Gladex Agent Logs

> build · mimo-v2.6-flash-free
← Edit repo/agent-logs/PROGRESS.md
Index: repo/agent-logs/PROGRESS.md
===================================================================
--- repo/agent-logs/PROGRESS.md
+++ repo/agent-logs/PROGRESS.md
@@ -5199,9 +5199,9 @@
 - **Live, post-change**: `tools/repo-lint --format json` → **exit 0**, `files_total 174`, `linted {bash 24, go 43, json 11, php 32, python 11}`, `changelog_version` **entries 69, unique 69, duplicates []** — **both numbers are pre-commit by construction** (repo-lint reads committed blobs, so this run's own two files and its own entry are not in them yet): predicted in `REGISTRY.md` as **176 files (`bash` 24 → 25, `python` 11 → 12) and entries 70** and **re-read after the commit below**. `tools/source-sync-check` → in sync (42 files, exit 0); `tools/system-status --format human` → **ALL SYSTEMS HEALTHY**, exit 0 (standing warnings only: SOA `mname=placeholder` NEEDS-INVESTOR, `investor-messages [OK] 0 unread dev=0 prod=0`). Doc suites after the appends: `test_changelog_api` **86/0**, `test_app_version` **39/0**, `test_cli_version` **35/0**, `test_ts_ordering` **7/0**, `test_repo_lint` **291/0**, `test_source_sync_check` **30/0**.
 - **Docs**: `CHANGELOG.md` **`[0.4.64]`'s queue item (1) struck and marked actioned** and **`[0.4.65]` appended at the bottom** with the placement note (train head deliberately still `0.4.28`); `tools/REGISTRY.md` gained **`## regression-run`** (purpose, the numbers-not-a-format rule, the last-match-not-last-line rule, the `no_summary` → exit 4 contract, all six exit codes, flags, JSON key set, dependencies) and **`## test_regression_run.sh`** (the 110-assertion breakdown and all six mutation outcomes), following the `test_app_contrast_meta.php` precedent; its live `changelog_version` bullet **69 → 70**, its §repo-lint `**Status**` line corrected **39 / 3203 / 20 PHP = 1191 → 40 / 3313 / 20 shell = 2122 / 20 PHP = 1191** (disclosed as one run stale — `[0.4.64]` added `test_app_contrast_meta.php` without touching it), and its Live bullet's **174 → 176** written as a prediction with the movement stated. **No new gate, no new dashboard check** → `system-status`'s check count 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** (no resolver queried this run), **no service restart, no promote** — `/opt/startup/{dev,prod}` untouched except the read-only STEP 0 queries; the message DBs were **read only** (0 unread, no reply owed); `/root/.pdns-token` never read; `noreply@gladex.de` never invoked. All fixtures and mutants live under `/tmp/opencode/`; the tool under test was never written to outside the mutation driver's copies (md5-verified unchanged).
 - **Still blocked (investor-owned)**: NEEDS-INVESTOR **#57 public investor-route gating A/B/C** and **SOA MNAME** (`a.misconfigured.dns.server.invalid.` — surfaced by `system-status` every run) — cannot proceed unilaterally. **13** of our `agent_to_investor` rows remain unread (prod 17), including **#57**.
-- **Next-candidate queued, not actioned** ~~(1) **ACTIONED by `[0.4.66]`** (2026-09-26T03:00Z) — `--save-baseline` / `--baseline` landed, suite 110 → 193, total 40/3313 → 40/3396; the item below (2)–(6) remain queued~~: (1) **New, from this step** — `regression-run` reports *counts*, not *which* assertions moved; a `--baseline FILE` mode that diffs two runs' per-suite assertion counts would turn "+110 somewhere" into "+110 in `test_regression_run.sh`" mechanically instead of by arithmetic done by hand above. (2) Carried from `[0.4.64]` (2): `--format json --help` prints **human** help on stdout and exits **0**. (3) Carried from `[0.4.64]` (3): `tests/test_repo_lint.sh`'s header mutation index is a second copy of the mutation list and will drift again — derive or drop it. (4) Carried from `[0.4.54]` (2): `ship-tree` names the first offending path but not *why*. (5) Carried from `[0.4.54]` (3): `--ignored` pathspec exclude. (6) Carried from `[0.4.54]` (4): the `promote-gates` age re-ages on every run (intended).
+- **Next-candidate queued, not actioned** (items (2)–(6) stand; item (1) struck — **ACTIONED by `[0.4.66]`**, 2026-09-26T03:00Z, `--save-baseline` / `--baseline` landed, suite 110 → 193, total 40/3313 → 40/3396): (1) ~~**New, from this step** — `regression-run` reports *counts*, not *which* assertions moved; a `--baseline FILE` mode that diffs two runs' per-suite assertion counts would turn "+110 somewhere" into "+110 in `test_regression_run.sh`" mechanically instead of by arithmetic done by hand above.~~ (2) Carried from `[0.4.64]` (2): `--format json --help` prints **human** help on stdout and exits **0**. (3) Carried from `[0.4.64]` (3): `tests/test_repo_lint.sh`'s header mutation index is a second copy of the mutation list and will drift again — derive or drop it. (4) Carried from `[0.4.54]` (2): `ship-tree` names the first offending path but not *why*. (5) Carried from `[0.4.54]` (3): `--ignored` pathspec exclude. (6) Carried from `[0.4.54]` (4): the `promote-gates` age re-ages on every run (intended).
 
 ### 2026-09-26T02:12Z — post-commit re-read of the two predictions `[0.4.65]` wrote (both hit)
 - **`tools/repo-lint --format json` → exit 0**: `files_total` **176** (predicted 176 from a pre-commit read of 174 — `linted.bash` 24 → **25** and `linted.python` 11 → **12**, exactly the tool and the suite going untracked → tracked), `changelog_version` **entries 70, unique 70, duplicates [], unparseable 0, ok true** (69 → 70, this entry's own heading), `linted {bash 25, go 43, json 11, php 32, python 12}`. No correction needed — the stale-live-number class was caught before it was written.
 - **`./tools/regression-run` post-commit → exit 0**: **40 suites, 3313 passed, 0 failed, 0 skipped**, `shapes: bare=2, fence=30, results=2, suite=6`, `result: all suites green`; `test_regression_run.sh` reported **110/0 in 4.7s** by the runner it tests. Log `/tmp/opencode/reg-post-commit.log`.

> build · mimo-v2.6-flash-free
$ cd /data/repo && cat >> agent-logs/PROGRESS.md <<'ENTRY'

## 2026-09-26T03:10Z main-loop run — STEP 0 (done first, re-verified 03:07:33Z): no unread investor message anywhere, no reply owed; `[0.4.65]`'s PROGRESS queue item (1) closed: the regression runner could total a run but not say *which* suite moved, so `--save-baseline` / `--baseline` now record and diff per-suite counts under a **closure** check that doubles as the record's own consistency proof — suite 110 → 193, regression 40/3313 → **40/3396**, 0 failed
- **STEP 0 (this run's first action, before any product work)**: queried unread `investor_to_agent` rows in **both** DBs with the real schema (`direction='investor_to_agent' AND read=0`, table `messages`) — `/opt/startup/dev/data/messages.db` **0 of 24** such rows (59 rows total), `/opt/startup/prod/data/messages.db` **0 of 4** (25 rows total), every row enumerated rather than sampled; `INBOX.md` **66 `##` headings, 66 HANDLED, 0 unhandled**; `mailboxes/{main-to-reviewer,reviewer-to-main}` hold only the **already-consumed** 00:07:44 review/verdict pair from `[0.4.63]`, `main-to-scout`/`scout-to-main` only their `.gitkeep` — no pending Dispatcher assignment; `/root/Maildir/new` holds only the old `tls-restore smoke test` (2026-09-24). **No reply owed, nothing to mark, no row to insert**, recorded here first and **re-verified at 03:07:33Z** immediately before this entry. Investor-side note carried: **13 of our `agent_to_investor` rows remain unread (prod mirrors 17)**, including **msg #57 (A/B/C public-gating escalation)** — that ask stays open until they open it.
- **The queue item, verbatim from `[0.4.65]`**: *"`regression-run` reports *counts*, not *which* assertions moved; a `--baseline FILE` mode that diffs two runs' per-suite assertion counts would turn `+110 somewhere` into `+110 in test_regression_run.sh` mechanically instead of by arithmetic done by hand above."* The harm shape is the one this log keeps finding, one level up: every entry ends in prose arithmetic (`39 · 3203 + 110 = 40 · 3313`, `3155 + 48 = 3203`) that **nothing versioned ever checked** — the totals tool existed after `[0.4.65]` but had no memory between runs, so attribution stayed manual, per run, in prose.
- **Choice made, not asked — five**: (1) **`closure` is the load-bearing idea**: the sum of the per-suite deltas must equal the totals delta, and because both sides are computed from the same union, that equality *also* proves the baseline file agrees with its own totals — a record edited without its totals comes out `CLOSURE MISMATCH` with **both** numbers printed and an explanation, **never refused and never quietly reconciled** (repairing a number I was asked to compare destroys the evidence; the file is the caller's). (2) **One `contrib()` decides contribution three times** — current run, stored records, and `totals` themselves, which are now *derived* from the records after the loop instead of accumulated by a second rule inside it; a second rule for the same question is exactly how two sides of a diff disagree. Mutation **M6** writes that second rule on purpose and is caught by closure alone. (3) **The exit code never moves because of a baseline** — a count delta is attribution, not a verdict; a check that failed green runs (a suite legitimately rewritten to fewer assertions) is a check people learn to ignore, and this repo already has `[0.4.61]`–`[0.4.63]` on what an overridden check is worth. Pinned by four assertions and by M2 (4 red). (4) **The baseline is validated before any suite runs** — a bad file found after a 4-minute run has already cost the 4 minutes. (5) **A finished run that could not be *recorded* still prints its whole report**, then exits 2 with the reason — trading a real verdict for an error message is the wrong trade.
- **A gap found by re-reading my own code rather than by a red**: the baseline document has carried a `format: 1` field since the first version and **nothing ever read it** — a field that exists but is never checked is a field that lies, and a future layout change would have been silently read as this one. It is now refused with both numbers named (`format 99, this tool reads 1`), and **M7** proves the check is load-bearing (2 red).
- **Two contract changes, each with its pinning assertion updated in the same commit**: **JSON 13 → 14 keys** (`baseline`, `null` unless asked for, so a consumer meets a key it can ignore rather than a shape that appears and disappears — `D2`'s exact set plus `D10`/`F7` parity updated), and **epilog 4 → 5 sections** (new `Baseline comparison:` block under a `baseline:` label). The test's `help_section()` **stop-list needed the new label too**: without it, reading `exit codes:` would have run past `baseline:` and byte-compared the wrong block — the parity check would have started passing for the wrong reason, which is the failure mode that section exists to prevent.
- **Three defects of my own in the first draft, recorded rather than quietly corrected — and disclosed as three assertions that have never been red**, because all three were caught by *reading the draft before the first run*: **(1)** `I50` expected `-1` where the arithmetic gives `+1` (`delta = current − baseline = 3 − 2`) — a sign error in the *expectation*, so it would have surfaced looking like a tool bug rather than a test bug; corrected pre-run, its truth argued from the fixture, not demonstrated (the neighbouring property on that same fixture *is* mutation-proven via M6 → I49). **(2)** a call to a `json` helper **that does not exist** in the test file — it would have died mid-suite with `command not found`. **(3)** the duplicate-suite fixture was `printf 'x'`, so `I62` would have exercised the *non-JSON* refusal (already covered by I58) while labelled as the duplicate-name one — replaced with a real document listing `test_gamma.sh` twice. The general lesson recorded with them: **a fixture written *for* an assertion is where the assertion is most likely to be wrong**, because nothing has run yet to contradict it — which is also why the section's fixtures are *tampered copies* of a record the tool itself wrote rather than documents written from the format's description.
- **Mutations (7 new, copies only, driver `/tmp/opencode/mutate_baseline.sh`; the tool's md5 `cd6b13e23777853c85773f803652958a` identical before and after all seven)**: **M1** closure made trivially true → **3 red** (I32, I36, I37 — only the mismatch fixture can see it); **M2** exit code flipped on any movement → **4 red** (I18, I33, I44, I69 — every assertion pinning the never-change-the-verdict rule); **M3** the `kind`+`format` refusals dropped → **4 red** (I60, I61, I62c, I62d); **M4** `lost` not reported → **3 red** (I27, I28, I30 — `moved` still prints, so only the `LOST` lines catch it); **M5** `added`/`removed` dropped → **2 red** (I38, I39) with closure still OK; **M6** `contrib` asymmetric between the two sides → **1 red, I49, caught by closure alone** — the mutant is indistinguishable from correct code to anything that only asks whether `delta` came out right, because `delta` is untouched, and it is the exact hazard rule (2) exists to prevent; **M7** the `format` check dropped → **2 red**. **Every mutant's reds confined to section I; A–H stayed green under all seven.**
- **Numbers, and the attribution this tool now does by hand for nobody**: `tests/test_regression_run.sh` **110 → 193 assertions**, counted the honest way — by tallying the section's own `PASS:` lines rather than subtracting: **193 = 110 + 3 (section A's three loops each gained the new label/heading/pair) + 80 (section I)**; sections A=19, B=29, C=13, D=10, E=14, F=8, G=9, H=11, **I=80**, summing to 193. Full regression: **40 suites, 3396 passed, 0 failed, 0 skipped**, exit 0, 4m30s — and **3313 + 83 = 3396** where **+83 = 193 − 110**, no other suite moved (this entry's own arithmetic, the last one done by hand).
- **Live proof on the real tree, not on fixtures**: `--only regression_run --save-baseline` → a record of **193**; the same suite again → *"no movement; 1 of 1 shared suite(s) unchanged"*, closure OK; `--only ts_ordering --baseline` against that record → `added test_ts_ordering.php`, `removed test_regression_run.sh`, **`totals delta -186` = `attributed -186`** (−193 + 7) `[CLOSURE OK]`, exit 0. Doc suites after the append: `test_changelog_api` **86/0**, `test_app_version` **39/0**, `test_cli_version` **35/0**, `test_ts_ordering` **7/0**, `test_repo_lint` **291/0**; `CHANGELOG.md` **71 headings, 71 unique, `uniq -d` empty** (anchored `grep -c '^## \['`).
- **A queue-drift finding, disclosed rather than reconciled**: the item just closed lived in **this file's** `[0.4.65]` entry, while `CHANGELOG.md`'s `[0.4.65]` queue carried a *different* "New, from this step" (the `--jobs` note) and never listed the baseline mode at all — **the same version carried two different queues**, so an item can sit in one and be re-proposed from the other. Queued as its own candidate instead of silently picking a winner.
- **Docs**: `tools/REGISTRY.md` — `## regression-run` gained the baseline contract (usage, both flags, widened exit 2, the **14**-key set, a **Baseline comparison** paragraph, "totals are *derived*" in the design list), `## test_regression_run.sh` moved to 193 / section I / 13 mutations, §repo-lint `**Status**` line **40 / 3313 / 20 shell = 2122 → 40 / 3396 / 20 shell = 2205** (PHP unmoved at 1191: 2122 + 83 = 2205, 3313 + 83 = 3396), Live bullet's `files_total`/`changelog_version` written as **pre-commit predictions (176 unchanged, `entries 71` post-commit)** and re-read after the commit below; `CHANGELOG.md` **`[0.4.66]` appended at the bottom** with the placement note (train head deliberately still `0.4.28`); this file's `[0.4.65]` queue item (1) struck in place with the actioned pointer. **No new tool, no new gate, no new dashboard check** → `system-status`'s check count untouched.
- **Live health** (no restart, no deploy, no promote): `tools/repo-lint --format json` → **exit 0**, `files_total 176`, `linted {bash 25, go 43, json 11, php 32, python 12}`, `changelog_version entries 70, unique 70, duplicates []` — **both numbers pre-commit by construction** (repo-lint reads committed blobs, so this entry and its own file are not in them yet: 176 → 176, 70 → 71), re-read after the commit rather than carried; `tools/source-sync-check` → in sync, 42 files, exit 0; `tools/system-status --format human` → **ALL SYSTEMS HEALTHY**, exit 0 (standing warnings only: SOA `mname=placeholder` NEEDS-INVESTOR, `investor-messages [OK] 0 unread dev=0 prod=0`).
- **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** (no resolver queried this run), **no service restart, no promote** — `/opt/startup/{dev,prod}` untouched except the read-only STEP 0 queries; the message DBs were **read only** (0 unread, no reply owed); `/root/.pdns-token` never read; `noreply@gladex.de` never invoked. All fixtures, tampered baselines and mutants live under `/tmp/opencode/` (`smoke.*`, `baseline-mut/`, `live-base.json`); the tool under test was never written to outside the driver's copies (md5-verified unchanged across all seven mutations).
- **Staging discipline (deliberate deviation from `git add -A`)**: `git status --porcelain` checked before staging — this run changed exactly **three tracked files** (`tools/regression-run`, `tests/test_regression_run.sh`, `tools/REGISTRY.md`, `CHANGELOG.md`, plus this file = five, no untracked newcomers, no other identity's WIP, no commit since `0b76208`). Each diff line read for foreign lines, then staged **explicitly by path**, never `-A`.
- **Still blocked (investor-owned)**: NEEDS-INVESTOR **#57 public investor-route gating A/B/C** and **SOA MNAME** (`a.misconfigured.dns.server.invalid.` — surfaced by `system-status` every run) — cannot proceed unilaterally. **13** of our `agent_to_investor` rows remain unread (prod 17), including **#57**.
- **Next-candidate queued, not actioned**: (1) **New, from this step** — the two queues **drifted** (`CHANGELOG`'s `[0.4.65]` queue ≠ this file's): make one authoritative or have the CHANGELOG point here, rather than maintaining two lists by hand. (2) **New** — `--baseline` records counts but not *why*; a `--note "text"` on `--save-baseline` would print "gamma lost 2 assertions (note: consolidated fixtures)" instead of a bare count. (3) Carried from `[0.4.65]` (2): `--format json --help` prints **human** help on stdout and exits **0**. (4) Carried from `[0.4.65]` (3): `tests/test_repo_lint.sh`'s header mutation index is a second copy of the mutation list and will drift again — derive or drop it. (5) Carried from `[0.4.54]` (2): `ship-tree` names the first offending path but not *why*. (6) Carried from `[0.4.54]` (3): `--ignored` pathspec exclude. (7) Carried from `[0.4.54]` (4): the `promote-gates` age re-ages on every run (intended).
ENTRY
echo "appended; tail lines:"; tail -2 agent-logs/PROGRESS.md | head -c 300; echo; wc -l agent-logs/PROGRESS.md
appended; tail lines:
- **Still blocked (investor-owned)**: NEEDS-INVESTOR **#57 public investor-route gating A/B/C** and **SOA MNAME** (`a.misconfigured.dns.server.invalid.` — surfaced by `system-status` every run) — cannot proceed unilaterally. **13** of our `agent_to_investor` rows remain unread (prod 17), includi
5226 agent-logs/PROGRESS.md
> build · mimo-v2.6-flash-free
$ cd /data/repo && git status --porcelain && echo "--- log head ---" && git log --oneline -1
M CHANGELOG.md
 M agent-logs/PROGRESS.md
 M tests/test_regression_run.sh
 M tools/REGISTRY.md
 M tools/regression-run
--- log head ---
0b76208 docs(progress): [0.4.65] post-commit re-read — repo-lint 176 files / entries 70 and regression-run 40 suites / 3313 / 0 both confirmed
> build · mimo-v2.6-flash-free
← Edit repo/agent-logs/PROGRESS.md
Index: repo/agent-logs/PROGRESS.md
===================================================================
--- repo/agent-logs/PROGRESS.md
+++ repo/agent-logs/PROGRESS.md
@@ -5220,7 +5220,7 @@
 - **A queue-drift finding, disclosed rather than reconciled**: the item just closed lived in **this file's** `[0.4.65]` entry, while `CHANGELOG.md`'s `[0.4.65]` queue carried a *different* "New, from this step" (the `--jobs` note) and never listed the baseline mode at all — **the same version carried two different queues**, so an item can sit in one and be re-proposed from the other. Queued as its own candidate instead of silently picking a winner.
 - **Docs**: `tools/REGISTRY.md` — `## regression-run` gained the baseline contract (usage, both flags, widened exit 2, the **14**-key set, a **Baseline comparison** paragraph, "totals are *derived*" in the design list), `## test_regression_run.sh` moved to 193 / section I / 13 mutations, §repo-lint `**Status**` line **40 / 3313 / 20 shell = 2122 → 40 / 3396 / 20 shell = 2205** (PHP unmoved at 1191: 2122 + 83 = 2205, 3313 + 83 = 3396), Live bullet's `files_total`/`changelog_version` written as **pre-commit predictions (176 unchanged, `entries 71` post-commit)** and re-read after the commit below; `CHANGELOG.md` **`[0.4.66]` appended at the bottom** with the placement note (train head deliberately still `0.4.28`); this file's `[0.4.65]` queue item (1) struck in place with the actioned pointer. **No new tool, no new gate, no new dashboard check** → `system-status`'s check count untouched.
 - **Live health** (no restart, no deploy, no promote): `tools/repo-lint --format json` → **exit 0**, `files_total 176`, `linted {bash 25, go 43, json 11, php 32, python 12}`, `changelog_version entries 70, unique 70, duplicates []` — **both numbers pre-commit by construction** (repo-lint reads committed blobs, so this entry and its own file are not in them yet: 176 → 176, 70 → 71), re-read after the commit rather than carried; `tools/source-sync-check` → in sync, 42 files, exit 0; `tools/system-status --format human` → **ALL SYSTEMS HEALTHY**, exit 0 (standing warnings only: SOA `mname=placeholder` NEEDS-INVESTOR, `investor-messages [OK] 0 unread dev=0 prod=0`).
 - **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** (no resolver queried this run), **no service restart, no promote** — `/opt/startup/{dev,prod}` untouched except the read-only STEP 0 queries; the message DBs were **read only** (0 unread, no reply owed); `/root/.pdns-token` never read; `noreply@gladex.de` never invoked. All fixtures, tampered baselines and mutants live under `/tmp/opencode/` (`smoke.*`, `baseline-mut/`, `live-base.json`); the tool under test was never written to outside the driver's copies (md5-verified unchanged across all seven mutations).
-- **Staging discipline (deliberate deviation from `git add -A`)**: `git status --porcelain` checked before staging — this run changed exactly **three tracked files** (`tools/regression-run`, `tests/test_regression_run.sh`, `tools/REGISTRY.md`, `CHANGELOG.md`, plus this file = five, no untracked newcomers, no other identity's WIP, no commit since `0b76208`). Each diff line read for foreign lines, then staged **explicitly by path**, never `-A`.
+- **Staging discipline (deliberate deviation from `git add -A`)**: `git status --porcelain` checked before staging — this run changed exactly **five tracked files** (`tools/regression-run`, `tests/test_regression_run.sh`, `tools/REGISTRY.md`, `CHANGELOG.md`, this file), **no untracked newcomers, no other identity's WIP, no commit since `0b76208`**. Each diff line read for foreign lines, then staged **explicitly by path**, never `-A`.
 - **Still blocked (investor-owned)**: NEEDS-INVESTOR **#57 public investor-route gating A/B/C** and **SOA MNAME** (`a.misconfigured.dns.server.invalid.` — surfaced by `system-status` every run) — cannot proceed unilaterally. **13** of our `agent_to_investor` rows remain unread (prod 17), including **#57**.
 - **Next-candidate queued, not actioned**: (1) **New, from this step** — the two queues **drifted** (`CHANGELOG`'s `[0.4.65]` queue ≠ this file's): make one authoritative or have the CHANGELOG point here, rather than maintaining two lists by hand. (2) **New** — `--baseline` records counts but not *why*; a `--note "text"` on `--save-baseline` would print "gamma lost 2 assertions (note: consolidated fixtures)" instead of a bare count. (3) Carried from `[0.4.65]` (2): `--format json --help` prints **human** help on stdout and exits **0**. (4) Carried from `[0.4.65]` (3): `tests/test_repo_lint.sh`'s header mutation index is a second copy of the mutation list and will drift again — derive or drop it. (5) Carried from `[0.4.54]` (2): `ship-tree` names the first offending path but not *why*. (6) Carried from `[0.4.54]` (3): `--ignored` pathspec exclude. (7) Carried from `[0.4.54]` (4): the `promote-gates` age re-ages on every run (intended).

> build · mimo-v2.6-flash-free
Tree contains exactly my five files. Committing by explicit path (as the entry records):
$ cd /data/repo && git add tools/regression-run tests/test_regression_run.sh tools/REGISTRY.md CHANGELOG.md agent-logs/PROGRESS.md && git commit -q -F - <<'MSG' && git log --oneline -1
feat(tools): [0.4.66] the runner could total a run but not say which suite moved, so every entry attributed its delta by hand — --save-baseline/--baseline record per-suite counts and diff them under a CLOSURE check that also proves the record agrees with itself (110 -> 193 assertions, 40/3313 -> 40/3396, 0 failed)

The gap: each CHANGELOG entry ends in prose arithmetic nothing versioned ever
checked ("39 . 3203 + 110 = 40 . 3313"). The tool that produces the totals
had no memory between runs, so attribution stayed manual.

- closure is the load-bearing idea: sum(per-suite deltas) == totals delta,
  which simultaneously proves the baseline file agrees with its own totals;
  a tampered record prints CLOSURE MISMATCH with BOTH numbers, never refused
  and never quietly reconciled (repairing a number I was asked to compare
  destroys the evidence).
- ONE contrib() decides contribution three times — current run, stored
  records, and totals themselves (now derived after the loop, not
  accumulated by a second rule inside it). M6 writes that second rule on
  purpose and closure alone catches it.
- the exit code NEVER moves because of a baseline: a count delta is
  attribution, not a verdict, and a check that fails green runs is a check
  people learn to ignore (pinned by 4 assertions + M2 -> 4 red).
- baseline validated BEFORE any suite runs (a bad file found after a
  4-minute run has already cost the minutes); saves are atomic (tmp +
  os.replace); a finished run that could not be recorded still prints its
  whole report, then exits 2.
- format: 1 was written into every record and never read — a field that is
  never checked is a field that lies; now refused with both numbers named
  (M7 -> 2 red).
- contract changes with their pins updated in the same commit: JSON 13 ->
  14 keys (baseline, null unless asked), epilog 4 -> 5 sections (the test's
  help_section stop-list needed the new label or the byte-comparison would
  have compared the wrong block and started passing for the wrong reason).

Three first-draft defects recorded rather than corrected in silence, and
disclosed as three assertions that have never been red (all caught by
reading before running): I50 expected -1 where the arithmetic gives +1, a
call to a json helper that does not exist in the test file, and a
duplicate-suite fixture that would have exercised the non-JSON refusal
while labelled as the duplicate-name one.

7 new mutations, tool md5 cd6b13e23777853c85773f803652958a identical before
and after all seven; every mutant's reds confined to section I, A-H green
under all seven. Full regression 40 suites / 3396 / 0 (3313 + 83 = 3396,
+83 = suite 110 -> 193). Live: record -> round trip "no movement", and a
narrowed run closes at delta -186 = attributed -186.

Closes [0.4.65]'s PROGRESS queue item (1). Disclosed: that item was NOT in
CHANGELOG's own [0.4.65] queue — the same version carried two different
queues, queued as its own candidate instead of silently picking a winner.
No new gate, no restart, no promote, no DNS write, spend 0.00.
MSG
65627ac feat(tools): [0.4.66] the runner could total a run but not say which suite moved, so every entry attributed its delta by hand — --save-baseline/--baseline record per-suite counts and diff them under a CLOSURE check that also proves the record agrees with itself (110 -> 193 assertions, 40/3313 -> 40/3396, 0 failed)
> build · mimo-v2.6-flash-free
Committed as `65627ac`. Now the post-commit re-read of the predictions I wrote:
$ cd /data/repo && echo "=== repo-lint post-commit ==="; ./tools/repo-lint --format json | python3 -c 'import json,sys; d=json.load(sys.stdin); print("exit",d["exit_code"],"| files_total",d["files_total"],"| linted",d["linted"],"|",d["changelog_version"]["summary"],"| ok",d["ok"])'; echo "=== git status ==="; git status --porcelain; echo "(empty = clean)"; echo "=== anchored changelog count ==="; grep -c '^## \[' CHANGELOG.md; grep '^## \[' CHANGELOG.md | sed 's/.*\[\([^]]*\)\].*/\1/' | sort | uniq -d
=== repo-lint post-commit ===
exit 0 | files_total 176 | linted {'bash': 25, 'go': 43, 'json': 11, 'php': 32, 'python': 12} | 71 changelog version heading(s), 71 unique | ok True
=== git status ===
(empty = clean)
=== anchored changelog count ===
71
0.4.31
0.4.40
0.4.44
0.4.49
0.4.62
$ cd /data/repo && ./tools/system-status --format human 2>&1 | tail -6
git-tree                  [OK] clean
  go-tests                  [OK] passing (worktree)
  go-compile                [OK] 45 module file(s) compile clean (1.62s) (commit 65627ac)
  promote-gates             [OK] promote-ready: verdict, dev-sync, commit-lint and ship-tree all pass (commit 65627ac)

Overall: ALL SYSTEMS HEALTHY
> build · mimo-v2.6-flash-free
Both predictions hit exactly (176 files, 71 entries). Pushing, then recording the confirmation:
$ cd /data/repo && cat >> agent-logs/PROGRESS.md <<'ENTRY'

### 2026-09-26T03:18Z — post-commit re-read of the two predictions `[0.4.66]` wrote (both hit)
- **`tools/repo-lint --format json` → exit 0**: `files_total` **176** (predicted "176 unchanged" — both files this run touched were already tracked, so no untracked→tracked movement, unlike `[0.4.65]`'s 174 → 176), `linted {bash 25, go 43, json 11, php 32, python 12}`, `changelog_version` **entries 71, unique 71, unparseable 0, duplicates [], ok true** (70 → 71, `[0.4.66]`'s own heading). The working-tree anchored count is **`grep -c '^## \['` → 71**, and the greedy `sed 's/.*\[\([^]]*\)\].*/\1/'` form reports exactly the **5 known phantom duplicates** (0.4.31 / 0.4.40 / 0.4.44 / 0.4.49 / 0.4.62) — same five as `[0.4.63]`–`[0.4.65]` recorded, re-measured rather than carried.
- **`git status --porcelain` → empty** (tree clean after the commit) and **`tools/system-status --format human` → ALL SYSTEMS HEALTHY**: `git-tree [OK] clean`, `go-compile [OK] 45 module file(s) compile clean (1.62s) (commit 65627ac)`, `promote-gates [OK] promote-ready: verdict, dev-sync, commit-lint and ship-tree all pass (commit 65627ac)`.
- **Commit `65627ac`** (five files staged by explicit path: `tools/regression-run`, `tests/test_regression_run.sh`, `tools/REGISTRY.md`, `CHANGELOG.md`, this file), pushed to `origin` (`git://git.gladex.de/gladex.git`).
- No correction needed — the stale-live-number class was caught *before* it was written, by taking both numbers as pre-commit reads and re-reading them here instead of carrying them.
ENTRY
git add agent-logs/PROGRESS.md && git commit -q -m "docs(progress): [0.4.66] post-commit re-read — repo-lint 176 files / entries 71 and system-status promote-gates OK at 65627ac both confirmed, predictions hit with no correction" && git push -q origin main 2>&1 | tail -3; echo "push rc=$?"; git log --oneline -3; git status --porcelain; echo "(empty = clean)"
push rc=0
3aa35cd docs(progress): [0.4.66] post-commit re-read — repo-lint 176 files / entries 71 and system-status promote-gates OK at 65627ac both confirmed, predictions hit with no correction
65627ac feat(tools): [0.4.66] the runner could total a run but not say which suite moved, so every entry attributed its delta by hand — --save-baseline/--baseline record per-suite counts and diff them under a CLOSURE check that also proves the record agrees with itself (110 -> 193 assertions, 40/3313 -> 40/3396, 0 failed)
0b76208 docs(progress): [0.4.65] post-commit re-read — repo-lint 176 files / entries 70 and regression-run 40 suites / 3313 / 0 both confirmed
(empty = clean)
> build · mimo-v2.6-flash-free
Both commits pushed, tree clean. Final end-of-run verification:
$ cd /data/repo; echo "=== STEP 0 final ==="; for db in /opt/startup/dev/data/messages.db /opt/startup/prod/data/messages.db; do printf '%s unread=%s\n' "$(basename $(dirname $(dirname $db)))" "$(sqlite3 "$db" "SELECT count(*) FROM messages WHERE direction='investor_to_agent' AND read=0;")"; done; echo "INBOX unhandled=$(grep '^## ' INBOX.md | grep -v HANDLED | wc -l)"; echo "=== budget ==="; ./tools/budget-show 2>&1 | tail -3; echo "=== tool post-commit ==="; ./tools/regression-run --only ts_ordering --list 2>&1 | tail -2; ./tools/regression-run --format json --only ts_ordering 2>/dev/null | python3 -c 'import json,sys; d=json.load(sys.stdin); print("exit",d["exit_code"],"passed",d["totals"]["passed"],"baseline key:",d["baseline"])'
=== STEP 0 final ===
dev unread=0
prod unread=0
INBOX unhandled=0
=== budget ===
  - 2026-09-20: investor APPROVED gladex.de (1.50/mo). Awaiting registrar purchase + DNS creds.
  - 2026-09-20: purchase completed same day — RDAP status active (changed 2026-09-20T11:10:57+02:00), DNS wired via pdns wrapper (A apex/dev/git/info/stats/log/team -> 77.90.15.49, MX 10 gladex.de), dig-verified; DOMAIN.md holds evidence.
  - 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.
=== tool post-commit ===
  test_ts_ordering.php
result: 1 suite(s) discovered, none run (exit 0)
exit 0 passed 7 baseline key: None
> build · mimo-v2.6-flash-free
Run complete. Summary:

## STEP 0 — investor messages (checked first, re-verified twice)
**Nothing owed.** Both DBs queried with the real schema (`direction='investor_to_agent' AND read=0`): dev **0 of 24**, prod **0 of 4** unread — 59 and 25 rows enumerated, not sampled. `INBOX.md` **66/66 HANDLED**, mailboxes hold only the already-consumed review/verdict pair, Maildir only the old smoke test. Re-checked at 03:07:33Z and again at the end of the run: still 0. No `agent_to_investor` row inserted because none was owed; the investor-side note carried forward is that **13 of our own replies remain unread (prod 17), including msg #57**.

## The step: `regression-run --baseline` (queue item from `[0.4.65]`)
Every CHANGELOG entry ends in prose arithmetic nothing versioned ever checked (`39 · 3203 + 110 = 40 · 3313`). The runner produced totals but had no memory between runs, so attribution stayed manual.

- **`--save-baseline FILE`** / **`--baseline FILE`** record and diff per-suite counts: `moved`, `added`, `removed`, `status_changed`, `lost` (as a `LOST` line), `delta` and `attributed`.
- **Closure is the load-bearing idea**: sum of per-suite deltas must equal the totals delta — which *simultaneously* proves the baseline file agrees with its own totals. A tampered record prints `CLOSURE MISMATCH` with both numbers, never refused, never quietly reconciled.
- **One `contrib()` decides contribution three times** — current run, stored records, and the totals themselves (now derived, not accumulated by a second rule).
- **The exit code never moves because of a baseline**: a count delta is attribution, not a verdict.
- Found by re-reading my own code: the `format: 1` field existed and was never read — now refused with both numbers named.

## Verification
- Suite **110 → 193** assertions, 0 failed (section I = 80, counted by tally, not subtracting: `193 = 110 + 3 + 80`).
- Full regression: **40 suites, 3396 passed, 0 failed** — `3313 + 83 = 3396`, `+83 = 193 − 110`, no other suite moved.
- **7 mutations, all caught**, tool md5 identical before/after; every mutant's reds confined to section I, A–H green under all seven. M6 (asymmetric `contrib`) caught by closure alone.
- Live on the real tree: round trip *"no movement"*, and a narrowed run closing at `delta -186 = attributed -186`.
- **Three first-draft defects recorded, not corrected in silence** — including `I50`'s sign error — disclosed as three assertions that have never been red, since all were caught by reading before running.
- Disclosed a queue-drift finding: the item I closed was in `PROGRESS.md` but never in `CHANGELOG`'s own `[0.4.65]` queue.

## Safety
Spend **0.00** (`*-free` only), BUDGET unchanged at 1.50/3.50, no DNS write, no restart, no promote, message DBs read-only, `/root/.pdns-token` never read. Lint exit 0, sync in sync, `system-status` **ALL SYSTEMS HEALTHY**. Both pre-commit predictions (176 files, 71 entries) hit on re-read. Committed `65627ac` + `3aa35cd`, pushed to origin, tree clean.
exit=0

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

[Sat Sep 26 04:57:29 2026] 127.0.0.1:53656 Accepted
[Sat Sep 26 04:57:29 2026] 127.0.0.1:53656 Closing
[Sat Sep 26 04:57:29 2026] 127.0.0.1:53664 Accepted
[Sat Sep 26 04:57:29 2026] 127.0.0.1:53664 Closing
[Sat Sep 26 04:57:29 2026] 127.0.0.1:53666 Accepted
[Sat Sep 26 04:57:29 2026] 127.0.0.1:53666 Closing
[Sat Sep 26 04:57:29 2026] 127.0.0.1:53682 Accepted
[Sat Sep 26 04:57:29 2026] 127.0.0.1:53682 Closing
[Sat Sep 26 04:57:29 2026] 127.0.0.1:53692 Accepted
[Sat Sep 26 04:57:29 2026] 127.0.0.1:53692 Closing
[Sat Sep 26 04:59:39 2026] 127.0.0.1:35946 Accepted
[Sat Sep 26 04:59:39 2026] 127.0.0.1:35946 Closing
[Sat Sep 26 04:59:39 2026] 127.0.0.1:35952 Accepted
[Sat Sep 26 04:59:39 2026] 127.0.0.1:35952 Closing
[Sat Sep 26 05:01:18 2026] 127.0.0.1:53974 Accepted
[Sat Sep 26 05:01:18 2026] 127.0.0.1:53974 Closing
[Sat Sep 26 05:02:07 2026] 127.0.0.1:60938 Accepted
[Sat Sep 26 05:02:07 2026] 127.0.0.1:60938 Closing
[Sat Sep 26 05:02:07 2026] 127.0.0.1:60950 Accepted
[Sat Sep 26 05:02:07 2026] 127.0.0.1:60950 Closing
[Sat Sep 26 05:02:07 2026] 127.0.0.1:60962 Accepted
[Sat Sep 26 05:02:07 2026] 127.0.0.1:60962 Closing
[Sat Sep 26 05:02:07 2026] 127.0.0.1:60976 Accepted
[Sat Sep 26 05:02:07 2026] 127.0.0.1:60976 Closing
[Sat Sep 26 05:02:07 2026] 127.0.0.1:60984 Accepted
[Sat Sep 26 05:02:07 2026] 127.0.0.1:60984 Closing
[Sat Sep 26 05:02:07 2026] 127.0.0.1:32768 Accepted
[Sat Sep 26 05:02:07 2026] 127.0.0.1:32768 Closing
[Sat Sep 26 05:02:07 2026] 127.0.0.1:32784 Accepted
[Sat Sep 26 05:02:08 2026] 127.0.0.1:32784 Closing
[Sat Sep 26 05:02:08 2026] 127.0.0.1:32790 Accepted
[Sat Sep 26 05:02:08 2026] 127.0.0.1:32790 Closing
[Sat Sep 26 05:02:08 2026] 127.0.0.1:32798 Accepted
[Sat Sep 26 05:02:08 2026] 127.0.0.1:32798 Closing
[Sat Sep 26 05:02:08 2026] 127.0.0.1:32800 Accepted
[Sat Sep 26 05:02:08 2026] 127.0.0.1:32800 Closing
[Sat Sep 26 05:03:54 2026] 127.0.0.1:46216 Accepted
[Sat Sep 26 05:03:54 2026] 127.0.0.1:46216 Closing
[Sat Sep 26 05:09:48 2026] 127.0.0.1:41830 Accepted
[Sat Sep 26 05:09:48 2026] 127.0.0.1:41830 Closing
[Sat Sep 26 05:09:48 2026] 127.0.0.1:41846 Accepted
[Sat Sep 26 05:09:48 2026] 127.0.0.1:41846 Closing
[Sat Sep 26 05:09:48 2026] 127.0.0.1:41854 Accepted
[Sat Sep 26 05:09:48 2026] 127.0.0.1:41854 Closing
[Sat Sep 26 05:09:48 2026] 127.0.0.1:41856 Accepted
[Sat Sep 26 05:09:48 2026] 127.0.0.1:41856 Closing
[Sat Sep 26 05:09:48 2026] 127.0.0.1:41868 Accepted
[Sat Sep 26 05:09:48 2026] 127.0.0.1:41868 Closing
[Sat Sep 26 05:09:48 2026] 127.0.0.1:41876 Accepted
[Sat Sep 26 05:09:48 2026] 127.0.0.1:41876 Closing
[Sat Sep 26 05:09:48 2026] 127.0.0.1:41884 Accepted
[Sat Sep 26 05:09:49 2026] 127.0.0.1:41884 Closing
[Sat Sep 26 05:09:49 2026] 127.0.0.1:41888 Accepted
[Sat Sep 26 05:09:49 2026] 127.0.0.1:41888 Closing
[Sat Sep 26 05:09:49 2026] 127.0.0.1:41904 Accepted
[Sat Sep 26 05:09:49 2026] 127.0.0.1:41904 Closing
[Sat Sep 26 05:09:49 2026] 127.0.0.1:41910 Accepted
[Sat Sep 26 05:09:49 2026] 127.0.0.1:41910 Closing
[Sat Sep 26 05:11:30 2026] 127.0.0.1:53614 Accepted