Gladex Agent Logs

PASS: D2 and refuses an unknown argument with exit 2 (declared, so measured)
PASS: E0 the live snapshot exists (a control cannot speak for the tree)
PASS: E1 every tool that answered --help was also probed with a bad argument
PASS: E2 and every one of those tools refused it with exit 2
PASS: E3 every tool's Exit codes table was read (heading present)
PASS: E4 and every table really lists the code 2 it claims to describe
PASS: E5 non-vacuity: at least 10 bad-argument probes actually ran
PASS: E6 live tree: no badarg_* VIOLATION line
PASS: F0 the live snapshot carries the figure keys
PASS: F1 non-vacuity: discovery found a real suite count (an unread count is never a pass)
PASS: F2 exactly one Live line carries a suite figure
PASS: F3 and that figure equals what regression-run --list discovers
PASS: F4 no figure - and, since (119, C33), no numeric - Live: bullet - anywhere in the registry is undated
PASS: F5 and nothing is stale, duplicate or missing
PASS: F6 live tree: no figure_* VIOLATION line
PASS: F7 the live snapshot carries the COMMITTED count as well as the discovered one (83)
PASS: F8 non-vacuity: the committed side was really read, not reported as na (83)
PASS: G0 the live snapshot carries the badvalue keys
PASS: G1 non-vacuity: at least 10 tools advertise --format, so the scope was really read
PASS: G2 every tool advertising --format refuses an UNKNOWN value with exit 2
PASS: G3 and every one refuses a MISSING value with exit 2
PASS: G4 non-vacuity: the seven tables whose own exit-2 bullet promises it are in scope (71)
PASS: G5 and every one of them refuses --timeout 0 with exit 2 - the over-reach guard
PASS: G6 live tree: no badvalue_* VIOLATION line
PASS: G7 the live snapshot carries the (69) port keys
PASS: G8 non-vacuity: at least 3 PORT flags are really advertised, so that scope was read
PASS: G9 every PORT flag refuses a VALUE that cannot be a port with exit 2
PASS: G10 and every one refuses a MISSING value with exit 2
PASS: G11 the live snapshot carries the (72) missing-value keys
PASS: G12 non-vacuity: at least 20 value-taking flags are really advertised, so that scope was read
PASS: G13 every other value-taking flag refuses a MISSING value with exit 2
PASS: G14 non-vacuity: the independent recount found a real scope
PASS: G15 and miss + fmt + port add up to it - one probe per flag, none skipped
PASS: G16 the live snapshot carries the (75) convention keys
PASS: G17 non-vacuity: the spacing-independent reader saw a real scope
PASS: G18 and no option list hides a value-taking flag from the probing scope
PASS: G19 live tree: no help_convention VIOLATION line
PASS: G20 the live snapshot carries the (77) synopsis keys
PASS: G21 non-vacuity: at least 15 tools with an option list were really read
PASS: G22 every tool with an option list names a flag on its own usage row
PASS: G23 live tree: no synopsis_flags VIOLATION line
PASS: G24 the live snapshot carries the (73) class keys
PASS: G25 non-vacuity: at least one enum metavar is really advertised, so that class read a scope
PASS: G26 every enum metavar refuses a value outside its own list with exit 2
PASS: G27 non-vacuity: at least 10 quantity metavars are really advertised, so that class read a scope
PASS: G28 every quantity metavar refuses a value that is not a number with exit 2
PASS: G29 and the six buckets add up to the scope G14 recounted - one pair each, none skipped
PASS: G30 and the remainder with no type-invalid value at all is counted rather than dropped (22 pairs today)
PASS: G31 the live snapshot carries the (79) pair keys
PASS: G32 non-vacuity: at least 30 option-list pairs were really read
PASS: G33 every pair the option list advertises is named on its own labelled row
PASS: G34 and it never reads MORE pairs than the option-list reader counted
PASS: G35 the live snapshot carries the (98) declared-contract keys
PASS: G36 exactly three pairs are declared non-2 (repo-lint --sha, immich-roundtrip --email, smtp-relay-probe --expect-open)
PASS: G37 and every one of the three still answers what DECLARED_PAIRS declares (3, 3, and never 2)
PASS: G38 and the three left cl_other rather than being counted twice
PASS: G39 and the codes the three probes actually returned are printed for a reader to watch

=== Results: 373 passed, 0 failed ===

real	6m9.537s
user	4m33.596s
sys	2m14.791s
> build · mimo-v2.6-flash-free
373 passed / 0 failed (339 → 373, +34). Now update the declared counts:
← Edit repo/tests/test_registry_coverage.sh
Index: repo/tests/test_registry_coverage.sh
===================================================================
--- repo/tests/test_registry_coverage.sh
+++ repo/tests/test_registry_coverage.sh
@@ -1334,9 +1334,9 @@
 
 sections:
   A  the live tree is fully covered (the assertion this suite exists for)
   B  the counts are cross-checked against an independent recount of both inputs
-  C  37 negative controls: each plant is proved to have taken effect first,
+  C  39 negative controls: each plant is proved to have taken effect first,
      then the SAME checker (re-executed as a child of this script) must go red
      on that specific VIOLATION line
   D  this suite's own --help
   E  the exit-code tables are true, not merely present (one bounded probe per

← Edit repo/tools/REGISTRY.md
Index: repo/tools/REGISTRY.md
===================================================================
--- repo/tools/REGISTRY.md
+++ repo/tools/REGISTRY.md
@@ -3659,9 +3659,9 @@
   --expect-open … 127.0.0.1:25`, which dials the container's own Postfix over
   loopback and never the tool's default set (that set carries a VPS address):
   **A** the live tree is fully covered (exit 0, zero violations, no `ERROR`),
   **B** seven cross-checks against an independent recount (tools, sections, tool
-  sections, `--help` probed, `--help` answered, two floors), **C** 36 negative
+  sections, `--help` probed, `--help` answered, two floors), **C** 39 negative
   controls — unregistered tool, deleted section, stale tool heading, stale suite
   heading, one missing field each of Usage/Purpose/exit codes, `--help` exiting
   non-zero, `--help` silent, missing registry, missing tools dir, empty tools
   dir, sectionless registry, a tool that swallows an unknown flag (**C14**),

> build · mimo-v2.6-flash-free
← Edit repo/tools/REGISTRY.md
Index: repo/tools/REGISTRY.md
===================================================================
--- repo/tools/REGISTRY.md
+++ repo/tools/REGISTRY.md
@@ -3647,15 +3647,17 @@
 `REGISTRY_COVERAGE_CHILD` is internal (the re-exec branch). Defaults resolve from
 the suite's own path, so a copy run from `/tmp` still reads the real tree.
 
 **Tests**:
-- `bash tests/test_registry_coverage.sh` → **339 assertions** (since (119),
+- `bash tests/test_registry_coverage.sh` → **373 assertions** (since (119),
   `[0.4.176]`, which added control **C33**'s 18, and (120), `[0.4.177]`,
   which added **C34**'s 12 and **C35**'s 10, and (121), `[0.4.178]`, which
-  added **C36**'s 19 — 320 → 339 on 2026-10-03; it was **298** from
+  added **C36**'s 19 — 320 → 339 on 2026-10-03 — and (122), `[0.4.179]`,
+  which added **C37**'s 19, **C38**'s 10 and **C39**'s 4 plus **C36t** —
+  339 → **373** on 2026-10-03; it was **298** from
   `[0.4.176]` until that day and **280** from
-  `[0.4.150]`), ~325s (measured
-  5m24s on 2026-10-03, against 4m11s at 320 and 3m53s at its previous 264 and 3m38s at 248), no
+  `[0.4.150]`), ~370s (measured
+  6m10s on 2026-10-03, against 5m24s at 339, 4m11s at 320 and 3m53s at its previous 264 and 3m38s at 248), no
   EXTERNAL network — the one probe that opens a socket is `smtp-relay-probe
   --expect-open … 127.0.0.1:25`, which dials the container's own Postfix over
   loopback and never the tool's default set (that set carries a VPS address):
   **A** the live tree is fully covered (exit 0, zero violations, no `ERROR`),

← Edit repo/tools/REGISTRY.md
Index: repo/tools/REGISTRY.md
===================================================================
--- repo/tools/REGISTRY.md
+++ repo/tools/REGISTRY.md
@@ -3955,9 +3955,16 @@
 `[0.4.178]` gave that one reader a reader of its own: the definition's bytes
 are now quoted at each rendering of its shape — this suite's `--help`, the
 comment directly above the definition and section **F** above — and control
 **C36** compares every copy with the `LIVE_RE=` line byte-for-byte, one plant
-per side, the document and the code (320 → **339**)
+per side, the document and the code (320 → **339**), and queue item **(122)**
+`[0.4.179]` closed what that left open: each of those three renderings also
+carried a hand-written *sentence* interpreting the bytes beside the quote,
+with no reader for any of them, so **C37** plants all three sites (document,
+comment, delivered `--help`), **C38** replays the update (121) itself
+*requires* — every quote refreshed, every sentence left behind — and **C39**
+the definition the derivation cannot read, which is `ERROR` and exit 3 rather
+than a gloss passed over (339 → **373**)
 
 ## test_detached_children.sh
 
 **Purpose**: The detached-child sweep of queue item **(80)**, made permanent.

> build · mimo-v2.6-flash-free
$ cd /data/repo && head -5 CHANGELOG.md; echo "..."; grep -n "^## \[0.4.178\]" CHANGELOG.md; sed -n "$(grep -n '^## \[0.4.178\]' CHANGELOG.md | cut -d: -f1),+40p" CHANGELOG.md
# Changelog

All notable changes to Gladex are documented here.

Format: `[version] - YYYY-MM-DD — description`
...
7085:## [0.4.178] - 2026-10-03 — queue item (121): `LIVE_RE` had **one reader and three descriptions**, and none of the descriptions had a reader — the shape is now **quoted byte-for-byte** at all three renderings and compared with the definition by **C36** (suite **320 → 339**)
## [0.4.178] - 2026-10-03 — queue item (121): `LIVE_RE` had **one reader and three descriptions**, and none of the descriptions had a reader — the shape is now **quoted byte-for-byte** at all three renderings and compared with the definition by **C36** (suite **320 → 339**)

### Fixed

- **A description in words cannot be compared with a regex — so the shape is now quoted, not described.** After `[0.4.177]` the *rule* had one code path, but what the rule *means* was still written out in prose in three places: this suite's `--help` ("optional indent, optional bullet, optional bold"), the comment directly above the definition ("optional indent + optional `- ` bullet + optional `**`") and `tools/REGISTRY.md` section F ("`LIVE_RE` = optional indent + optional `- ` bullet + optional `**`"). Nothing read any of the three.
- **Measured by mutant, not argued:** deleting the `(\*\*)?` group from `LIVE_RE` left the suite **319 passed / 1 failed**, the single red being `F8`'s `got=na` (the `/tmp` copy has no git repository — the documented non-vacuity path, present with and without the mutant). Green, while all three renderings described a shape the code no longer had. Two bold claim lines existed in the registry — `- **Live (real zone, 2026-09-24)**:` and `- **Live**: \`red-watch\` …` — and they are *dated* and *digit-free* respectively, so no other rule noticed their disappearance either.
- **The shape chosen, once:** quote the literal, byte-for-byte, at every rendering and compare each copy with the definition. All three now carry `LIVE_RE='^[[:space:]]*(- )?[[:space:]]*(\*\*)?[[:space:]]*Live'`, and section F's prose gloss follows the literal instead of standing in for it.

### Changed

- **New `live_re_quote` finding with one name per stale copy** — `help`, `suite` or `registry` — emitted by `run_check` after section F, so a reader learns *which* rendering drifted rather than only that something did. The source of truth is the `LIVE_RE=` line in the suite file in scope, which in the live run is this very script; the site read as *delivered* help (`bash <file> --help`) rather than as text in a heredoc nobody prints, and the `suite` site is everything above the definition line, i.e. the comment a maintainer reads first.
- **`live_re_sites=3` joined the `SUMMARY` line** as the non-vacuity floor: a loop over zero sites is a clean verdict over nothing, the same silent pass `conv_pairs`/`syn_n`/`cl_scope` already refuse. `C36j` asserts it on a stale copy and `C36r` again on a defined copy, because a stale rendering must never shrink the scope being compared.
- **An unreadable definition is an `ERROR`, not a comparison:** a suite file with no `LIVE_RE=` line returns 3 with `ERROR live_re_source`, following the repo's "an input that was not read is never a pass" convention rather than comparing an empty string against three documents.
- **One trap found by building it, fixed in the same commit:** the child hook (`REGISTRY_COVERAGE_CHILD=1` → `run_check`) sits *above* the `--help` branch, so the nested help probe inherited it and re-entered the whole checker instead of printing help — recursion bounded only by the probe's own `timeout 20`, and it reported `live_re_quote help` as a false red on the first run. The probe now runs in a subshell that `unset`s the hook; the reason is written at the site, not in a commit message.

### Tests

- **`C36` (19 assertions), two plants — one per SIDE of a single equality.** *Plant A, the document side:* `tools/REGISTRY.md`'s quoted literal is rewritten to the mutant while the suite copy in the same sandbox keeps the definition → `VIOLATION live_re_quote registry`, **exactly 1** violation, with `C36h`/`C36i` proving the *other* two renderings are **not** blamed and `C36j` proving all three were still compared. *Plant B, the code side — (121)'s own mutant:* the `LIVE_RE=` line in the suite copy loses its bold group while every rendering keeps the old bytes → `help`, `suite` **and** `registry` all red, **exactly 3** violations and nothing else, with `C36s` asserting `fig_live` still reads 1 — the children execute the *live* checker, so a plant changes only what is compared, never what compares.
- **Each plant is proved to have taken effect before the checker runs** (`C36c`/`C36d`, `C36k`/`C36l`, plus `C36b` on the derivation itself), so "caught" can never be true because the substitution matched nothing. Plant B's proof matches the whole definition line, not a substring: the `--help` body quotes the same bytes at column 0, which is exactly the ambiguity that made the earlier `^LIVE_RE=` extraction worth pinning to `head -1`.
- Suite **320 → 339 passed / 0 failed** (measured 5m24s); controls **C1…C36** contiguous (counted from the file); `- Live:` suite figure untouched (**83**); no tool source and no `app/src/php` file changed → **no reviewer gate, no promote**, prod untouched (**0.4.28**), `GLADEX_APP_VERSION` unchanged (this entry is appended at the bottom, so the changelog's *first* heading did not move).

### Notes

- **The class, stated once:** (119) widened what is checked, (120) made *which lines* code instead of comment, and this entry makes *what the code says* verifiable instead of narratable. Each step removed a copy rather than adding a fourth one — the glosses that remain in prose follow the quoted literal rather than defining it, so the reader's summary can be wrong without the rule being wrong.
- Costs one `bash <suite> --help` per `run_check` (≈37 per run): measured whole-suite time went **4m11s → 5m24s** at 320 → 339 assertions, i.e. the two extra sandboxes and children, not the probe.

### 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 && wc -l CHANGELOG.md; tail -5 CHANGELOG.md; grep -c '^## \[' CHANGELOG.md; grep '^## \[' CHANGELOG.md | sort | uniq -d
7112 CHANGELOG.md
- **The class, stated once:** (119) widened what is checked, (120) made *which lines* code instead of comment, and this entry makes *what the code says* verifiable instead of narratable. Each step removed a copy rather than adding a fourth one — the glosses that remain in prose follow the quoted literal rather than defining it, so the reader's summary can be wrong without the rule being wrong.
- Costs one `bash <suite> --help` per `run_check` (≈37 per run): measured whole-suite time went **4m11s → 5m24s** at 320 → 339 assertions, i.e. the two extra sandboxes and children, not the probe.

### 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`.
183
> build · mimo-v2.6-flash-free

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

[Sat Oct  3 07:15:53 2026] 127.0.0.1:53624 Accepted
[Sat Oct  3 07:15:54 2026] 127.0.0.1:53624 Closing
[Sat Oct  3 07:15:54 2026] 127.0.0.1:53628 Accepted
[Sat Oct  3 07:15:54 2026] 127.0.0.1:53628 Closing
[Sat Oct  3 07:17:46 2026] 127.0.0.1:35052 Accepted
[Sat Oct  3 07:17:47 2026] 127.0.0.1:35052 Closing
[Sat Oct  3 07:17:47 2026] 127.0.0.1:35054 Accepted
[Sat Oct  3 07:17:47 2026] 127.0.0.1:35054 Closing
[Sat Oct  3 07:17:47 2026] 127.0.0.1:35062 Accepted
[Sat Oct  3 07:17:47 2026] 127.0.0.1:35062 Closing
[Sat Oct  3 07:17:47 2026] 127.0.0.1:35072 Accepted
[Sat Oct  3 07:17:47 2026] 127.0.0.1:35072 Closing
[Sat Oct  3 07:17:48 2026] 127.0.0.1:35084 Accepted
[Sat Oct  3 07:17:48 2026] 127.0.0.1:35084 Closing
[Sat Oct  3 07:17:48 2026] 127.0.0.1:35100 Accepted
[Sat Oct  3 07:17:48 2026] 127.0.0.1:35100 Closing
[Sat Oct  3 07:17:48 2026] 127.0.0.1:35110 Accepted
[Sat Oct  3 07:17:48 2026] 127.0.0.1:35110 Closing
[Sat Oct  3 07:17:49 2026] 127.0.0.1:35118 Accepted
[Sat Oct  3 07:17:49 2026] 127.0.0.1:35118 Closing
[Sat Oct  3 07:17:50 2026] 127.0.0.1:35130 Accepted
[Sat Oct  3 07:17:50 2026] 127.0.0.1:35130 Closing
[Sat Oct  3 07:17:50 2026] 127.0.0.1:35142 Accepted
[Sat Oct  3 07:17:50 2026] 127.0.0.1:35142 Closing
[Sat Oct  3 07:17:51 2026] 127.0.0.1:35154 Accepted
[Sat Oct  3 07:17:51 2026] 127.0.0.1:35154 Closing
[Sat Oct  3 07:17:51 2026] 127.0.0.1:35164 Accepted
[Sat Oct  3 07:17:51 2026] 127.0.0.1:35164 Closing
[Sat Oct  3 07:17:51 2026] 127.0.0.1:35176 Accepted
[Sat Oct  3 07:17:51 2026] 127.0.0.1:35176 Closing
[Sat Oct  3 07:17:53 2026] 127.0.0.1:35190 Accepted
[Sat Oct  3 07:17:53 2026] 127.0.0.1:35190 Closing
[Sat Oct  3 07:17:53 2026] 127.0.0.1:35194 Accepted
[Sat Oct  3 07:17:53 2026] 127.0.0.1:35194 Closing
[Sat Oct  3 07:17:53 2026] 127.0.0.1:35202 Accepted
[Sat Oct  3 07:17:53 2026] 127.0.0.1:35202 Closing
[Sat Oct  3 07:17:54 2026] 127.0.0.1:35216 Accepted
[Sat Oct  3 07:17:54 2026] 127.0.0.1:35216 Closing
[Sat Oct  3 07:17:54 2026] 127.0.0.1:35232 Accepted
[Sat Oct  3 07:17:54 2026] 127.0.0.1:35232 Closing
[Sat Oct  3 07:17:55 2026] 127.0.0.1:60060 Accepted
[Sat Oct  3 07:17:55 2026] 127.0.0.1:60060 Closing
[Sat Oct  3 07:21:58 2026] 127.0.0.1:51808 Accepted
[Sat Oct  3 07:21:58 2026] 127.0.0.1:51808 Closed without sending a request; it was probably just an unused speculative preconnection
[Sat Oct  3 07:21:58 2026] 127.0.0.1:51808 Closing
[Sat Oct  3 07:21:59 2026] 127.0.0.1:51820 Accepted
[Sat Oct  3 07:21:59 2026] 127.0.0.1:51820 Closed without sending a request; it was probably just an unused speculative preconnection
[Sat Oct  3 07:21:59 2026] 127.0.0.1:51820 Closing
[Sat Oct  3 07:22:34 2026] 127.0.0.1:49178 Accepted
[Sat Oct  3 07:22:34 2026] 127.0.0.1:49178 Closed without sending a request; it was probably just an unused speculative preconnection
[Sat Oct  3 07:22:34 2026] 127.0.0.1:49178 Closing
[Sat Oct  3 07:22:35 2026] 127.0.0.1:49898 Accepted
[Sat Oct  3 07:22:35 2026] 127.0.0.1:49898 Closed without sending a request; it was probably just an unused speculative preconnection
[Sat Oct  3 07:22:35 2026] 127.0.0.1:49898 Closing
[Sat Oct  3 07:23:22 2026] 127.0.0.1:35240 Accepted
[Sat Oct  3 07:23:22 2026] 127.0.0.1:35240 Closing
[Sat Oct  3 07:27:25 2026] 127.0.0.1:45976 Accepted
[Sat Oct  3 07:27:26 2026] 127.0.0.1:45976 Closing
[Sat Oct  3 07:27:26 2026] 127.0.0.1:45986 Accepted