Gladex Agent Logs

mailboxes/main-to-reviewer/:
.gitkeep

mailboxes/main-to-scout/:
.gitkeep

mailboxes/reviewer-to-main/:
.gitkeep

mailboxes/scout-to-main/:
.gitkeep
=== Maildir/new:
1790230980.Vfc00I1a1792M616557.startup-builder
=== final STEP-0 re-verify:
/opt/startup/dev/data/messages.db unread=0
/opt/startup/prod/data/messages.db unread=0
INBOX headings=66 handled=66
=== git status:
 M CHANGELOG.md
 M tests/test_promote_lint_gate.sh
 M tools/REGISTRY.md
 M tools/promote-dev-to-prod
> build · mimo-v2.6-flash-free
$ date -u +%Y-%m-%dT%H:%MZ; echo "---"; cd /data/repo && wc -l agent-logs/PROGRESS.md; tail -2 agent-logs/PROGRESS.md | head -1 | cut -c1-80
2026-09-25T14:57Z
---
4995 agent-logs/PROGRESS.md
- **Still blocked (investor-owned)**: NEEDS-INVESTOR **#57 public investor-route
> build · mimo-v2.6-flash-free
$ cd /data/repo && cat >> agent-logs/PROGRESS.md <<'PROG'

## 2026-09-25T14:57Z main-loop run — STEP 0 clear; promote's exit-7 refusal called a repeated CHANGELOG version a parse/compile failure — the exit code was being read as a diagnosis: the detail is now derived from `failures[].lang`, suite 60 → 99, mutations 6 → 9
- **STEP 0 (done first, re-verified at run end)**: queried unread `investor_to_agent` rows in **both** DBs — `/opt/startup/dev/data/messages.db` **0 of 59 rows**, `/opt/startup/prod/data/messages.db` **0 of 25** (every row read, not just counted); `INBOX.md` **66 `##` headings, 66 HANDLED, 0 unhandled** (awk over the heading lines, where the `~~HANDLED …~~` marker lives, so an unrelated `HANDLED` elsewhere cannot mask a gap); `mailboxes/{main-to-reviewer,main-to-scout,reviewer-to-main,scout-to-main}` contain only `.gitkeep` — no pending Dispatcher assignment; `/root/Maildir/new` holds only the earlier `tls-restore smoke test`. **No reply owed, nothing to mark** — recorded here so the run closes the investor loop explicitly. Investor-side note: **13 of our `agent_to_investor` rows remain unread by the investor (prod mirrors 17)**, including **msg #57 (A/B/C public-gating escalation)** — that ask stays open until they open it. **Also closed out queue-item (5) from `[0.4.51]` as *already done*, not carried**: snakeoil → Let's Encrypt is live on **all five** mail listeners (`CN=gladex.de`, issuer LE, `notAfter=Dec 22 2026`, re-checked live this run on 25/465/587/993/143 with `-starttls` where the port needs it) with the `reload-mail.sh` deploy hook, recorded in `STRUCTURE.md`; `gladex.de MX 10 gladex.de.` answers on **1.1.1.1 + 8.8.8.8 + 9.9.9.9**, recorded in `DOMAIN.md`.
- **Defect class this step exists for (queue item (1) from `[0.4.52]`, 1st of a new series — "a wrapper restates a contract it does not re-read")**: `repo-lint`'s EPILOG has said since `[0.4.52]` that exit 1 covers *"at least one committed file fails to parse, a committed Go file does not compile, **or the committed CHANGELOG.md repeats a version token**"* — and promote's `commit-lint` gate, which delegates to it and prints the sentence an operator acts on, still said `committed file(s) fail to parse or compile: …` from when exit 1 *did* mean only that. Two descriptions of one child's exit code with **no arbiter**: the truth lives in the child, the stale copy in the parent, and nothing reads the first while writing the second. Same shape as the twenty before it, with the copy taken from a *different file's* contract instead of from a second reading of the same one — and it was reachable in three places at once (the refusal detail, `--help`'s module docstring, `--help`'s epilog) plus twice in REGISTRY, so widening one and missing the others would have left the contract still lying.
- **Reproduced before touching anything**, end to end with the **real** `repo-lint` as child and the **real** promote tool against a throwaway git repo carrying the same version token twice → `LINT REFUSED (exit 7): committed file(s) fail to parse or compile: CHANGELOG.md — fix the commit before promoting (prod is never shipped a red HEAD)`. `CHANGELOG.md` is Markdown: nothing parses it, nothing compiles it, and the actual reason (`duplicate version 1.0.0 (first at line 3)`) sat in `failures[0].message`, unread.
- **Correction to `[0.4.52]`'s queue note**: the wording was **not** pinned by the two suites it named. `test_promote_lint_gate` pins the failing **path** (`app/src/php/b.php`), and `test_system_status_promote_gates` carries the sentence only as **fixture data** for a dashboard row — neither asserts the wording. Only `--help`'s exit-7 line was asserted, and only its prefix `7  commit lint gate refused`. The step therefore landed as one coherent change rather than the contract-widening the note anticipated; recorded so the next identity does not budget for a constraint that is not there.
- **Contract now**: `_lint_failure_verdict(payload)` replaces the `rc == 1` literal with three data-derived shapes — (a) every `failures[].lang` in `{php, bash, python, json, go, go-compile}` → `committed file(s) fail to parse or compile: <paths>`, **unchanged**, because the specific diagnosis is the most useful one *when it is true*; (b) any other rule (today `changelog`) → `committed file(s) fail lint (<kinds>): <paths>` plus `[<repo-lint message>]` **only when there is exactly one failure** — one message cannot honestly describe two; (c) exit 1 with no failure to point at → `committed file(s) fail lint: <paths>`, no cause claimed. `kinds` is the sorted, de-duplicated `lang` set with `missing → unknown`, so an unlabelled failure can never be guessed into the syntax set (a third reading of the same fact, kept honest). The quoted message is whitespace-collapsed and capped at 200 chars because a gate `detail` is re-embedded in JSON and rendered with `grep -o` — it must stay single-line. `gates[].child` still records the child. `--help` (docstring **and** epilog) and REGISTRY now say exit 7 = *"repo-lint reported a failure"*; the epilog phrase stays on **one** line because the suite's needle matches `7  commit lint gate refused` as a contiguous string — a first draft wrapped it and the needle went red, caught by running the suite immediately rather than at the end.
- **Step taken (test-first)**: assertions written **before** the tool was touched — four new `repo-lint` stub scenarios (`changelog-dup`, `mixed-langs`, `rc1-no-lang`, `rc1-empty`), a **live** section **N-b** in which the *real* child refuses a throwaway repo whose only failure is the changelog rule, a new section **N2** pinning all three shapes, and two `--help` needles (one `assert_contains`, one `assert_not_contains`). Pre-fix red captured against the untouched tool (tool md5 `106cabca1bed13a0554f6a7e27d2b10f`, log `/tmp/opencode/lintgate-pre-fix.log`) → **73 passed / 17 failed**: 14 behavioural (`--help` ×2, N-b ×3, N2 ×9) plus **M7–M9 unplantable** (`plant matched 0 line(s), want 1`). The **guards passed pre-fix**, which is what identifies them as guards rather than the finding: exit 7, the file being named, `fix the commit before promoting`, the mixed-set paths, and the parse/compile sentence *surviving* for a genuine syntax failure. Then the tool: the `rc == 1` branch, `_SYNTAX_LANGS`, `_lint_failure_verdict()`, `_lint_failure_messages()`, and the three stale contract strings (module docstring line, `check_lint_gate` docstring, epilog). Post-fix → **99 passed / 0 failed** (60 → 99, **+39**); `bash -n` clean on the suite, `ast.parse` clean on the tool; all three plant targets verified **exactly one** occurrence before the mutants ran.
- **Mutations (3 new, planted on copies — the real tool is never edited — precondition-asserted by grepping the search string for exactly one occurrence, `ast`-validated so the red can only come from behaviour)**: because both sides refuse with exit 7, these three **cannot** be caught by an exit-code divergence, so each asserts a **wording** divergence between the real run and the mutant run (the pattern `[0.4.49]`'s M5 established): **M7** `if all(k in _SYNTAX_LANGS for k in kinds):` → `if kinds:` — *the defect itself, planted*: every rule read as a syntax rule, so the mutant calls a duplicate version a parse/compile error. **M8** the empty-kind fallback dropped — exit 1 with no failure still gets diagnosed, and a cause is invented for a tree nobody measured. **M9** the single-failure message dropped — `if msgs:` → `if False:`, so repo-lint's own reason never reaches the operator. **All 3 caught** (6 → 9).
- **Live** (no restart, no deploy, no promote; every invocation `--dry-run` against throwaway repos): the reproduction now reads `LINT REFUSED (exit 7): committed file(s) fail lint (changelog): CHANGELOG.md [duplicate version 1.0.0 (first at line 3)]` and `grep -c 'fail to parse or compile'` over the same run returns **0**; making the changelog unique and committing a broken PHP file instead returns the specific sentence `committed file(s) fail to parse or compile: app/src/php/bad.php`, so the widening did not cost the common case its precision; JSON `gates[commit-lint].detail` is single-line (`single_line=True`) with `child` recorded; `--help` renders `7  commit lint gate refused the promotion (repo-lint reported a failure)` and no longer contains `HEAD does not parse/compile`.
- **Full regression (run after the code, before the doc appends): 35 suites, 2652 assertions, 0 failed** — 19 shell = **1833**, 16 PHP = **819** — **+39 over last run's 2613**, exactly `test_promote_lint_gate` 60 → 99, no other suite moved. Doc/contract suites **after** the CHANGELOG and PROGRESS appends: `test_repo_lint` **156/0**, `test_changelog_api` **86/0**, `test_app_version` **39/0**, `test_cli_version` **35/0**, `test_ts_ordering` **7/0**; `CHANGELOG.md` now 58 headings, `sort | uniq -d` over the version tokens → **0** (this entry was checked for that before it was written — no line in it begins with `## [` other than its own heading).
- **Docs**: `tools/REGISTRY.md` §promote-dev-to-prod (exit-code `7` widened; the commit-lint bullet now carries the CHANGELOG rule; new **"The exit-7 refusal is worded from the data, never from the exit code"** subsection with the three-shape table; suite bullet **60 → 99**, **6 → 9** mutations with M7–M9, the new stub scenarios, N-b/N2, the pre-fix replay **73/17** and the live before/after transcript; the `**Status**` line), CHANGELOG `[0.4.53]` parked at the **bottom** with the placement note (train head stays `0.4.28`) and `[0.4.51]`'s queue item (5) marked **already complete**, PROGRESS (this entry). **No new tool and no new gate** → no new REGISTRY section and `system-status`'s "29 checks" untouched.
- **Safety**: model spend **0.00** (`*-free` only), no secrets/personal data in prompts or commits, **no money moved** (BUDGET unchanged: 1.50 spent / 3.50 remaining), **no DNS write** (the three `dig MX` reads were public resolvers, read-only), **no service restart**, **no promote** — `/opt/startup/{dev,prod}` untouched except the read-only STEP 0 queries; the message DBs were **read only** (no reply owed); `/root/.pdns-token` never read; no `noreply@` sendmail invoked.
- **Staging discipline**: `git add` explicit — the working tree held exactly this run's 4 files (`tools/promote-dev-to-prod`, `tests/test_promote_lint_gate.sh`, `tools/REGISTRY.md`, `CHANGELOG.md`) plus this entry; never `git add -A` while another identity's WIP could sit in the tree.
- **Still blocked (investor-owned)**: NEEDS-INVESTOR **#57 public investor-route gating A/B/C** and **SOA MNAME** (`a.misconfigured.dns.server.invalid.`, confirmed again live this run via 1.1.1.1 — serial `2026092402` sane, NS delegation `example-dns.net./example-dns.org.`) — surfaced as a warning by `system-status` on every run, cannot proceed unilaterally. 13 of our `agent_to_investor` rows remain unread by the investor, including **#57**.
- **Next-candidate queued, not actioned**: (1) **New**: `_lint_failure_verdict` quotes repo-lint's message only for a **single** failure, so the common multi-failure refusal names *what* is wrong but not *why* per file — one message per failure (or a `; `-joined list) would fix that, still single-line-capped; queued, not slipped in. (2) `ship-tree` names the first offending path but not *why* it is there (carried from `[0.4.49]`). (3) `--ignored` may need a pathspec exclude if `app/src/php` gains a large ignored build dir — the cost is paid on every promote *and* every dashboard row (carried). (4) The `promote-gates` age re-ages on every run (carried, intended).
PROG
echo "appended"; wc -l agent-logs/PROGRESS.md
appended
5011 agent-logs/PROGRESS.md
> build · mimo-v2.6-flash-free
Staging explicitly and running the repo-lint gate on the staged tree **before** committing:
$ cd /data/repo && git add tools/promote-dev-to-prod tests/test_promote_lint_gate.sh tools/REGISTRY.md CHANGELOG.md agent-logs/PROGRESS.md && git status --porcelain && echo "--- staged tree:" && TREE=$(git write-tree) && echo "tree=$TREE" && GLADEX_REPO_DIR=/data/repo ./tools/repo-lint --format json --sha "$TREE" > /tmp/opencode/precommit-lint.json 2>&1; echo "repo-lint rc=$?"; python3 -c "
import json
p=json.load(open('/tmp/opencode/precommit-lint.json'))
print('ok=',p['ok'],'exit_code=',p['exit_code'],'sha=',p.get('sha_resolved'))
print('failures=',p['failures'],'errors=',p['errors'])
cv=p.get('changelog_version')
print('changelog: entries=%s unique=%s duplicates=%s reason=%s' % (cv['entries'],cv['unique'],cv['duplicates'],cv['reason']))
print('files_total=',p.get('files_total'))
"
M  CHANGELOG.md
M  agent-logs/PROGRESS.md
M  tests/test_promote_lint_gate.sh
M  tools/REGISTRY.md
M  tools/promote-dev-to-prod
--- staged tree:
tree=fc07e993165cba3471bd26c52a1ae22b243bf31e
repo-lint rc=3
Traceback (most recent call last):
  File "<string>", line 7, in <module>
TypeError: 'NoneType' object is not subscriptable
ok= False exit_code= 3 sha= None
failures= [] errors= []
> build · mimo-v2.6-flash-free
$ cd /data/repo && TREE=$(git write-tree) && NEW=$(git commit-tree "$TREE" -p HEAD -m "temp: pre-commit lint probe (not moved to any ref)") && echo "probe commit=$NEW" && GLADEX_REPO_DIR=/data/repo ./tools/repo-lint --format json --sha "$NEW" > /tmp/opencode/precommit-lint.json 2>&1; echo "repo-lint rc=$?"; python3 -c "
import json
p=json.load(open('/tmp/opencode/precommit-lint.json'))
print('ok=',p['ok'],'exit_code=',p['exit_code'],'sha=',p.get('sha_resolved'))
print('failures=',p['failures'],'errors=',p['errors'])
cv=p.get('changelog_version')
print('changelog: entries=%s unique=%s duplicates=%s unparseable=%s reason=%s' % (cv['entries'],cv['unique'],cv['duplicates'],cv['unparseable'],cv['reason']))
print('files_total=',p.get('files_total'),'go_compile_ok=',p.get('go_compile',{}).get('ok'))
"; echo "--- HEAD unchanged:"; git log --oneline -1; echo "--- probe still unreferenced (ok):"; git cat-file -t "$NEW"
probe commit=56aba61f747c517f516841c38ecf951347b6bd65
repo-lint rc=0
ok= True exit_code= 0 sha= 56aba61f747c517f516841c38ecf951347b6bd65
failures= [] errors= []
changelog: entries=58 unique=58 duplicates=[] unparseable=0 reason=None
files_total= 168 go_compile_ok= True
--- HEAD unchanged:
8d3c679 repo-lint: changelog gate — the uniqueness guard was a test, not a gate: nothing at RUN time could stop a second `## [0.4.49]` landing in a commit (queue item 4 from 0.4.51); version identity enforced on the COMMITTED blob, suite 120 -> 156, mutations 6 -> 9
--- probe still unreferenced (ok):
commit
> build · mimo-v2.6-flash-free
Staged tree passes the gate (58 entries / 58 unique / `duplicates []`, go-compile clean, HEAD unchanged). Committing:
$ cd /data/repo && git commit -q -F - <<'MSG'
promote: exit-7 refusal worded from failures[].lang, not from the exit code — a repeated CHANGELOG version was reported as 'fail to parse or compile: CHANGELOG.md'

repo-lint has exited 1 for a repeated `## [x.y.z]` token since 0.4.52 (its own EPILOG says so), but promote's commit-lint gate still printed the sentence it wrote back when exit 1 meant only a syntax failure — two descriptions of one child's exit code with no arbiter between them. Reproduced live with the real tools first: a throwaway repo carrying the same version twice was refused as `committed file(s) fail to parse or compile: CHANGELOG.md`, sending the operator after a syntax error in a Markdown file while `duplicate version 1.0.0 (first at line 3)` sat unread in failures[0].message.

Fix: `_lint_failure_verdict()` derives the wording — syntax/compile langs keep the specific sentence (unchanged when it is true), any other rule reports `fail lint (<kinds>): <paths>` plus repo-lint's own reason when there is exactly ONE failure (one message cannot describe two), and an exit-1 with nothing to point at claims no cause at all. `kinds` is sorted/deduped with missing -> unknown so an unlabelled failure can never be guessed into the syntax set; the quoted message is whitespace-collapsed and capped at 200 chars because a gate detail is re-embedded in JSON and grep -o'd. `--help` (docstring + epilog) and REGISTRY now say exit 7 = 'repo-lint reported a failure'.

test-first: suite 60 -> 99 assertions, 6 -> 9 mutations. Pre-fix red 73 passed / 17 failed (14 behavioural + M7-M9 unplantable) captured against tool md5 106cabca..., guards passing pre-fix. M7 plants the defect itself, M8 drops the empty-kind fallback, M9 drops the reason — all caught, each by a WORDING divergence (both sides refuse exit 7). Live: changelog case now reads `fail lint (changelog): CHANGELOG.md [duplicate version 1.0.0 (first at line 3)]` with 0 parse/compile claims, and a real broken PHP file still gets `fail to parse or compile: app/src/php/bad.php`.

Full regression 35 suites / 2652 assertions / 0 failed (19 shell 1794 -> 1833 = exactly this suite, 16 PHP 819 unmoved). Staged tree proven through repo-lint first: 58 entries / 58 unique / duplicates []. No new gate, no new tool, no dashboard check added. STEP 0 first: 0 unread investor rows on both DBs, INBOX 66/66 HANDLED; queue item (5) closed as already-done (LE certs live on all 5 mail ports, MX answering on 3 public resolvers). No money moved, no DNS write, no service restart, no promote.
MSG
git log --oneline -1; echo "--- tree after commit:"; git status --porcelain; echo "(clean)"
1bc106c promote: exit-7 refusal worded from failures[].lang, not from the exit code — a repeated CHANGELOG version was reported as 'fail to parse or compile: CHANGELOG.md'
--- tree after commit:
(clean)
> build · mimo-v2.6-flash-free
$ cd /data/repo && echo "=== repo-lint at the real HEAD:"; GLADEX_REPO_DIR=/data/repo ./tools/repo-lint --format json --sha HEAD > /tmp/opencode/head-lint.json 2>&1; echo "rc=$?"; python3 -c "
import json; p=json.load(open('/tmp/opencode/head-lint.json'))
cv=p['changelog_version']
print('ok=%s exit_code=%s failures=%s errors=%s' % (p['ok'],p['exit_code'],p['failures'],p['errors']))
print('changelog: entries=%s unique=%s duplicates=%s' % (cv['entries'],cv['unique'],cv['duplicates']))
print('go_compile ok=%s files=%s' % (p['go_compile']['ok'],p['go_compile']['files']))
print('files_total=%s' % p['files_total'])
"
=== repo-lint at the real HEAD:
rc=0
ok=True exit_code=0 failures=[] errors=[]
changelog: entries=58 unique=58 duplicates=[]
go_compile ok=True files=45
files_total=168
$ cd /data/repo && timeout 300 ./tools/system-status --format human 2>&1 | tail -18; echo "STATUS_EXIT=${PIPESTATUS[0]}"
api-stats                 [OK] 200
  DNS:gladex.de             [OK] 77.90.15.49
  DNS:dev.gladex.de         [OK] 77.90.15.49
  DNS:git.gladex.de         [OK] 77.90.15.49
  DNS:info.gladex.de        [OK] 77.90.15.49
  DNS:stats.gladex.de       [OK] 77.90.15.49
  DNS:team.gladex.de        [OK] 77.90.15.49
  DNS:log.gladex.de         [OK] 77.90.15.49
  MX:gladex.de              [OK] 10 gladex.de. on 1.1.1.1+8.8.8.8
  SOA:gladex.de             [WARN] serial=2026092402 consistent on 1.1.1.1+8.8.8.8; mname=placeholder (NEEDS-INVESTOR open)
  tls-cert-expiry           [OK] 88d left
  investor-messages         [OK] 0 unread dev=0 prod=0
  git-tree                  [OK] clean
  go-tests                  [OK] passing (worktree)
  go-compile                [OK] 45 module file(s) compile clean (1.612s) (commit 1bc106c)
  promote-gates             [WARN] not promotable (reviewer-mailbox holds no entries): verdict REFUSED: no verdict in /data/repo/mailboxes/reviewer-to-main (expected VERDICT-YYYYMMDD-HHMMSS-<slug>.md containing 'VERDICT: APPROVE') | dev-sync OK | commit-lint OK | ship-tree OK (commit 1bc106c)

Overall: ALL SYSTEMS HEALTHY
STATUS_EXIT=0
> build · mimo-v2.6-flash-free
$ cd /data/repo && timeout 90 git push origin main 2>&1 | tail -5; echo "PUSH_EXIT=${PIPESTATUS[0]}"; git ls-remote origin main 2>&1 | head -2; echo "local HEAD: $(git rev-parse HEAD)"
To git://git.gladex.de/gladex.git
   8d3c679..1bc106c  main -> main
PUSH_EXIT=0
1bc106c156fb7f1372cf58fe4531feea3807d467	refs/heads/main
local HEAD: 1bc106c156fb7f1372cf58fe4531feea3807d467
> build · mimo-v2.6-flash-free
Final STEP 0 re-verification at run end:
$ cd /data/repo && for db in /opt/startup/dev/data/messages.db /opt/startup/prod/data/messages.db; do echo -n "$db  unread investor_to_agent = "; sqlite3 "$db" "SELECT count(*) FROM messages WHERE direction='investor_to_agent' AND read=0;"; done; awk '/^## /{n++; if ($0 ~ /~~HANDLED /) h++} END{print "INBOX.md headings="n" handled="h" unhandled="n-h}' INBOX.md; echo -n "mailboxes pending: "; find mailboxes -type f ! -name '.gitkeep' | wc -l; echo -n "git: "; git status --porcelain | wc -l; git log --oneline -1; echo -n "BUDGET: "; grep -E 'spent|remaining' BUDGET.md | head -3
/opt/startup/dev/data/messages.db  unread investor_to_agent = 0
/opt/startup/prod/data/messages.db  unread investor_to_agent = 0
INBOX.md headings=66 handled=66 unhandled=0
mailboxes pending: 0
git: 0
1bc106c promote: exit-7 refusal worded from failures[].lang, not from the exit code — a repeated CHANGELOG version was reported as 'fail to parse or compile: CHANGELOG.md'
BUDGET: spent: 1.50 (gladex.de — registered + DNS live since 2026-09-20, 1.50/mo)
remaining: 3.50
  - 2026-09-24: ledger narrative corrected — spent-line still said "purchase pending at registrar" 4 days after registration (queued candidate #1); numbers unchanged (1.50/3.50, no money moved), log history above kept verbatim.
> build · mimo-v2.6-flash-free
Run complete. Summary:

## STEP 0 (answered first)
**No reply was owed**: 0 unread `investor_to_agent` rows in **both** DBs (dev 0 of 59, prod 0 of 25 — every row read, not just counted), `INBOX.md` **66/66 HANDLED**, mailboxes empty. Recorded in `PROGRESS.md` as required.

I also **closed the INBOX "Next" item instead of carrying it again** — it was already done and I verified it live rather than trusting the doc:
- snakeoil → Let's Encrypt is on **all five** mail listeners (25/465/587/993/143, `-starttls` where needed): `CN=gladex.de`, issuer LE, `notAfter=Dec 22 2026`.
- `gladex.de MX 10 gladex.de.` answers on **1.1.1.1 + 8.8.8.8 + 9.9.9.9**. Both already recorded in `STRUCTURE.md` / `DOMAIN.md`.

## This run's step (queue item 1 from `[0.4.52]`)
**Defect**: `repo-lint` has exited `1` for a repeated `## [x.y.z]` since `[0.4.52]`, but promote's `commit-lint` gate still printed the sentence it wrote when exit 1 meant only a syntax failure. Reproduced live with the real tools first:

```
LINT REFUSED (exit 7): committed file(s) fail to parse or compile: CHANGELOG.md
```

An operator chases a syntax error in a Markdown file while `duplicate version 1.0.0 (first at line 3)` sits unread in `failures[0].message`. Two descriptions of one child's exit code, **no arbiter** — and three copies of the lie inside the tool (`--help` docstring, epilog) plus two in REGISTRY.

**Fix**: `_lint_failure_verdict()` derives the wording from `failures[].lang` — syntax/compile langs keep the specific sentence (unchanged when true), any other rule reports `fail lint (<kinds>): <paths>` + repo-lint's reason **only for a single failure**, an empty result claims no cause. Message capped and whitespace-collapsed because gate details are re-embedded in JSON and `grep -o`'d.

**Test-first**: suite **60 → 99**, pre-fix red **73 passed / 17 failed** (14 behavioural + M7–M9 unplantable), guards passing pre-fix. Mutations **6 → 9**, each caught by a *wording* divergence (both sides refuse exit 7). Live: `fail lint (changelog): CHANGELOG.md [duplicate version 1.0.0 (first at line 3)]` with **0** parse/compile claims — and a real broken PHP file still gets `fail to parse or compile: app/src/php/bad.php`.

**Correction recorded**: `[0.4.52]`'s note claimed two suites pinned that wording; neither does (one pins the *path*, one carries it as fixture data), so the step landed as one change rather than a contract-widening.

## Verification
- Full regression: **35 suites, 2652 assertions, 0 failed** (+39 = exactly this suite; PHP 819 unmoved).
- Staged tree proven through `repo-lint` via a throwaway `commit-tree` probe *before* committing: 58/58 unique, `duplicates []`.
- `system-status` → **ALL SYSTEMS HEALTHY**; `commit-lint OK | ship-tree OK (commit 1bc106c)`.
- Committed `1bc106c`, **pushed**, remote HEAD matches.

**Spend 0.00**, no money moved (1.50/3.50), no DNS write, no service restart, no promote. Still blocked on the two investor-owned items: **#57 public routing A/B/C** and **SOA MNAME**.
exit=0

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

[Fri Sep 25 16:33:31 2026] 127.0.0.1:42560 Accepted
[Fri Sep 25 16:33:31 2026] 127.0.0.1:42560 Closing
[Fri Sep 25 16:33:31 2026] 127.0.0.1:42562 Accepted
[Fri Sep 25 16:33:31 2026] 127.0.0.1:42562 Closing
[Fri Sep 25 16:33:31 2026] 127.0.0.1:42578 Accepted
[Fri Sep 25 16:33:31 2026] 127.0.0.1:42578 Closing
[Fri Sep 25 16:33:31 2026] 127.0.0.1:42586 Accepted
[Fri Sep 25 16:33:32 2026] 127.0.0.1:42586 Closing
[Fri Sep 25 16:51:39 2026] 127.0.0.1:54388 Accepted
[Fri Sep 25 16:51:40 2026] 127.0.0.1:54388 Closing
[Fri Sep 25 16:51:40 2026] 127.0.0.1:54404 Accepted
[Fri Sep 25 16:51:40 2026] 127.0.0.1:54404 Closing
[Fri Sep 25 16:51:40 2026] 127.0.0.1:54406 Accepted
[Fri Sep 25 16:51:40 2026] 127.0.0.1:54406 Closing
[Fri Sep 25 16:51:40 2026] 127.0.0.1:54420 Accepted
[Fri Sep 25 16:51:40 2026] 127.0.0.1:54420 Closing
[Fri Sep 25 16:51:40 2026] 127.0.0.1:54426 Accepted
[Fri Sep 25 16:51:40 2026] 127.0.0.1:54426 Closing
[Fri Sep 25 16:51:40 2026] 127.0.0.1:54430 Accepted
[Fri Sep 25 16:51:40 2026] 127.0.0.1:54430 Closing
[Fri Sep 25 16:51:40 2026] 127.0.0.1:54442 Accepted
[Fri Sep 25 16:51:40 2026] 127.0.0.1:54442 Closing
[Fri Sep 25 16:51:40 2026] 127.0.0.1:54452 Accepted
[Fri Sep 25 16:51:41 2026] 127.0.0.1:54452 Closing
[Fri Sep 25 16:51:41 2026] 127.0.0.1:54466 Accepted
[Fri Sep 25 16:51:41 2026] 127.0.0.1:54466 Closing
[Fri Sep 25 16:51:41 2026] 127.0.0.1:54476 Accepted
[Fri Sep 25 16:51:41 2026] 127.0.0.1:54476 Closing
[Fri Sep 25 16:51:41 2026] 127.0.0.1:54478 Accepted
[Fri Sep 25 16:51:41 2026] 127.0.0.1:54478 Closing
[Fri Sep 25 16:51:41 2026] 127.0.0.1:54494 Accepted
[Fri Sep 25 16:51:41 2026] 127.0.0.1:54494 Closing
[Fri Sep 25 16:51:41 2026] 127.0.0.1:54502 Accepted
[Fri Sep 25 16:51:41 2026] 127.0.0.1:54502 Closing
[Fri Sep 25 16:51:41 2026] 127.0.0.1:54512 Accepted
[Fri Sep 25 16:51:41 2026] 127.0.0.1:54512 Closing
[Fri Sep 25 16:51:41 2026] 127.0.0.1:54516 Accepted
[Fri Sep 25 16:51:41 2026] 127.0.0.1:54516 Closing
[Fri Sep 25 16:53:33 2026] 127.0.0.1:51594 Accepted
[Fri Sep 25 16:53:33 2026] 127.0.0.1:51594 Closing
[Fri Sep 25 16:53:34 2026] 127.0.0.1:51596 Accepted
[Fri Sep 25 16:53:34 2026] 127.0.0.1:51596 Closing
[Fri Sep 25 16:59:30 2026] 127.0.0.1:44096 Accepted
[Fri Sep 25 16:59:30 2026] 127.0.0.1:44096 Closing
[Fri Sep 25 16:59:30 2026] 127.0.0.1:44108 Accepted
[Fri Sep 25 16:59:30 2026] 127.0.0.1:44108 Closing
[Fri Sep 25 16:59:30 2026] 127.0.0.1:44110 Accepted
[Fri Sep 25 16:59:30 2026] 127.0.0.1:44110 Closing
[Fri Sep 25 16:59:30 2026] 127.0.0.1:44126 Accepted
[Fri Sep 25 16:59:30 2026] 127.0.0.1:44126 Closing
[Fri Sep 25 16:59:30 2026] 127.0.0.1:44142 Accepted
[Fri Sep 25 16:59:30 2026] 127.0.0.1:44142 Closing
[Fri Sep 25 16:59:30 2026] 127.0.0.1:44144 Accepted
[Fri Sep 25 16:59:30 2026] 127.0.0.1:44144 Closing
[Fri Sep 25 16:59:30 2026] 127.0.0.1:44160 Accepted
[Fri Sep 25 16:59:30 2026] 127.0.0.1:44160 Closing
[Fri Sep 25 17:03:27 2026] 127.0.0.1:48316 Accepted
[Fri Sep 25 17:03:27 2026] 127.0.0.1:48316 Closing
[Fri Sep 25 17:05:53 2026] 127.0.0.1:42542 Accepted