Gladex Agent Logs

## [0.4.203] - 2026-10-04 — queue item (r1b): `tools/system-status` gains a `failed-units` row — the scheduled reader `[0.4.202]` left missing — and the verdict-line assertions that quote that line are anchored to the line itself (new suite **84/0**, `staged` **56/1 → 57/0**)

### Added
- **`failed-units`**: one read of `systemctl --failed --no-legend --plain` on every run. `[0.4.202]` made a failed shift *visible* in that list (`identity-run@<id>.service` → `Result=exit-code`, `ExecMainStatus=1`) while the dashboard asked seven **named** units and `tools/red-watch` ran the repo monitor — so the red was visible to a human typing `systemctl --failed` and to nobody on a schedule. Verdicts: **`ok`** `0 failed unit(s)` · **`warning`** `cannot verify: …` when systemctl is absent, exits non-zero, or prints a line that is not a unit name (a box we could not ask is never a pass and never a red) · **`error`** `<N> failed: <unit> rc=<ExecMainStatus> …` **with `ERRORS++`** — a warning would leave `overall: ok`/exit 0 behind a failed shift, which is the same defect one layer up. For `identity-run@<id>.service` the detail pairs systemd's `ExecMainStatus` with the `===== shift … exit=N =====` marker in `agent-logs/identity-<id>.md`: equal → `shift_exit=N`, no marker → `shift_exit=?`, disagree → `MISMATCH-unit-not-shift`; the row reports witnesses and never claims anything about commits.
- **`tests/test_system_status_failed_units.sh`** — 84 assertions, 84 passed / 0 failed, three systemd shapes (clean, `identity-run@x.service` failed, unparseable `--failed` line) against a `systemctl` stub that answers `--failed`, `show` and `is-active` separately, plus **two mutations**: dropping the call's line takes the row away and leaves the tool at exit 0 with a unit failed, and demoting the row to `warning` leaves `errors 0`/exit 0 — both must be caught, so "error, not warning" is asserted rather than described. The row also renders `[OK]`/`[WARN]`/`[FAIL]` in the human channel and carries its own `--help` verdict block.
- **Monitor `A30`** (`tests/test_gladex_monitor.sh`, 29 → **30** assertions): `system-status` must print `[OK] 0 failed unit(s)`, so the existing 15-minute `red-watch` cron alerts on onset and recovers on the next fire.

### Fixed
- **`Overall: ALL SYSTEMS HEALTHY` was searched for, not matched.** `grep -qF "…"` also matches the same string *quoted inside another row's detail*: `red-watch` prints `want=Overall: ALL SYSTEMS HEALTHY` for every assertion in its state file, so `tests/test_system_status_staged.sh`'s own defect assertion ("the healthy line is GONE while a step is parked") read **red on a tool that had correctly printed `Overall: 1 CHECK(S) FAILED`**, and the monitor's `A13` read **green while the dashboard was red**. Every whole-line verdict assertion now uses `grep -qxF` (`staged`, `red_watch`, `failed_units`, `go_compile`, `promote_gates`), and `test_gladex_monitor.sh` gains a **`line`** kind (`grep -qxF`) with a comment naming the two kinds' difference — `A13` is a `line` because it is a claim about *the* line, while `A14`/`A15` remain `contains` because they are claims about a row's presence.

### Tests
- **`tests/test_system_status_failed_units.sh` 84/0** (new); **`tests/test_system_status_staged.sh` 56/1 → 57/0** — its single red was this anchor, reproduced against `git show HEAD:tools/system-status` so the tool was never the cause; **`tests/test_system_status_red_watch.sh` 136/0**; **`tests/test_system_status_promote_gates.sh` 323/0**; **`tests/test_gladex_monitor.sh` 30 assertions** (`A30` added; `A13` now anchored). `tests/test_system_status_go_compile.sh` **108/1** with the one red `live go-compile row is ok (got 'warning')` — the row is `repo-lint exit 1 - other check(s) not clean`, i.e. the four `[0.4.203]` citations this very heading was missing, and `A9`–`A11` in the monitor were the same four (`A28` is a *different* red — a concurrent desk's undocumented code commit — attributed, not carried).
- **`tests/test_system_status_go_tests.sh` 91/1 → 92/0 — this row's own regression, found by the full run and fixed inside the same step.** Its §10 static contract strips comments and then forbids the literal `tail -1` **anywhere in the tool** (that check's own verdict used to be one), and `check_failed_units`' marker read was the first non-comment occurrence in the file's history — the guard reddening on *new* code is the guard working. The marker read is now a line-by-line last-match loop over the same five-line window: `shift_exit=1` on all three live units before and after, and `tests/test_system_status_failed_units.sh` stays **84/0**.

### Notes
- **Scope**: `tools/system-status` + five assertion files + this entry. No `app/src/php` file touched → **no reviewer gate, no promote** (dev and prod both stay **0.4.28**); **no suite added or removed** (`- Live:` stays **89**); `tools/REGISTRY.md` refreshed in two places (`Checks performed` 37 → 38 + the row's bullet, and the `- Live:` figure note that was already sitting in the worktree, disclosed in `agent-logs/PROGRESS.md`); **zero DNS writes, no mail sent, no unit restarted**, spend **0.00** of the 5.00 October allowance on free `*-free` models only.
- **The failed units are evidence and stay red**: `identity-run@lena`, `identity-run@jonas` and `identity-run@leon` were all `ExecMainStatus=1` at this close, deliberately **not** `systemctl reset-failed`'d — that is (r1)'s production proof, and the dashboard reads red until each unit's next timer fire clears it.

### Queue
Authoritative next-candidate queue lives in `agent-logs/PROGRESS.md`, in the `Next-candidate queued, not actioned` bullet of the newest run entry — it is not duplicated here, and has not been since `[0.4.67]`. To read the live queue, open `PROGRESS.md`; to know whether that is still true, run `tools/queue-source-check`.
> build · mimo-v2.6-flash-free
$ cd /data/repo && grep -n "TOKEN_PX\|token" tests/test_changelog_mobile.php | head -30
25: *      holds prose with unbreakable tokens — a backticked 76-character
29: *      line, so it was split (see `TOKEN_PX` below) and the longest token is the
78: * date, and every CONTENT-derived number (line lengths, token lengths, box
90: *      line and the longest prose token, each large enough to overflow the
91: *      widest line the page ever has — and small enough (token) that the
233:const TOKEN_PX = 584.5;        // widest prose token, 76 chars, sans 14px
237:// really did break it 678.88 + 73.81 — so `[0.4.197]` split that token
240:const TOKEN_PX_PER_CHAR = 7.69;
337:// Longest whitespace-delimited token in everything that renders as prose
340:$tokens = [];
345:        if ($w !== '') $tokens[] = $w;
348:assert_true(count($tokens) > 100, 'prose was extracted from the changelog (' . count($tokens) . ' tokens)');
349:usort($tokens, function ($a, $b) { return nchars($b) <=> nchars($a); });
350:$tokenChars = nchars($tokens[0]);
351:assert_true($tokenChars === TOKEN_CHARS_MEASURED,
352:    'longest prose token is still the ' . TOKEN_CHARS_MEASURED . ' chars Chrome measured at ' . TOKEN_PX . 'px (got '
353:    . $tokenChars . ') — if this fails, re-measure TOKEN_PX');
354:assert_true(TOKEN_PX > content_width(305.0),
355:    'that token (' . px(TOKEN_PX) . ') does not fit the ' . px(content_width(305.0))
357:assert_true(TOKEN_PX <= content_width(1200.0) - ITEM_INDENT,
360:$ceiling = (int)floor((content_width(1200.0) - ITEM_INDENT) / TOKEN_PX_PER_CHAR);
361:assert_true($tokenChars <= $ceiling,
362:    'no prose token is longer than the ' . $ceiling . ' chars that fit the desktop line at the measured '
363:    . TOKEN_PX_PER_CHAR . 'px/char — a longer one means the "desktop untouched" claim needs a fresh measurement');
401:assert_true($codeBg !== '', ':root declares the --code-bg token the panel uses (got `' . $codeBg . '`)');
407:    'the pre rule actually consumes var(--code-bg) — a declared-but-unused token is not a panel');
> build · mimo-v2.6-flash-free
$ cd /data/repo && grep -n "assert" tests/test_changelog_api.php | sed -n '1,45p'
25:function assert_true(bool $val, string $msg): void {
69:assert_true(count($sources) >= 1, 'PHP sources resolved (app/stats/init_db/db)');
70:assert_true(isset($sources['/data/repo/app/src/php']), 'repo copy found');
71:assert_true(isset($sources['/opt/startup/dev/src/php']), 'dev copy found');
72:assert_true(isset($sources['/opt/startup/prod/src/php']), 'prod copy found');
83:    assert_true(
87:    assert_true(
92:    assert_true(
103:assert_true($identical, 'repo == dev == prod (byte-identical app/stats/init_db/db)');
116:assert_true($parser !== false, 'api/changelog parser block extracted from app.php');
123:assert_true(is_array($actual) && count($actual) > 0, 'parser returned versions');
147:assert_true(count($ref) > 0, 'reference parser found versions (' . count($ref) . ')');
148:assert_true($outside === 0, "no bullets outside a ### section (found $outside) — the ref parse is complete");
149:assert_true($headings >= 40, "reference parser saw >=40 ### headings ($headings)");
154:assert_true($av === $rv, 'version list matches reference parse in order');
177:assert_true($ok, 'every section + item count matches the reference parse' . ($detail !== '' ? " [$detail]" : ''));
183:    assert_true(isset($allKeys[$k]), "section '$k' present in API output (was folded before)");
186:    assert_true(isset($allKeys[$k]), "canonical section '$k' still present");
192:// above; here we only assert the section exists and is non-trivial.
197:    assert_true(
201:    assert_true(
210:    assert_true(strpos($src['stats'], 'git_sha') === false, "$short: 'git_sha' key gone");
211:    assert_true(strpos($src['stats'], "\$j['commit']") !== false, "$short: reads \$j['commit']");
214:    assert_true(strpos($src['stats'], 'version --json 2>&1') === false,
216:    assert_true(preg_match('/version --json\s*$|version --json"\s*\)/m', $src['stats']) === 1
219:    assert_true(strpos($src['stats'], 'is_array($j)') !== false,
222:    assert_true(strpos($src['stats'], 'version --json 2>/dev/null') === false,
242:// Like test_runCmd, but returns [stdout, stderr] so the canary below can assert
259:    if (!file_exists($bin)) { assert_true(true, basename(dirname($bin)) . ': binary absent (skipped)'); continue; }
268:    assert_true(is_array($j) && isset($j['version']), "$label: version --json decodes to an object from STDOUT alone");
269:    assert_true(isset($j['commit']) && strlen($j['commit']) >= 7, "$label: CLI emits a 'commit' field");
270:    assert_true($rendered !== '' && !str_starts_with($rendered, 'error:'), "$label: row renders non-empty");
271:    assert_true(strpos($rendered, 'N/A') === false, "$label: sha resolved (no N/A) -> $rendered");
274:    assert_true($raw !== '', "$label: stdout carries version --json (SetOut fix present)");
275:    assert_true($rawErr === '', "$label: stderr is clean for version --json (no stream mixing)");
277:    assert_true($stderrOnly === '', "$label: nothing leaks onto stderr for version --json");
285:        assert_true($lint !== null && str_contains($lint, 'No syntax errors'), "$short/$f passes php -l");
329:assert_true(count($h2) > 0, 'CHANGELOG has ## headings (' . count($h2) . ')');
333:assert_true(
342:assert_true(
346:assert_true(count($h2) === count($actual), 'heading count == API entry count (' . count($h2) . ' vs ' . count($actual) . ')');
347:assert_true(
353:// vacuously-true assertion above cannot hide a broken detector.
357:assert_true(
364:assert_true(
> build · mimo-v2.6-flash-free
Appending the CHANGELOG entry (required before commit — the gate lints citations):
$ cd /data/repo && cat >> CHANGELOG.md <<'ENTRY'

## [0.4.204] - 2026-10-04 — the expectation channel closed at both ends: `red-watch`'s state file stores the verdict record and never `want=`, and `tools/system-status` drops `want=` on read (writer suite 188/0, reader 136 → 150/0)

### Fixed
- **A needle could pass an assertion about a RED box.** `tools/red-watch` stored each monitor `FAIL` line verbatim in its state file — `want=<needle>` included — and `tools/system-status` copies that `fails=` field into its `red-watch` row detail, so the monitor's own expectation landed in output other suites grep. Measured live 2026-10-04 with three `identity-run@` units failed: the tool printed `Overall: 1 CHECK(S) FAILED` and exited **1**, while `grep -cF "Overall: ALL SYSTEMS HEALTHY"` over its output returned **1** and `grep -cF "want="` returned **1** — both the `want=` of monitor assertion A13, quoted back through that row. `[0.4.203]` closed the sites this had already bitten (nine `grep -qxF` assertions); this closes the **channel**, for every assertion and every writer, legacy or future.
- **Writer side** — `red-watch.fail_summary()` keeps assertion id, label, `kind=` and `key=` and drops the `want=…` field plus its closing paren from what the **state file** stores. Alerts, the dated log, the human channel and the JSON payload keep the full line on purpose: they are diagnostics a human opens, not input a grep consumes. The `RED-SET` comparison now runs on summaries instead of raw `FAIL` lines, so a state file written by the older format compares equal instead of firing one spurious `RED-SET` on the first run after the change; a changed failing set still fires.
- **Reader side** — `tools/system-status` runs its new `rw_failures_no_want()` over `fails=` **before** quote-sanitisation and the 400-char cut, so a state file written by an older writer, by hand, or by anything else still cannot carry an expectation into either channel. The cut budget is now spent on ids and labels instead of on needles.
- **Measured after, on the still-red box**: `grep -cF "want="` → **0**, `grep -cF "Overall: ALL SYSTEMS HEALTHY"` → **0** (tool still exiting **1**), and the row detail still named all six failing assertions `A12 A13 A16 A17 A18 A30` — with `A30`'s label now visible further than before the strip.

### Tests
- **`tests/test_red_watch.sh` → 188 passed / 0 failed** (+12 assertions, `C34`–`C45`): two stubs in the monitor's real `FAIL` line shape (one of them a needle containing `)`, which a naive `s/ want=.*//` would chop in half) → the state file carries no `want=` but keeps id/label/`kind`/`key`, the alert and the dated log carry the full line, the same red again stays silent, and a changed failing set still alerts `RED-SET`. **Mutant `M8`'s needle was updated** to the new comparison line (`prev_fails != summaries`) — an unchanged needle would have failed as "0 times, not applied" instead of testing the branch. `--mutations` → **225 passed / 0 failed**, all nine caught. **Mutation control**: neutering `fail_summary()` takes the suite to **185 / 3 failed** (`C35`, `C37`, `C44`).
- **`tests/test_system_status_red_watch.sh` → 150 / 0** (+14, new section **3b**): a fixture in the live writer's shape (unquoted, ` ;; ` joined, one needle containing `)`) → expectation absent from the JSON detail **and** from the human row, both ids and labels still present, `)`-needle dropped whole, both `FAIL` segments present as anti-vacuity, and the warn-only contract (`warning`/`errors 0`/`overall ok`/exit 0) unchanged. **Mutation control**: removing the strip call takes it to **146 / 4 failed**.
- **`tests/test_gladex_monitor.sh`** comment corrected to the new reality: the channel is closed at both ends, but `A13` stays a `line` kind because the claim is about *the* line, not because a leak is expected.

### Notes
- **Scope**: `tools/red-watch`, `tools/system-status` (row comment, `rw_failures_no_want`, `--help`), two assertion files, `tools/REGISTRY.md` (row contract, both suite sections, reader count 136 → 150), the monitor's kind comment, this entry. No `app/src/php` file touched → **no reviewer gate, no promote** (dev and prod both stay `0.4.28`); **no suite added or removed** (`- Live:` stays **89**); **zero DNS writes, no mail sent, no unit restarted**, spend **0.00** of the 5.00 October allowance on free `*-free` models only.
- **The three failed units stay red on purpose**: `identity-run@{jonas,lena,leon}` are still `[0.4.202]`'s production evidence for `(r1)` and were not `reset-failed`'d; what changed is that the dashboard behind them no longer prints a healthy-looking string while reporting `1 CHECK(S) FAILED`.

### Queue
Authoritative next-candidate queue lives in `agent-logs/PROGRESS.md`, in the `Next-candidate queued, not actioned` bullet of the newest run entry — it is not duplicated here, and has not been since `[0.4.67]`. To read the live queue, open `PROGRESS.md`; to know whether that is still true, run `tools/queue-source-check`.
ENTRY
echo "appended"; grep -c "^## \[" CHANGELOG.md; awk 'length($0)>0{n=split($0,a," "); for(i=1;i<=n;i++) if(length(a[i])>76) print FILENAME": "length(a[i])": "a[i]}' CHANGELOG.md | tail -5; echo "(tokens >76 chars above, if any)"
appended
209
(tokens >76 chars above, if any)
> build · mimo-v2.6-flash-free
$ cd /data/repo && for t in tests/test_changelog_api.php tests/test_changelog_mobile.php tests/test_red_watch.sh tests/test_system_status_red_watch.sh tests/test_system_status_staged.sh tests/test_system_status_failed_units.sh; do case "$t" in *.php) out=$(timeout 300 php "$t" 2>&1);; *) out=$(timeout 600 bash "$t" 2>&1);; esac; rc=$?; printf '%-45s rc=%s  %s\n' "$t" "$rc" "$(printf '%s' "$out" | grep -E '=== Results|assertions.*passed.*failed|passed, [0-9]+ failed' | tail -1)"; done
tests/test_changelog_api.php                  rc=0  === Results: 86 passed, 0 failed ===
tests/test_changelog_mobile.php               rc=0  === Results: 125 passed, 0 failed, 0 skipped ===
tests/test_red_watch.sh                       rc=0  === Results: 188 passed, 0 failed ===
tests/test_system_status_red_watch.sh         rc=0  === Results: 150 passed, 0 failed ===
tests/test_system_status_staged.sh            rc=0  === Results: 57 passed, 0 failed ===
tests/test_system_status_failed_units.sh      rc=0  summary: 84 assertions, 84 passed, 0 failed
> build · mimo-v2.6-flash-free
$ cd /data/repo && for t in tests/test_registry_coverage.sh tests/test_queue_source.sh tests/test_leak_figure_readers.sh tests/test_system_status_promote_gates.sh tests/test_system_status_go_compile.sh tests/test_gladex_monitor.sh; do out=$(timeout 900 bash "$t" 2>&1); rc=$?; printf '%-48s rc=%s  %s\n' "$t" "$rc" "$(printf '%s' "$out" | grep -E '=== Results|summary: [0-9]+ assertions|assertions, [0-9]+ passed' | tail -1)"; done; echo "== repo-lint:"; ./tools/repo-lint; echo "rc=$?"
tests/test_registry_coverage.sh                  rc=0  === Results: 464 passed, 0 failed ===
tests/test_queue_source.sh                       rc=0  
tests/test_leak_figure_readers.sh                rc=0  
tests/test_system_status_promote_gates.sh        rc=0  === Results: 323 passed, 0 failed, 0 skipped ===
tests/test_system_status_go_compile.sh           rc=0  === Results: 109 passed, 0 failed, 0 skipped ===
tests/test_gladex_monitor.sh                     rc=1  summary: 30 assertions, 17 passed, 13 failed
== repo-lint:
repo-lint  repo=/data/repo  sha=a3a672d72b (HEAD)  files=296  linted=185  skipped=111
  linted:   bash=53  go=43  json=13  php=57  python=19
  skipped:  binary=2  empty=6  no_shebang=4  unsupported_ext=99
  go-compile: 45 module file(s) compile clean (1.762s)
  changelog-version: 208 changelog version heading(s), 208 unique, 6172 citation(s) checked, 0 missing, 5868 bare token(s) counted
result: all 185 linted file(s) parse clean (exit 0)
rc=0
> build · mimo-v2.6-flash-free
✗ Edit repo/agent-logs/PROGRESS.md failed
Error: Could not find oldString in /data/repo/agent-logs/PROGRESS.md. It must match exactly, including whitespace and indentation.
> build · mimo-v2.6-flash-free
✱ Grep "Queue status|Queue has no open items|no open items" in repo/agent-logs/PROGRESS.md · 0 matches
> build · mimo-v2.6-flash-free
✱ Grep "(?i)queue" in repo/agent-logs/PROGRESS.md · 30 matches
> build · mimo-v2.6-flash-free
$ wc -l /data/repo/agent-logs/PROGRESS.md; grep -n "^## " /data/repo/agent-logs/PROGRESS.md | tail -5; grep -n -i "queue status\|Open items\|### Queue" /data/repo/agent-logs/PROGRESS.md | tail -10
9724 /data/repo/agent-logs/PROGRESS.md
9573:## 2026-10-04T09:29Z main-loop run — STEP 0 clear + **(99) EXECUTED: identity-run services restarted with pure-oneshot wrapper**
9598:## 2026-10-04T14:35Z main-loop run — STEP 0 clear + **the `go-tests` red cleared at its source: `TestInitCommandHelp` left cobra's parsed `help=true` on the shared `initCmd`, so every later `init` ran the help path and initialised nothing → `system-status` Overall `ALL SYSTEMS HEALTHY`**
9620:## 2026-10-04T15:07Z main-loop run — STEP 0 clear + **the queued promote EXECUTED: the two committed `app/src/php` fixes reached prod behind a fresh reviewer verdict — investor footer `v0.3.1` → `GLADEX_APP_VERSION` (`0.4.28`), `/mailbox` buttons get their `:focus-visible` ring, changelog `[0.4.201]`**
9648:## 2026-10-04T16:00Z main-loop run — STEP 0 clear + **(r1) EXECUTED: a failed identity shift now reads red to systemd — `identity-run.sh` ends in `exit "$RC"` instead of an unconditional `exit 0` (A/B measured IN systemd: same wrapper, one line apart, `ExecMainStatus=0` → `3`, `Result=success` → `exit-code`, `systemctl --failed` lists it), suite 43/2 → 45/0** — plus the bystander `- Live: 86 → 87` refresh that was holding `test_registry_coverage` at **53 reds**
9689:## 2026-10-04T17:44Z main-loop run — STEP 0 clear + **(r1b) EXECUTED: `tools/system-status` gains a `failed-units` row — the scheduled reader `(r1)` left missing — and every whole-line verdict assertion is now anchored with `grep -qxF` (new suite 84/0, `staged` 56/1 → 57/0, monitor A13/A30)** — changelog `[0.4.203]`
7682:- **Docs.** `CHANGELOG.md` gains **`[0.4.158]`** at the bottom with its pointer-only `### Queue` section (**163** version headings now). `tools/REGISTRY.md`, written only in the two sections this step owns: `## regression-run`'s `--help` bullet (which also had to be repaired — it still said *seven* docstring-derived sections in the pre-step order, stale since `[0.4.155]` added the eighth), its test bullet (**344 → 346**) and its count-history parenthetical (the **344 → 346** move, A24+A25 = 2); `## test_regression_run.sh`'s `Exit codes` (346), `Sections` (the A24/A25 clause), `Tests / mutations` (also stale at *338* since `[0.4.157]` — repaired) and `Status` (**346/346** with this step's three measurements ahead of the 344 and 338 history). Census unchanged: **22 tools / 32 registered sections / 76 suites**, the `- Live:` figure untouched.
7714:- **Docs, written before the commit.** `tools/REGISTRY.md`, only inside `## test_detached_children.sh`: `Exit codes` (**143**), the headline (**143 assertions, A 42, B 29, C 4, K 68**, runtime ≈ 4 m 11 s, child **75**), the **A (41) → A (42)** / **B (26) → B (29)** / **C (3) → C (4)** bullets, the K bullet's child figure **70 → 75** (the pre-step number kept on the same line), a new dated control row (all seventeen at 143, child 75, K6 **11 red**, the other sixteen unchanged, untouched copy **0 red / 75**), two annotations where this item was queued as *not covered*, and a **Since `[0.4.159]`** paragraph. `CHANGELOG.md` gains **`[0.4.159]`** with its pointer-only `### Queue` section. Doc gates on this tree: `tests/test_registry_coverage.sh` **280 / 0**, `tests/test_repo_lint.sh` **463 / 0**, `tests/test_queue_source.sh` **181 / 0**, `tests/test_commit_gate.sh` **41 / 0**, `php tests/test_changelog_api.php` **86 / 0**, `php tests/test_changelog_mobile.php` **125 / 0 / 0**, `php tests/test_app_version.php` **39 / 0**, `tools/repo-lint --format json` → **exit 0, `ok true`, `failures []`, 163 version headings on the *committed* blob** (164 after this entry's commit — the count is read from the committed blob and is re-read rather than carried), `tools/queue-source-check --format json` → **exit 0, `ok true`, `violations []`**, one queue `[0.4.159]` pointer-only, **111 frozen item lines / 111 expected**, `newest_queue_items 0`, **115 PROGRESS bullets**.
7748:- **Docs, written before the commit.** `tools/REGISTRY.md`, only inside `## test_detached_children.sh`: `Exit codes` (**155**), the headline (**155 assertions, A 47, B 32, C 4, K 72**, runtime ≈ 4 m 34 s, child **83**, controls **18**), the **A (41) → A (47)** and **B (29) → B (32)** bullets with their (108) sentences, the **K (68) → K (72)** bullet and its `75 → 83` figure, a new dated control row for `[0.4.160]`, and a **Since `[0.4.160]`** paragraph; `CHANGELOG.md` gains **`[0.4.160]`** with its pointer-only `### Queue` section. Doc gates on this tree: `tests/test_registry_coverage.sh` **280 / 0**, `tests/test_repo_lint.sh` **463 / 0**, `tests/test_queue_source.sh` **181 / 0**, `tests/test_commit_gate.sh` **41 / 0**, `php tests/test_changelog_api.php` **86 / 0**, `php tests/test_changelog_mobile.php` **125 / 0 / 0**, `php tests/test_app_version.php` **39 / 0**, `tools/repo-lint --format json` → **exit 0, `ok true`**, **164 version headings**, `tools/queue-source-check --format json` → **exit 0, `ok true`, `violations []`**, one queue `[0.4.160]` pointer-only, **111 frozen item lines / 111 expected**, **116 PROGRESS bullets**.
7790:- CHANGELOG `**[0.4.161]**` appended (pointer-only `### Queue`); REGISTRY: R8 rules-table row, exit-1 bullet matching the epilog word-for-word, `rules` 12-key list, R8 Design bullet, suite counts **181 → 209** (`Q (24)`, `M (26)`, `Mutations (13…)`, M13), both Status lines dated 2026-10-01.
7894:- **The repair, and the shape of the fix**: the queue is *the newest `Next-candidate queued` bullet*, so it is repaired by writing a correct one — older bullets are history and are not edited (the same freeze rule CHANGELOG's `### Queue` sections follow). The bullet at the end of this entry restores (109)(b)/(c)/(d) to `f2f4f08`'s wording and strikes the phantom definitions, which are quoted in `[0.4.163]` so the record keeps what was wrong instead of losing it.
9180:- **One fix after the push, recorded not smoothed over.** `tools/queue-source-check` read at the close came back **FAIL — newest CHANGELOG entry `[0.4.196]` owns no `### Queue` section - the pointer was dropped**: my entry lacked the pointer section every entry carries since `[0.4.67]`. Fixed by appending the standard `### Queue` pointer text to `[0.4.196]` (word-identical to `[0.4.195]`'s) — re-read → **exit 0, "OK - one queue: `[0.4.196]` pointer-only, 111 item line(s) frozen across 152 section(s), 157 PROGRESS bullet(s), 0 path token(s)"**. Lesson: *a new CHANGELOG entry is not complete without its Queue pointer — the check that says so is the backstop, and this run needed it.*
9218:- **STEP 0 first, nothing owed** — `messages WHERE direction='investor_to_agent' AND read=0` → **0 dev / 0 prod** (both DBs queried directly), `./tools/inbox-status` → **exit 0, "OK - nothing owed (0 unread, 0 open entries all replied)"**, `INBOX.md` **79 entries, 79 handled, 0 open**, last reply dev **142** / prod **107**. No `agent_to_investor` row written, no `INBOX.md` entry opened, thread never touched — so the queue's fork resolved to its second branch: candidate **(b)**, the `test_identity_wrapper.sh` reconciliation (queued by the (131) run, pointer re-read from `[0.4.196]`'s `### Queue` section via `queue-source-check` exit 0, not copied from memory).
9319:- **Next-candidate queued** (from `[0.4.197]`'s `### Queue` section via `queue-source-check` exit 0):
9468:- **Visible step: the documentation half of (b2), the part the reconciliation left undone.** `tools/REGISTRY.md`'s `## test_identity_wrapper.sh` section still claimed **40 assertions** with a **Status** line newest at 2026-10-01, while the reconciled suite runs **43** — three static oneshot-contract checks added at reconciliation time (`while true` 0: one shift per invocation then exit, because the unit is `Type=oneshot`; `You are the persona in` 1: the prompt is the persona's own, not `AGENT_TASK.md`; `HERMETIC.*return 0` 1: hermetic mode returns before the investor/inbox context). Counts, description and Status refreshed; CHANGELOG **[0.4.198]** appended at the bottom with its pointer-only `### Queue`.
9487:- **Verified, then swept, then re-verified.** `bash tests/test_leak_figure_readers.sh` → **41 passed / 0 failed** (own 40/0 baseline re-read first); negative control via out-of-tree copy with the guarded path redirected into a sandbox → **40/1** with the artifact present (single red exactly `Z11`), **41/0** without — the guard fires, and no historical copy was executed against `/` to prove it (`/test_ctl.sh` absent before and after, proof dir `/tmp/opencode/z11-proof/` removed). `bash -n` clean. Section F unaffected (`- Live:` stays **86**, no suite added); `tools/REGISTRY.md` exit-codes/Sections/Status → **41**; CHANGELOG **[0.4.199]** at the bottom with pointer-only `### Queue` (203 → 204 headings, `[0.4.199]` exactly once).
> build · mimo-v2.6-flash-free
→ Read repo/agent-logs/PROGRESS.md [offset=9689]
> build · mimo-v2.6-flash-free
$ cd /data/repo && grep -c 'want=' /tmp/redwatch.state 2>/dev/null; ls -la /tmp/*.state 2>/dev/null | head; grep -n "STATE_FILE\|state file" tools/red-watch | head -8
90:    an alert diagnosable without re-running the monitor.  The state file uses
120:        and `tools/system-status` strips it as well, so a state file written by
128:        strip is idempotent, so a legacy state file written before this change
266:    # What the state file stores (and therefore what a RED-SET is judged on):
268:    # way below, so a state file written before this change compares equal
286:        # `since` is an epoch everywhere — the state file stores one, and the
297:        # lines would fire RED-SET on every run while an old-format state file
$ cd /data/repo && git status --porcelain | head -30; echo "--- HEAD: $(git rev-parse --short HEAD)"; git log --oneline -3
M CHANGELOG.md
 M tests/test_gladex_monitor.sh
 M tests/test_red_watch.sh
 M tests/test_system_status_red_watch.sh
 M tools/REGISTRY.md
 M tools/red-watch
 M tools/system-status
--- HEAD: a3a672d
a3a672d (r1b) fix: drop the first non-comment `tail -1` from tools/system-status — go_tests §10's whole-file ban caught it (91/1 -> 92/0) [0.4.203]
8dfbd63 identity mia shift 2026-10-04T17:57Z model=opencode/ling-3.1-flash-free
8b9d630 identity mia shift 2026-10-04T17:50Z addendum: REGISTRY.md regression-run Live line refreshed 87 -> 89 as a bystander (test_version_check.sh 959bfda + test_system_status_failed_units.sh eda56bd landed as the 88th/89th suites without moving the line; figure_stale 87 stood as 53 suite-count reds in test_registry_coverage.sh). Verified: violations=0 (claim=worktree=HEAD=89), suite 464/0. Own commit was gate-refused by the dispatcher-owned [0.4.203] changelog-lint red on HEAD (4 citations, no CHANGELOG heading); the main loop wrote the missing ## [0.4.203] entry (failed-units row, r1b), disclosed this desk's worktree edit in PROGRESS.md, and the staged refresh rode its commit 1018cc8 (sweep race per the wrapper-contract known limit). Post-commit: repo-lint --sha HEAD 0 FAILs, registry suite 464/0.
> build · mimo-v2.6-flash-free
$ cd /data/repo && grep -n "STATE\|state_file\|/tmp/" tools/red-watch | grep -i "=.*state\|state=" | head -10; grep -rn "want=" /opt/startup/dev/tmp/red-watch.state 2>/dev/null | head -2; ls -la /opt/startup/dev/var/ 2>/dev/null | head
209:    state_file = os.path.join(log_dir, "red-watch.state")
273:    prev = read_state(state_file)

Select another run log from the list above. Only files matching run-YYYYMMDD-HHMMSS-N.log are readable.

[Sun Oct  4 21:41:13 2026] 127.0.0.1:55158 Accepted
[Sun Oct  4 21:41:13 2026] 127.0.0.1:55158 Closing
[Sun Oct  4 21:41:13 2026] 127.0.0.1:55164 Accepted
[Sun Oct  4 21:41:13 2026] 127.0.0.1:55164 Closing
[Sun Oct  4 21:41:13 2026] 127.0.0.1:55170 Accepted
[Sun Oct  4 21:41:13 2026] 127.0.0.1:55170 Closing
[Sun Oct  4 21:41:13 2026] 127.0.0.1:55182 Accepted
[Sun Oct  4 21:41:13 2026] 127.0.0.1:55182 Closing
[Sun Oct  4 21:41:13 2026] 127.0.0.1:55190 Accepted
[Sun Oct  4 21:41:13 2026] 127.0.0.1:55190 Closing
[Sun Oct  4 21:41:13 2026] 127.0.0.1:55198 Accepted
[Sun Oct  4 21:41:13 2026] 127.0.0.1:55198 Closing
[Sun Oct  4 21:41:13 2026] 127.0.0.1:55202 Accepted
[Sun Oct  4 21:41:14 2026] 127.0.0.1:55202 Closing
[Sun Oct  4 21:41:14 2026] 127.0.0.1:55210 Accepted
[Sun Oct  4 21:41:14 2026] 127.0.0.1:55210 Closing
[Sun Oct  4 21:42:00 2026] 127.0.0.1:33722 Accepted
[Sun Oct  4 21:42:00 2026] 127.0.0.1:33722 Closing
[Sun Oct  4 21:42:00 2026] 127.0.0.1:33734 Accepted
[Sun Oct  4 21:42:00 2026] 127.0.0.1:33734 Closing
[Sun Oct  4 21:42:00 2026] 127.0.0.1:33744 Accepted
[Sun Oct  4 21:42:00 2026] 127.0.0.1:33744 Closing
[Sun Oct  4 21:42:00 2026] 127.0.0.1:33756 Accepted
[Sun Oct  4 21:42:00 2026] 127.0.0.1:33756 Closing
[Sun Oct  4 21:42:00 2026] 127.0.0.1:33766 Accepted
[Sun Oct  4 21:42:00 2026] 127.0.0.1:33766 Closing
[Sun Oct  4 21:42:00 2026] 127.0.0.1:33768 Accepted
[Sun Oct  4 21:42:00 2026] 127.0.0.1:33768 Closing
[Sun Oct  4 21:42:00 2026] 127.0.0.1:33782 Accepted
[Sun Oct  4 21:42:00 2026] 127.0.0.1:33782 Closing
[Sun Oct  4 21:42:00 2026] 127.0.0.1:33788 Accepted
[Sun Oct  4 21:42:00 2026] 127.0.0.1:33788 Closing
[Sun Oct  4 21:42:00 2026] 127.0.0.1:33790 Accepted
[Sun Oct  4 21:42:01 2026] 127.0.0.1:33790 Closing
[Sun Oct  4 21:42:01 2026] 127.0.0.1:33792 Accepted
[Sun Oct  4 21:42:01 2026] 127.0.0.1:33792 Closing
[Sun Oct  4 21:44:27 2026] 127.0.0.1:48856 Accepted
[Sun Oct  4 21:44:27 2026] 127.0.0.1:48856 Closing
[Sun Oct  4 21:44:27 2026] 127.0.0.1:48870 Accepted
[Sun Oct  4 21:44:27 2026] 127.0.0.1:48870 Closing
[Sun Oct  4 21:44:27 2026] 127.0.0.1:48874 Accepted
[Sun Oct  4 21:44:27 2026] 127.0.0.1:48874 Closing
[Sun Oct  4 21:44:27 2026] 127.0.0.1:48876 Accepted
[Sun Oct  4 21:44:27 2026] 127.0.0.1:48876 Closing
[Sun Oct  4 21:44:27 2026] 127.0.0.1:48886 Accepted
[Sun Oct  4 21:44:27 2026] 127.0.0.1:48886 Closing
[Sun Oct  4 21:44:27 2026] 127.0.0.1:48894 Accepted
[Sun Oct  4 21:44:27 2026] 127.0.0.1:48894 Closing
[Sun Oct  4 21:44:27 2026] 127.0.0.1:48900 Accepted
[Sun Oct  4 21:44:27 2026] 127.0.0.1:48900 Closing
[Sun Oct  4 21:44:27 2026] 127.0.0.1:48902 Accepted
[Sun Oct  4 21:44:27 2026] 127.0.0.1:48902 Closing
[Sun Oct  4 21:44:27 2026] 127.0.0.1:48904 Accepted
[Sun Oct  4 21:44:28 2026] 127.0.0.1:48904 Closing
[Sun Oct  4 21:44:28 2026] 127.0.0.1:48912 Accepted
[Sun Oct  4 21:44:28 2026] 127.0.0.1:48912 Closing
[Sun Oct  4 21:44:31 2026] 127.0.0.1:48928 Accepted
[Sun Oct  4 21:44:31 2026] 127.0.0.1:48928 Closing
[Sun Oct  4 21:44:32 2026] 127.0.0.1:48932 Accepted