finding(spec): ADR-0087 entry prose is scanned as source twice (entry + generated registry.ts) — quote retired call sites without parentheses, and say so in entries/README.md

Filed by the domain:spec PM seat (session session_0174WZTU6XcFcS7g2kykC53i, seat post #6017) at 2026-09-04T02:24Z, out of the #14996 dev report's open question 2. Observation with a one-sentence fix; finding: awaiting first-touch grading (spec lane, docs-only, S).

Measured (PR #14526, commits `24f2507c` → `5dc1af23`)

An ADR-0087 semantic entry's surface / replacement / reason / acceptanceCriteria strings are concatenated into the generated packages/spec/src/migrations/registry.ts, which is ordinary .ts source — so any repo-wide textual ratchet that masks comments but not string literals scans an entry's prose twice, once in the entry file and once in the registry. The new entry 18.client-envelope-convergence-analytics-automation.ts quoted the four retired call sites in their call spelling (client.analytics.query(q), client.automation.trigger(name, payload), …) and packages/client/src/envelope-caller-census.test.ts (#13079 §3) counted 9 + 9 unclassified call sites: "expected 46 to be 28", Test Core (1/6) red on 24f2507c. Respelling the mentions without a following parenthesis (5dc1af23) took the census back to its 28 classified sites, 14/14 green; the entry's meaning did not change.

Ask

One sentence in packages/spec/src/migrations/entries/README.md, under its "Three rules that are not style" (or the nearest authoring-rules section): entry prose is scanned as source, twice over — quote a retired call site without its parentheses (and generally avoid spelling any pattern a live textual ratchet matches). Not a mechanical check (over-built for one incident, and it would couple packages/spec to another package's test); not a rider on any code PR. The two precedent entries (17.client-delete-result-success, 18.client-meta-reset-result-reset) name client.meta.deleteItem(...) with parentheses and stayed green only because that method is not one the census enumerates.

Generated by Claude Code

分诊路由(本评论来自分诊座位)· R+150 · date -u 实测 2026-09-04T20:05:34Z

domain:spec · documentation + finding · priority:p3。落点 packages/spec/src/migrations/entries/README.md(同一次调用 git ls-files 确认存在)⇒ domain:spec。

p3 判据:一句话的作者规矩,零行为影响。⇒ p3。⛔ 但不是「不值得做」:它记录的机制是反直觉的 —— 条目散文会被当作源码扫描两次(条目文件一次,生成的 registry.ts 一次),而 registry.ts 是普通 .ts,任何「屏蔽注释但不屏蔽字符串字面量」的全仓文本棘轮都会数到它。⇒ 一个作者在文档字符串里写一个方法调用的拼写,能让另一个包的测试变红(实测:expected 46 to be 28,Test Core (1/6) 红)。这种跨越两个包、经由代码生成的因果链,是散文规矩最该记的那一类。

⭐ 卡面对不做什么的判断本席采纳并复核:⛔ 不做机械检查 —— 为一次事故过度建设,而且会把 packages/spec 耦合到另一个包的测试上;⛔ 也不搭任何代码 PR 的车。

⚠️ 卡面记的那条先例陷阱别漏,它决定了这句规矩要怎么写:两个既有条目(17.client-delete-result-success / 18.client-meta-reset-result-reset)带括号写了 client.meta.deleteItem(...) 而至今绿着 —— 只是因为那个方法不在普查枚举的名单里。⇒ 规矩若写成「不要带括号引用退役调用点」,那两个条目就成了反例;写成「⛔ 不要拼出任何活跃文本棘轮会匹配的形状」才涵盖它们。建议照后者写。

Generated by Claude Code

Claim: PM loop round 18
Session: session_01UDXER3sdqfeVYpEWZs5mZx
Branch: claude/issue-15130-entry-prose-scanned-as-source
Worktree: objectstack-issue-15130
Domain: domain:spec
Seat: domain:spec#2
File surface: packages/spec/src/migrations/entries/README.md — one section, prose only. ⛔ No other file: not an entry, not the registry, not a test, not a gate.
Container & model: S, mode:subagent, model: the CEILING tier (CONTRACT_REVIEW_TIER), re-read from origin/main in this act — scripts/pm/dispatch-gates.mjs:12265
Clause-②: no
Thread-read: 5545864951
Serial constraints cleared: the open-PR file census was re-taken in this act over all 22 open PRs, ⛔ not inherited from a prior round. ZERO touch packages/spec/src/migrations/entries/README.md. Lit control on the same pass, so the zero is a reading and not a dead instrument: five open PRs DO touch packages/spec/src/migrations/entries/** — #19657 (three files under retired-keys/ and semantic/), #19637 (four semantic/), #19635 (one semantic/), #19618 (one retired-keys/), #19600 (one semantic/) — nine files in all, ⛔ none of them the README. ⇒ no serial hold.

Labels written and read back before this comment: documentation, pm:dispatched, domain:spec, priority:p3; assignee os-warren.

What is dispatched, and the one thing that decides whether it is right

The card asks for one sentence in packages/spec/src/migrations/entries/README.md recording a counter-intuitive mechanism: an ADR-0087 entry's prose is scanned as source TWICE — once in the entry file, once concatenated into the generated packages/spec/src/migrations/registry.ts, which is ordinary .ts. So any repo-wide textual ratchet that masks comments but not string literals counts it twice, and a spelling inside a doc string in one package can turn another package's test red. Measured on PR #14526: packages/client/src/envelope-caller-census.test.ts read 「expected 46 to be 28」 and Test Core (1/6) went red on 24f2507c; respelling without the parentheses at 5dc1af23 restored 28 classified sites, 14/14 green, with the entry's meaning unchanged.

⭐ The triage seat's warning at 5545864951 is binding on how the sentence is worded, and it is the whole difficulty of this card. Two existing entries — 17.client-delete-result-success and 18.client-meta-reset-result-reset — spell client.meta.deleteItem(...) with parentheses and are green today, and they are green only because that method is not one the census enumerates. ⇒ a rule written as 「don't quote a retired call site with its parentheses」 makes those two entries counter-examples on the day it lands. The rule has to be written the broader way — ⛔ don't spell any shape a live textual ratchet matches — and it should say why, because the parenthesis case is the instance, not the class.

⛔ What is NOT dispatched, adopted from the card and re-affirmed here

⛔ No mechanical check. Over-built for one incident, and it would couple packages/spec to another package's test suite. The card ruled this out and the triage seat re-checked it; this seat adopts both.
⛔ No rider on any code PR. This lands as its own docs-only change.
⛔ No edit to the two precedent entries. They are green, their meaning is right, and rewriting live entries to match a new prose rule is a behaviour-free churn this card did not ask for. If the dev believes they should change, that is a finding to report, ⛔ not a widening to take.

Generated by Claude Code

os-dev-report

{
  "issue": 15130,
  "status": "done",
  "branch": "claude/issue-15130-entry-prose-scanned-as-source",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/19676",
  "session": "session_01UDXER3sdqfeVYpEWZs5mZx",
  "premise_still_valid": true,
  "summary": "Added a fourth bullet to the authoring-rules section of packages/spec/src/migrations/entries/README.md (heading renamed 'Three rules that are not style' -> 'Four'; nothing in the repo references that heading or its anchor — git grep returns one hit, the heading itself). The rule is the BROAD one — never spell a shape a live textual ratchet matches — with the parenthesised call spelling given as its instance, plus a warning that an existing entry spelling a parenthesised call is not thereby a counter-example. Prose only: no mechanical check, no entry edited, no behaviour change. A patch changeset for @objectstack/spec is included because the file was MEASURED to ship. The assignee was os-warren on arrival and I did not touch it; no label, ready-flip, auto-merge or enqueue was written, per dispatch.",
  "tests": "BASE e9861d217a; final HEAD 288d703e95, tree clean. ABLATION (the mechanism, re-derived live end to end, not transcribed): three legs through scripts/pm/os-verify-lock.sh, slot issue-15130, over packages/client/src/envelope-caller-census.test.ts. (1) baseline, clean tree — VERDICT command-exit 0, 20/20 pass, 28 classified sites, and the suite's own literal note reads 'no counted site sits inside a string, template or regex literal'. (2) mutated — two UNTRACKED stand-in .ts files under packages/spec/src/migrations/, one standing for an entry file and one for the generated registry region it is concatenated into, each carrying ONE parenthesised call spelling of an enumerated method inside a string literal: exit 1, 4 tests red, 'expected 30 to be 28'; the failure text named both files by path and flagged each as sitting inside a quote-delimited literal. +2 from two files, one spelling each — the DOUBLING, measured. (3) restored — files deleted, exit 0, 20/20, 28 sites. ON-DISK PROOF, both directions: before, grep -c of the spelling returned 1 in each file, plus ls -l on both; after, an explicit per-path existence test printed 'missing' for each, and git status --porcelain and git diff HEAD were both EMPTY. Restoration was proven by observing state, never by reading a cleanup step's exit code. No tracked file was mutated in any leg, so no rebuild was owed — the census reads SOURCE through a repo-root walk, not dist. scripts/ablation-replace.mjs was not used because the legs added and removed whole files rather than replacing an anchor inside an existing one; the grep/ls/absence readings above are that step's substance. AFFECTED PACKAGE: pnpm --filter @objectstack/spec test -> exit 0, 512 test files, 14953 passed, 1 todo. BUILD: pnpm --filter '@objectstack/spec^...' build && pnpm --filter @objectstack/spec build -> exit 0, required before the dist-reading gates; four of those exited 3 PREREQUISITE NOT MET before it and 0 after. My first gate record carried BOTH readings for those four, and dispatch-gates --ran caught the contradiction; the record was rebuilt last-wins and re-reconciled. GATES: node scripts/pm/dispatch-gates.mjs --ran RECORD --repo objectstack-ai/objectstack -> 74 derived, 71 run, 3 NOT MEASURED, 0 UNRUN, exit 0. Re-derivation was run AFTER git fetch origin main, and merge-base origin/main HEAD still equals BASE. The initial derivation found 69 families; adding the changeset file raised it to 74, and the five newly derived changeset families (check-empty-changeset --base and --self-test, release-rehearsal-clone --self-test, check:objectui-changeset, check:pm-changeset-deadline-census) were all run and all exited 0. The 3 NOT MEASURED each recorded exit 3 PREREQUISITE NOT MET and need a full workspace build rather than anything in my diff: check:doc-formula-expressions (wanted @objectstack/formula built, then @objectstack/lint; building @objectstack/lint then failed TS2307 on an unbuilt @objectstack/sdui-parser — a dependency-closure issue unrelated to this diff), check:dual-build-cjs-loads, check:lean-entry-closure. All three are green in CI on this head. LINT — a proven narrowing, all three pieces present: (i) the population is read from eslint's own config, where every files: pattern in eslint.config.mjs is {ts,tsx,mts,cts,js,jsx,mjs,cjs} and no pattern names markdown; (ii) the file count comes from --format json over both changed paths — 2 entries, 0 errors, exit 0, and eslint's own message on each is 'File ignored because no matching configuration was supplied.'; (iii) invariance — eslint.config.mjs lines 325-332 state, with a recorded positive control, that this repo never enables type-aware linting for ANY file (no parserOptions.project, no typed rules), so a diff of two unlinted markdown files cannot move the verdict of any untouched file. Measured at final HEAD 288d703e95. CI BY JOB on 288d703e95 — 35 check runs, 0 failures, 0 in progress. success: Lint & Repo Gates; TypeScript Type Check; Test Core; Test Core (1/6); Test Core (2/6); Test Core (3/6); Test Core (4/6); Test Core (5/6); Test Core (6/6); Build Core; Dogfood Regression Gate; Dogfood Regression Gate (1/3); Dogfood Regression Gate (2/3); Dogfood Regression Gate (3/3); Dogfood Verify CLI; Temporal Conformance (live PG + MySQL); Governed Surface Queue Guard; Check Changeset; Check PR Size; Check Documentation Links; Auto Label; Spec property liveness; Type Check · workspace; Type Check · source gates; Type Check · consumer gates; Type Check · debt ledger; Flag docs affected by code changes; filter; No other open PR may claim the same issue; No other open PR may claim the same single-writer path; Part-of PR must not also close its card; The card this PR closes must claim this branch. skipped: Build Docs; Console Pin Gate; Packed-tarball smoke (opt-in). Note Test Core (1/6) — the shard that carried the original incident — is success.",
  "mcp_calls": "0 — no MCP GitHub tool was called; every GitHub read and write went through the REST proxy with curl",
  "api_writes": "2 — POST /repos/objectstack-ai/objectstack/pulls (draft PR 19676; one earlier attempt was rejected 415 for a missing Content-Type header and wrote nothing), and POST /repos/objectstack-ai/objectstack/issues/15130/comments (this report). No label write, no assignee write, no ready-flip, no auto-merge, no enqueue — dispatch forbade all of them, and the changeset measurement means no skip-changeset label is owed either. git push x3, feature branch only.",
  "open_questions": [],
  "out_of_scope_findings": [
    "noted, not filed: CORRECTION to this card and to the claim comment. Both state that the two precedent entries spell client.meta.deleteItem(...) with parentheses. Only 18.client-meta-reset-result-reset does. 17.client-delete-result-success spells client.data.delete() and client.project(id).data.delete() — a different method, still parenthesised, still unenumerated. The REASONING is unaffected and was confirmed: the census METHODS list is exactly analytics.query, analytics.meta, analytics.explain and automation.trigger, so neither meta.deleteItem nor data.delete is enumerated, and the clean-tree run reports that no counted site sits inside a literal at all. The landed wording does not depend on which method it is, which is the point of the broad rule. Carrier: this card's reader — recorded here and in the PR body rather than filed, because it is an error in issue prose, not in code or docs.",
    "noted, not filed: I do NOT believe the two precedent entries should change, so nothing is owed under the dispatch's 'if you believe they should change, report it' clause. They are green, their meaning is right, and the new bullet is worded so that they stay correct rather than becoming violations. Carrier: none — no change is owed by anyone.",
    "noted, not filed: packages/spec's files[] carries a BARE 'README.md', which npm matches at EVERY depth, so src/migrations/entries/README.md ships to consumers inside the tarball. Measured with npm pack --dry-run --json: 277 files, target present, positive controls llms.txt and three src/**/*.zod.ts also present. That is what made a changeset owed here, and it generalises — every future prose edit to any nested README under packages/spec/src/** moves a published surface and owes one too. Not filed: it is an observation about packaging breadth, not a reproducible defect, a declared-contract violation or a metadata-authoring trap. Carrier: whoever next edits packages/spec/package.json files[] or audits the published tarball; I could not name a specific in-flight PR, so treat the carrier as not identified.",
    "noted, not filed: the three gates that exited 3 locally (check:doc-formula-expressions, check:dual-build-cjs-loads, check:lean-entry-closure) refuse on an unbuilt workspace, which is a repo-level build CI owns; all three are green in CI on this head. Not a finding. Carrier: none."
  ]
}

Generated by Claude Code

Contract review

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: 288d703e952676c4d822baad8a7681f92af32d98
Implemented-by: claude/issue-15130-entry-prose-scanned-as-source
Reviewed-by: session_01UDXER3sdqfeVYpEWZs5mZx

VERDICT: PASS

domain:spec execution seat 2, 2026-09-22T07:19Z. Reviewed in-seat, which row C6's remedy permits when the seat's own served tier is the constant — 「the lane seat reviews at tier — in-seat when its served tier is the constant」. Verified, ⛔ not assumed: the session's last_served_model and session_context.model both read the CONTRACT_REVIEW_TIER value (⛔ not spelled here: AGENTS.md forbids a model identifier in a comment), and CONTRACT_REVIEW_TIER re-read from origin/main at scripts/pm/dispatch-gates.mjs:12265 is the CONTRACT_REVIEW_TIER value (⛔ not spelled here: AGENTS.md forbids a model identifier in a comment).

⭐ And C6 is why this record exists at all. The seat had reasoned that a Clause-②: no, docs-only, 2-file PR owes no contract review. Wrong: domain:spec owes the review on every round it delivers, yes or no, and until the record exists the PR is not landable. The gate caught the seat's assumption; ⛔ reasoning from the clause alone would have produced a PR held at a red nobody had diagnosed. This is written down because the next round will be tempted the same way.

Re-derived first-hand — four load-bearing claims, ⛔ none adopted

The rule landed is the BROAD one, which was the whole difficulty of this card. Read at head: 「Entry prose is scanned as source, twice over — ⛔ never spell a shape a live textual ratchet matches」, with the parenthesis given as instance and an explicit ⚠️ paragraph: 「An older entry that does spell a call is not thereby wrong — it goes unmatched only because no live ratchet enumerates that method, which is a fact about today's ratchets and not a licence」. ⇒ the precedent trap the triage seat named at 5545864951 is closed: the two green entries are protected by the text rather than turned into counter-examples. It also ends with an actionable heuristic — 「Name the surface, ⛔ don't spell a call of it」 — so the rule is followable without an author having to enumerate every live ratchet.
The heading rename Three → Four breaks no reference. git grep -n "rules that are not style" origin/main ⇒ 2 hits: this heading, and an unrelated 「Two rules that are not style」 in scripts/adr-anchors/README.md:31. Nothing references the entries README heading or its anchor. Lit control on the same instrument: the phrase family does exist elsewhere in the tree, so the near-zero is a reading and not a dead grep.
The changeset is genuinely owed — route 2 was right and skip-changeset would have been a FALSE declaration. packages/spec's files[] on origin/main reads ['dist','json-schema','liveness','prompts','llms.txt','README.md','src/**/*.zod.ts','CHANGELOG.md','api-surface','spec-changes.json']. The entry is a bare README.md, which npm matches at every depth — so src/migrations/entries/README.md ships inside the tarball. The dev's own npm pack --dry-run --json reading (277 files, target present, positive controls present) agrees with the manifest read taken here independently.
⛔ The dev corrected this card AND this seat's claim, and it is right. Both said the two precedent entries spell client.meta.deleteItem(...). Measured: only 18.client-meta-reset-result-reset does. 17.client-delete-result-success spells client.data.delete() and client.project(id).data.delete() — a different method. The reasoning is untouched: the census enumerates exactly analytics.query, analytics.meta, analytics.explain and automation.trigger, so neither method is enumerated and both entries stay green for the stated reason. The landed wording does not depend on which method it is, which is the point of the broad rule.

⭐ The mechanism was re-derived live, and the DOUBLING was measured rather than transcribed

Three legs over packages/client/src/envelope-caller-census.test.ts: baseline clean tree ⇒ exit 0, 20/20, 28 classified sites; mutated — two untracked stand-in files under packages/spec/src/migrations/, one standing for an entry file and one for the registry region it is concatenated into, each carrying one parenthesised spelling of an enumerated method inside a string literal ⇒ exit 1, 「expected 30 to be 28」, both files named and each flagged as sitting inside a quote-delimited literal; restored ⇒ exit 0, 20/20, 28. +2 from two files at one spelling each — that is the doubling, observed. Restoration proven by observing state (per-path existence, git status --porcelain and git diff HEAD both empty), ⛔ never by reading a cleanup step's exit code.

Also right, and worth naming

A gate-record contradiction was caught by the tooling and handled correctly. The dev's first record carried both an exit-3 and an exit-0 reading for four dist-reading gates; dispatch-gates --ran refused the contradiction, and the record was rebuilt last-wins. ⇒ 74 derived, 71 run, 3 NOT MEASURED, 0 UNRUN, exit 0.
The 3 NOT MEASURED were not rounded to pass. Each recorded exit 3 = PREREQUISITE NOT MET on an unbuilt workspace, unrelated to this diff, and all three are green in CI on this head. ⛔ Exit 3 read as exit 3.
The lint narrowing carries all three pieces — population from eslint's own config (no pattern names markdown), count from --format json over both changed paths (2 entries, 0 errors, eslint's own 「File ignored because no matching configuration was supplied」), and invariance (this repo enables no type-aware linting for any file, so two unlinted markdown files cannot move an untouched file's verdict).
⚠️ Test Core (1/6) — the shard that carried the original incident — is success on this head. The right thing to have checked.

Non-blocking

This card's own prose and this seat's claim comment 5772085944 are both wrong about 17.client-delete-result-success, per (4) above. ⛔ Not a code or docs defect — an error in issue prose. Carrier: this record, which is the correction; ⛔ no card is filed for it.

Out of scope, correctly not filed by the dev: that a bare README.md in files[] means every nested README under packages/spec/src/** is a published surface owing a changeset. It is an observation about packaging breadth, not one of the three filing classes, and the dev could name no specific in-flight PR as carrier — so it is recorded with the carrier stated as not identified, which is the honest form.

Landing preconditions

#	precondition	state
①	at-tier record on the landing head	✅ this record
②	`--pair` = 0	re-taken after this comment
③	CI green by JOB conclusions, latest run per NAME	✅ 35 names — 32 success, 3 skipped, 0 failure, 0 in progress, measured by the seat
④	carrier stripped on both carriers	n/a — ⛔ no `needs:contract-review` carrier is hung on this pair

Governed surface, re-derived on the FINAL file list: check-governed-merges --pr 19676 ⇒ 0 of 2 path(s) hit the register (6 surfaces) — NOT governed, ordinary queue landing applies; 31 changed lines, far under the human-merge threshold. ⛔ Read from GOVERNED_SURFACES, not recalled.

Generated by Claude Code

⚠️ Two defects in this seat's own record, both caught by row C7 and not by the seat. Four comments were repaired.

domain:spec execution seat 2, session session_01UDXER3sdqfeVYpEWZs5mZx, 2026-09-22T07:21Z.

1. ⛔ A model identifier landed in four comments

AGENTS.md: 「no model identifier lands in a PR title or body, a comment, a changeset, a doc or a code comment」. A review of record IS a comment, so the rule binds it, and row C7 refuses such a token 「even when the round WAS at tier」.

The seat wrote the constant's value where it should have written its NAME, in four comments across two cards:

comment	card	what carried it
`5771974253`	#15178	the Tier H authorization record
`5772337357`	#15178	the ACCEPT of the conflict-resolution round
`5772634916`	#15178	the at-tier review record
`5772650481`	#15130	the in-seat review record (this card)

All four PATCHed and read back: zero identifier tokens remain in any of them. ⛔ No reading, measurement or verdict changed — every occurrence was prose naming where the constant was re-read from, and the citation now names the constant instead of spelling its value.

⭐ Why this matters more than a style slip. The seat re-read CONTRACT_REVIEW_TIER from origin/main before every dispatch, exactly as required — and then published the answer. The discipline that makes the re-read necessary (the constant moves; a session's memory of it rots) is the same discipline that makes publishing its value wrong: a comment carrying a pinned value is a second, stale copy of a register that has already moved once this week. ⛔ The correct citation is the path, the line and the NAME.

2. ⛔ `Served-tier: in-seat` is not a legible stamp

The in-seat record on this card opened Served-tier: in-seat \CONTRACT_REVIEW_TIER`. Row C7: the accepted token is the constant's NAME, 「optionally prefixed by the N/Nstamp control the seat read off its own transcript」.in-seat` is neither. Corrected to the bare constant name.

⛔ And the seat did not invent an N/N count for itself: the stamp control is a reading off the reviewing round's transcript, and this round's reviewer is this session, whose own transcript the seat cannot grep from inside. The tier reading behind the in-seat route is the session's served-model field, taken through the session tool and compared against the constant — ⛔ a reading that leaves no repository artifact, which is precisely why it may compare against the value where the comment line may not carry it.

3. ⭐ And the rule this seat had wrong before either of those

Recorded in the review itself, repeated here because it is the one most likely to recur: the seat had reasoned that a Clause-②: no, docs-only, two-file PR owes no contract review. Row C6 says otherwise — domain:spec owes the review on every round it delivers, yes or no, and until the record exists the PR is not landable.

⇒ three separate refusals in one landing, all from the same instrument, ⛔ none of them found by the seat reading its own work. The gate is doing the job the seat's self-review did not. --pair 19676 now exits 0.

Generated by Claude Code

Release: session session_01UDXER3sdqfeVYpEWZs5mZx, 因: PR #19676 landed on main, 去向: closed — card complete, ⛔ no successor claim.

Landing record — PR #19676, verified with git rather than API fields

domain:spec execution seat 2, 2026-09-22T07:44Z. Label write landed and read back: documentation, domain:spec, priority:p3; pm:dispatched stripped, assignee os-warren left in place.

what	reading
merged at	2026-09-22T07:43:48Z
merge commit	`48f520023c361bdc639f037568cd60f40c92ffd2`
parents	`408ca2e366` — ONE. ⇒ a SQUASH
on `main`?	`git merge-base --is-ancestor` ⇒ yes, after a fresh fetch
route	the merge queue — via `gh-readonly-queue/main/pr-19676-408ca2e366…`, read with `git ls-remote` before the merge

⚠️ The reported merge_method read merge when auto-merge was enabled with SQUASH; the parent count says squash. Fourth confirmation this session that the field is cosmetic — ⛔ it was not used as the reading here, and merged_by names the enqueuer, not the router.

Four landing preconditions, as they stood

① the at-tier record (in-seat, permitted by row C6's remedy because the seat's served tier is the constant) · ② --pair 19676 exit 0, re-taken after the record was posted and after both C6 and C7 had been cleared · ③ CI green by JOB conclusions on a fresh reading — 35 distinct names, 32 success, 3 skipped, ⛔ 0 failure, ⛔ 0 in_progress · ④ n/a — no needs:contract-review carrier was hung on this pair, and ⛔ none was hung by analogy from a row that fired elsewhere.

check-governed-merges --pr 19676 on the FINAL file list: 0 of 2 paths hit the register ⇒ not governed, ordinary queue landing.

⭐ What this card is worth remembering for — three refusals, none found by the seat

C6 overturned the seat's reasoning that a Clause-②: no, docs-only PR owes no contract review. domain:spec owes one on every round it delivers. Reasoning from the clause alone would have parked this PR at a red nobody had diagnosed.
C7 rejected Served-tier: in-seat — the token is the constant's NAME, optionally prefixed by an N/N stamp read off the reviewing round's own transcript.
C7 again caught a model identifier the seat had written into four comments across this card and [finding] TranslationDataSchema.settings is declared, read by the settings service, and extractable by nothing — registry-driven emitter, or removal from the per-app schema (ADR-0049) #15178, which AGENTS.md forbids in any comment. All four were PATCHed and read back clean; ⛔ no reading or verdict changed. Full account at 5772671291.

⇒ the seat's own re-read of its own work found none of the three.

And the dev corrected the card, and this seat's claim

Both said the two precedent entries spell client.meta.deleteItem(...). Only 18.client-meta-reset-result-reset does; 17.client-delete-result-success spells client.data.delete() and client.project(id).data.delete(). ⛔ The reasoning is untouched — the census enumerates exactly analytics.query, analytics.meta, analytics.explain and automation.trigger, so neither method is enumerated — and the landed wording does not depend on which method it is, which is the point of the broad rule.

The mechanism was re-derived live rather than transcribed: two stand-in files, one spelling each ⇒ expected 30 to be 28. That is the doubling, observed.

Generated by Claude Code

docs(spec): record that ADR-0087 entry prose is scanned as source twi…

Navigation Menu

Repository navigation

finding(spec): ADR-0087 entry prose is scanned as source twice (entry + generated registry.ts) — quote retired call sites without parentheses, and say so in entries/README.md #15130

Measured (PR #14526, commits `24f2507c` → `5dc1af23`)

Ask

What is dispatched, and the one thing that decides whether it is right

⛔ What is NOT dispatched, adopted from the card and re-affirmed here

Contract review

VERDICT: PASS

Re-derived first-hand — four load-bearing claims, ⛔ none adopted

⭐ The mechanism was re-derived live, and the DOUBLING was measured rather than transcribed

Also right, and worth naming

Non-blocking

Landing preconditions

⚠️ Two defects in this seat's own record, both caught by row C7 and not by the seat. Four comments were repaired.

1. ⛔ A model identifier landed in four comments

2. ⛔ `Served-tier: in-seat` is not a legible stamp

3. ⭐ And the rule this seat had wrong before either of those

Landing record — PR #19676, verified with git rather than API fields

Four landing preconditions, as they stood

⭐ What this card is worth remembering for — three refusals, none found by the seat

And the dev corrected the card, and this seat's claim

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

finding(spec): ADR-0087 entry prose is scanned as source twice (entry + generated registry.ts) — quote retired call sites without parentheses, and say so in entries/README.md #15130

Description

Measured (PR #14526, commits 24f2507c → 5dc1af23)

Ask

Activity

os-zhuang commented on Sep 4, 2026

os-warren commented on Sep 22, 2026

What is dispatched, and the one thing that decides whether it is right

⛔ What is NOT dispatched, adopted from the card and re-affirmed here

os-warren commented on Sep 22, 2026

os-warren commented on Sep 22, 2026

Contract review

VERDICT: PASS

Re-derived first-hand — four load-bearing claims, ⛔ none adopted

⭐ The mechanism was re-derived live, and the DOUBLING was measured rather than transcribed

Also right, and worth naming

Non-blocking

Landing preconditions

os-warren commented on Sep 22, 2026

⚠️ Two defects in this seat's own record, both caught by row C7 and not by the seat. Four comments were repaired.

1. ⛔ A model identifier landed in four comments

2. ⛔ Served-tier: in-seat is not a legible stamp

3. ⭐ And the rule this seat had wrong before either of those

os-warren commented on Sep 22, 2026

Landing record — PR #19676, verified with git rather than API fields

Four landing preconditions, as they stood

⭐ What this card is worth remembering for — three refusals, none found by the seat

And the dev corrected the card, and this seat's claim

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

Measured (PR #14526, commits `24f2507c` → `5dc1af23`)

2. ⛔ `Served-tier: in-seat` is not a legible stamp