; a live connector holding a {pid, start} record is unknown, unlock refuses and nothing changes; after it exits, unlock clears it (81.131754ms)
 ✔ lock: a live pid whose record carries a malformed or noncanonical start or boot string is unknown, not a mismatch; nothing signals, removes, or claims over it (821.907635ms)
 ✔ lock: identity syntax; only canonical unsigned decimal start ticks and lowercase boot uuids are identities (1.169369ms)
 ✔ lock: a process whose start marker or boot id cannot be read refuses to claim (44.588245ms)
 ✔ lock: a live pid whose identity cannot be read right now is unknown: never signaled, never removed, never claimed over (3.410371ms)
 ✔ lock: four processes racing for the same binding; exactly one claims it and the others refuse (102.949689ms)
 ✔ lock: stale handoff; concurrent starts over a stale lock all refuse, nothing reclaims, one unlock then exactly one live owner (258.084232ms)
 ✔ lock: four-party schedule; claims landing inside an unlock's gap never survive, one unlock leaves no owner and no residue (176.49723ms)
 ✔ notices: a kind is recorded per UTC day and found again (1.053503ms)
 ✔ notify: the DM channel opens once, every send carries the nonce, and only the message id comes back (10.557984ms)
 ✔ notify: refusals and unknowns surface as RestOutcome without ids; a refused send reopens the channel next time (3.7748ms)
 ✔ notify: no dmRecipient, a non-0600 binding or token, or a missing binding refuse before any network use (4.592249ms)
 ✔ recover: nothing to do is clean; a lock whose owner is gone or that has no record is cleared and STOP ends up absent (75.563709ms)
 ✔ recover: an operator STOP refuses with exit 3 and is never removed, whatever the lock says (61.586462ms)
 ✔ recover: a brake written during the unlock wins; STOP stays with both lines and the start is refused (56.812654ms)
 ✔ recover: a held binding refuses with exit 3 and writes no STOP: live owner, alive pid without verifiable identity, unreadable record (207.860034ms)
 ✔ cli: recover exits 0 when ready, 3 behind a brake or a held binding, and run's own STOP refusal is 3 (1510.849171ms)
 ✔ rest: createMessage sends nonce, enforce_nonce, empty allowed_mentions and a soft reply reference (5.531414ms)
 ✔ rest: 429 waits retry_after and retries; 4xx is refused; 5xx and socket errors are unknown (3.650468ms)
 ✔ rest: content and nonce limits are enforced locally; typing never throws (0.932379ms)
 ✔ rest: react PUTs the encoded emoji on the inbound message for @me; 2xx is true, anything else is false and never throws (0.954723ms)
 ✔ rest: createDm posts the recipient and resolves the channel id; 4xx refused, 5xx unknown, bad id never sent (2.00831ms)
 ✔ setspark config: a bare https or loopback origin, a private key file, a principal (10.254659ms)
 ✔ setspark config: reaches the tools config and the binding as a fixed key (4.714592ms)
 ✔ setspark config: the binding's key survives resolveToolRoots and the engine's JSON hand-off to the extension (2.900566ms)
 ✔ setspark config: approvers come from the binding's users, never from the binding's setspark key (3.163685ms)
 ✔ setspark verbs: required_approvers go out as discord ids from names and come back as names (84.247403ms)
 ✔ setspark verbs: no Discord user id reaches tool text, whatever shape the service returns it in (24.969462ms)
 ✔ setspark contract: a decision made with names opens a request the connector accepts; names stored by an old record still refuse (14.721499ms)
 ✔ setspark keys: read per call, one printable token per file, rotation without a restart (17.167329ms)
 ✔ setspark idempotency keys: principal, turn id, call index; connector keys name a step (3.160519ms)
 ✔ setspark http core: json in and out, bearer header, idempotency header, fixed user agent, no key anywhere else (9.572244ms)
 ✔ setspark http core: error bodies become fixed refusals with code and the 409 fields; server text is data, cut (1042.749224ms)
 ✔ setspark verbs: a setspark key enables the eight verbs and no counters (6.688099ms)
 ✔ setspark verbs: writes carry the turn's key and the asserted requester, reads carry no key, and the api key never appears in text or details (13.997535ms)
 ✔ setspark verbs: no turn refuses every write before any request; bad arguments refuse before any request; reads still work (8.115785ms)
 ✔ setspark verbs: renderRecord caps long output and hides the accepted snapshot (1.376507ms)
 ✔ setspark api: bind, add_approval (button and reply) and get use integer request ids and the connector's keys (7.689174ms)
 ✔ tools: config refuses a missing, symlinked, dotted, non-directory or duplicate root and bad limits (8.854561ms)
 ✔ tools: every escape is refused with a fixed reason and nothing outside the root is read (17.340196ms)
 ✔ tools: happy paths list, read a window, and search case-insensitively; dotfiles and symlinks never appear (9.218508ms)
 ✔ tools: the tool set renders text for the model, records details for the journal, and enforces the per-run budget (10.983915ms)
 ✔ tools: listing and search caps hold (32.381356ms)
 ✔ tools: credential shapes are caught; ordinary prose and ids are not (7.784107ms)
 ✔ tools: the read uses the checked file itself; a symlink, a swapped file, a FIFO, a grown file or a hard link at read time is refused (35.08534ms)
 ✔ tools: an unreadable file under the root is skipped by search and refused by read (6.387642ms)
 ✔ tools: config accepts write: true only as a boolean, and enables the write tools only then (6.48996ms)
 ✔ tools: every write outside the fence is refused before any byte lands, and no temp file remains (26.868965ms)
 ✔ tools: write_file leaves the exact bytes, edit_file replaces one exact match, and the set renders the change as uncommitted (15.200985ms)
 ✔ tools: a target that changed between the check and the rename is refused and the temp file is removed (6.559184ms)
 ✔ web: config takes an https or loopback-http SearXNG base url and a bounded fetch cap (13.287236ms)
 ✔ web: address rules refuse every private, loopback, link-local, mapped and multicast form (4.204019ms)
 ✔ web: web_fetch refuses bad urls, private hosts, rebinding names, non-https redirects, too many hops, error status, non-text bodies, and times out (1105.741841ms)
 ✔ web: web_fetch returns html as text with the title, follows an https redirect, keeps plain text and json, and cuts at the cap (16.202457ms)
 ✔ web: html to text drops scripts, styles and comments, decodes entities and keeps block breaks (0.804354ms)
 ✔ web: web_search asks the instance for json, returns at most ten clean results, and refuses a bad query, a down instance or an unusable answer (10.647487ms)
 ✔ web: the tool set enables the web tools only with a web key, counts them in the budget, and records url, status and hits (8.722026ms)
 ℹ tests 238
 ℹ suites 0
 ℹ pass 238
 ℹ fail 0
 ℹ cancelled 0
 ℹ skipped 0
 ℹ todo 0
 ℹ duration_ms 5109.413025
@@ -0,0 +1,25 @@
 PROBE T1 sends at +0s +30s +90s +210s +450s
 PROBE T1 gave-up at +450s = 7.5 min
 PROBE T2 after 4 h: outcomes refused,refused,refused,refused,refused,gave-up
 PROBE T2 sends 5
 PROBE T3 sends before restart 2 after one tick post-restart 3 (same clock second)
 PROBE T4 dm sends 17 gave-up false
 PROBE T5 digest sends in 1 h 7 gave-up false
 PROBE A1 tick {"dms":1,"digest":false,"failed":0} line {"at":"2026-10-08T05:00:00.000Z","kind":"dm","decision":"624d0407-56e6-4686-b5af-21dbc41002fb","outcome":"confirmed","messageId":123}
 PROBE A1 reopen 3 notify journal line 1 is malformed (messageId): <root>/notify/demo/sent.jsonl
 PROBE A2 ok sent d1 days 2026-10-08 refusals [["d2",1]]
 ✔ F2-T1 time from first refusal to gave-up, polling every POLL_MS (86.95343ms)
 ✔ F2-T2 binding fixed after 60 min: the DM never lands (172.659536ms)
 ✔ F2-T3 restart drops the backoff: next attempt is immediate (60.500209ms)
 ✔ F2-T4 500s for 6 h: no gave-up (243.575372ms)
 ✔ F2-T5 digest refusals: retried, never gave-up (75.006372ms)
 ✔ J4-A1 append does not type-check: a numeric messageId bricks the next open (61.251173ms)
 ✔ J4-A2 a line the row 39 writer produced still opens (57.483971ms)
 ℹ tests 7
 ℹ suites 0
 ℹ pass 7
 ℹ fail 0
 ℹ cancelled 0
 ℹ skipped 0
 ℹ todo 0
 ℹ duration_ms 844.895562
@@ -0,0 +1,33 @@
 nocb connected before send true
 nocb process exit code 1
 node:events:505
    throw er; // Unhandled 'error' event
    ^
 -- exit 0
 nocb connected before send true
 nocb process exit code 1
 node:events:505
    throw er; // Unhandled 'error' event
    ^
 -- exit 0
 nocb connected before send true
 nocb process exit code 1
 node:events:505
    throw er; // Unhandled 'error' event
    ^
 -- exit 0
 cb connected before send true
 cb callback got EPIPE
 cb process exit code 3
 -- exit 0
 cb connected before send true
 cb callback got EPIPE
 cb process exit code 3
 -- exit 0
 cb connected before send true
 cb callback got EPIPE
 cb process exit code 3
 -- exit 0
@@ -0,0 +1,212 @@
 # Row 45, S4 follow-up, round 1 review (Filbert)
 Issue #1527, request comment 26869, queue rev 207. Packet:
 `agents/rocko/work/s4-follow-up/BUILD.md`. Candidate: manifest
 `candidate-manifest.sha256`, sha256
 `b329fdcbdf8cb62659f239359570dfa9771559e3112b743df55429244769ed14`, 8
 files, over `521597bb` with `build.patch` (sha256
 `4ed9ff61b94b4c9d2426e5703c2e6c0dbcd6bc0c64a88d59347a8b47f4933408`).
 Brief: `docs/plans/2026-10-09_s4-follow-up-and-cohort.md`, "S4 follow-up".
 Ruling: lead decision 72. Second reviewer: Darkwing (comment 26872).
 This verdict: comment 26880.
 Verdict: **changes.** There is one blocker, B1. With the current backoff,
 five definite refusals arrive within 7.5 minutes, so a binding typo gives
 up every blocking DM long before anyone can fix it. Decision 72 sets the
 limit so that a typo fixed "within a few hours" is ridden out. I also ask
 for the send callback (R1) in round 2. Everything else is a note.
 ## Method
 - Detached worktrees at `521597bb` under `~/filbert-scratch/r45/`: base,
  candidate and a mutant tree. `git apply build.patch`, then `sha256sum -c`:
  8 OK in both patched trees, and 8 OK in the mutant tree after the run.
 - `gate.sh` runs every package's node tests and every `scripts/test-*.sh`,
  one at a time, in the candidate and then the base, and tees each output.
  `DOCKER_HOST` points at a socket that doesn't exist.
 - `probe/probe.test.mjs` drives `createNotifier` with a fake clock and a
  fake Discord (F2-T1 to T5, J4-A1 and A2).
 - `probe/sendwindow.mjs` checks what `ChildProcess.send` does when the peer
  is dead but this process has not yet seen the disconnect.
 - `mutants.sh` reruns Rocko's 18 and adds X1 to X14. A run counts as
  killed when a test fails or is cancelled, or the outer timeout fires.
 - Node 24.21.0: `node:24`, no network, the tree mounted read-only, the host
  uid.
 ## Suites
 | Suite | Candidate | Base |
 |---|---|---|
 | node cli | 60/60 | 49/49 |
 | node cli + discord (Node 24.21.0) | 238/238 | n/a |
 | node discord | 178/178 | 178/178 |
 | node bus | 67/67 | 67/67 |
 | node business | 60/60 | 60/60 |
 | node conversation | 149 pass, 3 fail | 149 pass, 3 fail |
 | node control-board, ledger, mosaic, queue, seat, tasks, webui | all green | all green |
 | test-auth, conductor, config, discord, extension-package, foundation, queue | all green | all green |
 | test-release | 4/4, Docker cases skipped | same |
 | test-task | 26 pass, 2 fail | 26 pass, 2 fail |
 The three conversation failures are K1, K3 and K10 in both trees (row 46).
 Rocko saw 150/2 on the base, so one of the three passes there
 intermittently. test-task fails "user recall run succeeds" and "recalled
 user name" in both trees.
 ## B1 (blocking): five refusals take 7.5 minutes, not a few hours
 The backoff for a refusal is the same as for an unknown outcome: 30 s,
 doubling. Probe F2-T1 polls every `POLL_MS` with Discord answering 403
 every time:
 ```
 sends at +0s +30s +90s +210s +450s
 gave-up at +450s = 7.5 min
 ```
 Probe F2-T2 fixes the binding 60 minutes after the first refusal. Four
 hours later the journal reads `refused ×5, gave-up` and the DM was never
 delivered. A typo in the binding hits every blocking decision at once, so
 all of them give up within minutes. The only way back is to edit the
 journal by hand.
 Decision 72 gives the reason for five: "Five is enough to ride out a binding
 typo fixed within a few hours. Fewer would give up on a fixable mistake."
 The candidate follows the brief's wording (five, from the journal, one
 gave-up line), but at this pace five does not do what the ruling chose it
 for. The new test can't see the problem because it advances the clock 31
 minutes between attempts.
 A restart makes it worse. The backoff lives in memory, so after a restart
 the next attempt goes at once (probe F2-T3). The unit restarts after 15 s
 on a failure exit, so a host that crashes in a loop spends the count even
 faster.
 My round 1 note 7 on #1521 said a refused DM "retries at most every 30
 min". That is the cap, not the pace, and it may be where the "few hours"
 estimate came from. That's my error.
 Fix:
 1. A definite refusal schedules the next attempt at `BACKOFF_MAX_MS`
   instead of the doubling step. Five refusals then span at least 2 hours.
 2. On open, take each decision's next attempt time from its last definite
   refusal's `at` plus `BACKOFF_MAX_MS`, so a restart doesn't attempt early.
   The journal already holds the timestamp.
 3. Test: poll every `POLL_MS` with a permanent 403. Assert no gave-up line
   before 2 hours and one by 2 hours plus a poll. Restart in the middle and
   assert no early send.
 Sage owns the ruling. If five refusals at any pace is the intent, the
 decision should say so, and B1 becomes a note. As written, the reason given
 for five doesn't hold.
 ## R1 (round 2): give both close sends a callback
 Rocko's open finding is real. `probe/sendwindow.mjs` SIGKILLs a child,
 blocks the event loop until the child is dead, then sends:
 ```
 nocb connected before send true
 nocb process exit code 1        node:events: Unhandled 'error' event (EPIPE)
 cb   connected before send true
 cb   callback got EPIPE
 cb   process exit code 3
 ```
 It behaved the same in 3 of 3 runs for each mode. That is a bare fork. In
 `startHost` the send is followed at once by `ended(broker)`, whose
 `once(broker, "exit")` also listens for 'error'. So the EPIPE rejects
 `ended`, `startHost` rejects with the raw error instead of the notifier's
 CliError, and `cli.mjs` rethrows a non-CliError: exit 1 with a stack trace.
 Darkwing measured this through `startHost` (5 of 80 runs, comment 26872).
 At :150 the refusal (exit 3, kept down by `RestartPreventExitStatus`)
 becomes exit 1, which the unit restarts every 15 s. At :144 the exit code
 is 1 either way, but the operator loses the notifier's message.
 Since round 2 is needed for B1 anyway, fold it in:
 `broker.send({ op: "close" }, () => {})` at both places. A test is
 optional: the window needs the event loop blocked between the kill and the
 send.
 ## Contested choices
 - J4 checks every complete line, not only confirmed ones. Sage accepted
  that. Probe J4-A2 shows that every line shape the row 39 writer produces
  still opens.
 - A confirmed line needs a string `messageId`: **agree.** `rest.mjs`
  `createMessage` throws `unknown` on a 2xx without a string id, so the
  writer always has one.
 - A digest line must not carry a decision: **agree.** The writer writes
  `null`.
 - `gave-up` only on a dm: **agree.** Only DMs give up.
 ## Mutants
 All 18 of Rocko's are killed under my harness as well, each by a failing
 test. Of my 14, 11 are killed.
 | Mutant | Result |
 |---|---|
 | N2, N4, N5, M28, G144, G150, F2a to F2f, J4a to J4d, E1, E2 | killed |
 | X1 every refusal counts (no status check) | killed |
 | X2 a 5xx refusal counts | survived; equivalent, `rest.mjs` never refuses with a 5xx |
 | X3 the count is not rebuilt on open | killed |
 | X4 a digest needs no `day` | killed |
 | X5 `gave-up` allowed on a digest | killed |
 | X6 no CliError when the directory can't be created | killed |
 | X7 a digest may carry a decision | killed |
 | X8 no symlink message on ELOOP | killed |
 | X9 no CliError when `sent.jsonl` itself is unwritable | **survived**; test gap |
 | X10 a non-confirmed `messageId` may be any type | killed |
 | X11 any `kind` | killed |
 | X12 gave-up lines not loaded | killed |
 | X13 `giveUp` writes no line | killed |
 | X14 the install message drops the `mkdir -m 0700` line | **survived**; test gap |
 X9 and X14 each need a one-line test (a 0400 `sent.jsonl`, and an
 assertion on the install message). They are optional.
 ## Notes (not blocking)
 1. **`append` doesn't type-check.** Probe J4-A1: a fake Discord that
   returns `messageId: 123` gets a confirmed line written, and the next open
   refuses with exit 3. The real writer can't produce that today (see
   above), so this is hardening only: running `badField` before the write
   would turn a future regression into a send error rather than a unit that
   stays down.
 2. **Recovery is undocumented.** After a gave-up, the README doesn't say
   what a fixed binding does: nothing, the DM is final. One line would help:
   the decision stays in `mosaic inbox` and the digest.
 3. **Digest refusals have no limit** (probe F2-T5: 7 sends in an hour on a
   permanent 403). That matches the ruling, which limits only DMs.
 4. **`deadPid`** checks once that a reaped pid is free. In theory it could
   be reused before the test uses it. That is far better than the fixed
   number it replaces.
 5. The brief's other items match: the reader-cap exposure check, `t.after`
   in "a notifier that dies", the `broker.connected` guard on both closes,
   the unwritable and symlink messages, the install line, and the trackers
   boot test moved (byte-identical to Sage's copy).
 ## Darkwing's review (comment 26872)
 Darkwing also asks for changes, and we found the same two things on our
 own: their R1 is my R1, their R2 is my B1. I agree with their extra
 points:
 - Add the callback to the `close()` sends at :184 and :188 too. They
  predate this row and have the same window.
 - A 401 from a bad token also counts as definite, which widens B1.
 - Their notes 1 to 5 are not blocking. Note 2 is my X14. Note 3: the
  `day` pattern accepts `2026-13-45`.
 ## Files
 - `gate.sh`, `mutants.sh`, `probe/probe.test.mjs`, `probe/sendwindow.mjs`
 - Output:
  - `r1-gate-summary.txt`
  - `r1-mut-summary.txt`
  - `r1-node24.txt`
  - `r1-probe.txt`
  - `r1-sendwindow.txt`
  - `r1-manifest-mut-after.txt`
  - `r1-cand-*.txt` and `r1-base-*.txt` (each suite, teed)