test/docs-drift.test.ts is at 97.8% of the 100KB hygiene cap — the next rule added to it turns the gate red

Found while landing #1187. Filing rather than fixing: out of that card's scope.

Problem

test/docs-drift.test.ts is 100,190 bytes on main @ 540e4887. The cap in
scripts/check-source-hygiene.mjs is MAX_FILE_BYTES = 100 * 1024 = 102,400, so
the file has 2,210 bytes of headroom — 97.8% full.

That is not a hypothetical. #1187 asked for a new doc-drift rule and pointed it at
this file's "docs match the built stack" family, which is the right family for it.
Appending the rule took the file to 107KB:

✗ no source file over 100KB — 1 violation(s)
    test/docs-drift.test.ts:0  107KB
    → split the file — oversized modules defeat review

#1187 shipped its rule as a sibling test/docs-readme-token-figures.test.ts
instead, which is where most of that family already lives, so nothing is blocked
today. But the next author who reaches for this file — reasonably, because it is
the named home of the drift rules — pays a detour to discover the same thing.

Why this is worth a card rather than a shrug

Two closed precedents show the same condition being noticed only at the moment it
bites: #814 (test/metadata-references.test.ts, 817 bytes of headroom) and #635
(src/data/index.ts, ~1.5KB under). In both, the discovery cost landed on an
unrelated PR. The cap has no early warning: the gate is silent at 99% and red at
101%, so the signal arrives as a failed run on someone else's change.

Worth considering, in rough order of cost:

Split docs-drift.test.ts along the seams it already has. It holds several
independent describe families (the flow-threshold rules, the [17.0-rc2验收] README 与 marketplace 文档的对象/flow 数量仍是旧值(15 objects / 23 flows),rc.2 实际为 17 / 24 #729 count rule,
the [17.0-rc2验收] objectstack.config.ts 的 engines.protocol 仍是 ^17.0.0-rc.1,与 objectstack.manifest.json(rc.2)及 build 产物不一致 #728 protocol-version rule, the docs/STATUS.md 的「当前状态」全表过期：对象/流程/仪表盘/动作计数、平台版本、测试规模无一与 main 相符 #1011 STATUS transcript rule), each with
its own long header comment. Most would move out whole.
Warn before failing. A hygiene advisory at, say, 90% of the cap would have
surfaced all three instances of this while the fix was still cheap — the same
shape as the token ratchet's re-anchor nag.

No behavior change either way; the tests all pass at their current size.

Not in scope here

The 100KB cap itself is well-argued and this is not a request to raise it.

定级（首触）→ `pm:queue`。它现在还挡着 #1135，所以不是「等它咬人再说」。

按本席常设授权定级。finding 同笔摘除。

前提对 `origin/main` @ `6ed7b8d` 重测

$ wc -c test/docs-drift.test.ts       →  99,746
  cap (MAX_FILE_BYTES = 100 * 1024)   →  102,400
  headroom                            →   2,654 B   (97.4% 满)

卡片记的是 100,190 / 2,210 余量；文件这期间小了一点，但结论一字未变：下一条规则塞进去就红。

为什么晋级 —— 它已经不是假设了

卡片说「nothing is blocked today」。那句话今天不成立了：本轮定级把 #1135 判进了队列，而 #1135 的可交付正是往 test/docs-drift.test.ts 里加断言（它要复用那里已有的 capCel 机器，那是正确的落点）。

⇒ 两张卡在同一个文件上硬串行，而顺序是确定的：#1196 先，#1135 后。反过来做，#1135 会在实施到一半时撞上 100KB 上限，然后面对一个它没被授权做的选择（拆文件 / 抬上限 / 换文件）—— 这正是 #814 和 #635 各自付过的账，卡片把它们列为先例。

裁定：走选项 1（沿既有接缝拆分），⛔ 不抬上限

✅ 拆分：卡片指出该文件已有若干独立的 describe 家族（flow 阈值规则、[17.0-rc2验收] README 与 marketplace 文档的对象/flow 数量仍是旧值(15 objects / 23 flows),rc.2 实际为 17 / 24 #729 计数规则、[17.0-rc2验收] objectstack.config.ts 的 engines.protocol 仍是 ^17.0.0-rc.1,与 objectstack.manifest.json(rc.2)及 build 产物不一致 #728 协议版本规则、docs/STATUS.md 的「当前状态」全表过期：对象/流程/仪表盘/动作计数、平台版本、测试规模无一与 main 相符 #1011 STATUS transcript 规则），各带自己的长头注释，多数可以整块搬走。这是沿着已有接缝切，不是重构。
⛔ 不抬 100KB 上限：卡片自己说了「The 100KB cap itself is well-argued and this is not a request to raise it」。本席同意，并补一条纪律理由：抬门禁阈值属门禁削弱，恒在人工地板上（AI 有把 CI 弄绿的结构性动机，削弱农场必须是人的动作）。⛔ 实施者不得以「拆起来麻烦」为由改那个常数。
⏭️ 选项 2（90% 预警）不在本卡：它是好主意，但它是给 check-source-hygiene.mjs 加能力，与拆分是两件事。⚠️ 若实施者认为不加预警就会再犯，单独立卡，⛔ 不要搭车。

拆分时的两条要求

⚠️ 拆完要留一张指路表。 test/metadata-references.test.ts 距 100KB hygiene 上限只剩 817 字节 —— 下一个往里加测试的 PR 会被 gate 拦住 #814 拆 metadata-references.test.ts 时留了 SPLIT BY FAMILY (#814) 一节逐族列出兄弟文件，那是对的做法 —— 但它也制造了 src/ 与 bulk-action-dispatch.test.ts 里 4 条「见 metadata-references.test.ts 的某守卫」的注释，指向的守卫已随 #814 拆分搬家（需多跳一次） #931（四条注释指向搬走了的守卫）。所以：拆分的同时，grep 全仓对被搬走的 describe 的引用并一并更新，⛔ 不要留给下一张卡。
拆完复测每个新文件的字节数并写进报告 —— 拆成两个 51KB 的文件等于什么都没解决。

面：test/docs-drift.test.ts + 拆出的兄弟文件 + 指向它们的注释。⚠️ 这个文件是本仓最热的守卫文件之一，派发前查热文件队；#1135 明确排在本卡之后。

先例：#814（metadata-references.test.ts，817 B 余量）· #635（src/data/index.ts，约 1.5KB）—— 两次都是在咬人的那一刻才被发现，代价落在无关的 PR 上。

Generated by Claude Code

Claim: PM loop round 2
Session: session_01STpYzSP2LiZwZinh6m6Qgf
Branch: claude/issue-1196-docs-drift-split
Worktree: hotcrm-issue-1196
Domain: (hotcrm has no domain:* taxonomy — repo-wide seat, objectstack#10282)
File surface: test/docs-drift.test.ts + the sibling test/*.test.ts files you split it into, + comments that reference the moved describes. ⛔ The reference sweep is fenced OUT of src/pages/**, src/actions/**, src/flows/** and src/translations/** — those are #1207's, dispatched in the same batch. If a reference lives there, report it, do not edit it (stop on breach; explain in the report)
Container & model: M, mode:subagent, model: opus
Clause-②: no — test-file reorganisation.
Serial constraints cleared: 4 open PRs, all dependabot, none touching test/. Batch siblings #1207 (fenced above) and #1169 (.github/workflows/docs-app.yml) — disjoint. ⚠️ #1135 is pm:queue and is explicitly serialised BEHIND this card — it adds assertions to this very file and would hit the 100KB cap mid-implementation. ⛔ Do not implement #1135's assertions here.

Read the full issue body and every comment on GitHub yourself before starting, and verify the body is not truncated.

Ruling (⛔ not re-adjudicable)

✅ Split along the seams the file already has (the card's option 1).
⛔ Do NOT raise MAX_FILE_BYTES. The card says the 100KB cap is well-argued and is not asking for it to be raised; independently, raising a gate threshold is 门禁削弱 and sits on the maintainer floor — AI has a structural incentive to make CI green, so weakening the farm must be a human action. ⛔ Not negotiable, including "just this once for the split".
⏭️ The 90%-of-cap advisory (the card's option 2) is NOT in this card. It is a capability change to scripts/check-source-hygiene.mjs, a different file and a different design. If you conclude the split will re-fill without it, report it and I will file it — ⛔ do not ride it along.

Measured starting point (falsify if wrong)

On origin/main @ 6ed7b8d:

wc -c test/docs-drift.test.ts   →  99,746
MAX_FILE_BYTES (100 * 1024)     → 102,400
headroom                        →   2,654 B   (97.4% full)

Twelve describe families, which are the seams:

:196  package docs do not drift from the flows they document
:318  maintainer docs do not point at directories that no longer exist
:371  product docs do not point at directories that no longer exist
:392  docs that draw the src/ tree only draw directories that exist
:442  documented agent names resolve to a platform agent
:568  the dashboards docs page lists tiles that exist
:785  translated docs pages keep the English page callouts
:882  the action example teaches a selection key the platform can deliver (#813)
:989  product docs do not name a retired copilot persona (#612)
:1190 the docs print the version the manifest declares (#612)
:1421 one protocol version, declared in three files (#728)
:1534 product docs state the metadata counts the stack registers (#729)

⚠️ How to group them into files is your judgement — the card names four independent families; the listing above shows twelve. Group by subject, keep each describe whole with its header comment, and say in the report why you grouped as you did.

⚠️ Two requirements the card does not state, and I do

Grep the repo for references to the moved describes and update them in the same PR. test/metadata-references.test.ts 距 100KB hygiene 上限只剩 817 字节 —— 下一个往里加测试的 PR 会被 gate 拦住 #814 split metadata-references.test.ts and left a SPLIT BY FAMILY pointer table — good — but it also produced src/ 与 bulk-action-dispatch.test.ts 里 4 条「见 metadata-references.test.ts 的某守卫」的注释，指向的守卫已随 #814 拆分搬家（需多跳一次） #931: four comments still pointing at the old file, one of which is still stale today. ⛔ Do not create src/ 与 bulk-action-dispatch.test.ts 里 4 条「见 metadata-references.test.ts 的某守卫」的注释，指向的守卫已随 #814 拆分搬家（需多跳一次） #931's successor. Known referencing files include src/objects/_thresholds.ts, src/docs/*.md, test/helpers/*, test/sharing-coverage.test.ts, test/persona-copy.test.ts, test/docs-revenue-approvals-navigation.test.ts — re-derive the list yourself, and honour the src/ fence above.
Report the byte count of every resulting file. Splitting 99,746 into two 50KB files solves nothing durable — state the numbers and the resulting headroom per file.

Non-negotiable

The suite must be identical before and after. This is a move, not a rewrite: same test count, same names, same assertions. Report the before/after Test Files and Tests lines from vitest run side by side. ⚠️ Note test counts can legitimately differ CI vs local (describe.runIf in test/dataset-granularity.test.ts) — compare local before vs local after, same command.
Gates: pnpm verify = validate → typecheck → lint → lint:i18n-gate → hygiene → hygiene:tokens → build → test. Exit codes after redirect, never through a pipe. ⚠️ pnpm hygiene is the gate that was red at 107KB — it is the one that must now be green and must have actually run.
Live changeset gate (newly-added .changeset/*.md vs base sha, or skip-changeset label, or empty frontmatter). private: true does not exempt it.
PR draft, base main, first body line Fixes #1196.

Report

Terminal JSON report as a comment on this issue first (first line the literal os-dev-report), then as the return message. Leave your own claim comment with your own session ID and branch.

⚠️ Out-of-scope findings: report to the PM, do not try to file. os-dev seats get HTTP 403 from api.github.com (measured this round).

Generated by Claude Code

Claim: os-dev seat (dev subagent dispatched by the PM loop)
Session: session_01STpYzSP2LiZwZinh6m6Qgf (same session as the PM claim above — this seat runs as a subagent inside it, so the session ID is shared by construction; the branch below is what distinguishes this claim)
Branch: claude/issue-1196-docs-drift-split (pushed empty at 6ed7b8dd before the first edit — write route confirmed)
Worktree: /home/user/hotcrm-issue-1196, based on origin/main @ 6ed7b8dd
Assignee: untouched (PM's claim)

Premise re-measured on origin/main @ 6ed7b8dd: wc -c test/docs-drift.test.ts → 99746, cap MAX_FILE_BYTES = 100 * 1024 → 102400, headroom 2654 B (97.4% full). Matches the PM's numbers exactly.

Scope as dispatched: split along the twelve existing describe seams, keep each describe whole with its header comment, sweep and update in-repo references to the moved families — fenced OUT of src/pages/**, src/actions/**, src/flows/**, src/translations/** (#1207's surface; will report, not edit). ⛔ Not raising MAX_FILE_BYTES. ⛔ Not implementing #1135's assertions. Option 2 (90%-of-cap advisory) not ridden along.

Generated by Claude Code

test(docs): split docs-drift.test.ts along its describe seams (#1196)

os-dev-report

{
  "issue": 1196,
  "status": "done",
  "branch": "claude/issue-1196-docs-drift-split",
  "pr": "https://github.com/objectstack-ai/hotcrm/pull/1285",
  "premise_still_valid": true,
  "summary": "Split test/docs-drift.test.ts (99,746 B / 2,654 B headroom / 97.4% of the 102,400-byte MAX_FILE_BYTES) along the describe seams it already had, into the kept file plus seven siblings. The file carried THIRTEEN describes, not the twelve the dispatch listed - 'docs/STATUS.md states the current repository (#1011)' at line 1810 was missing from the list; all thirteen are accounted for. Grouping was by subject, except docs-declared-versions.test.ts where the mechanical seam decided it too: #612 version, #728 protocol and #1011 STATUS all read PACKAGE_JSON / PLATFORM_VERSIONS / h2Sections, so splitting them would have meant a duplicated derivation or a new helper module. capCel/CRON_LABEL/RULES stayed in docs-drift.test.ts, which is where #1135 lands and now has 91,622 B of room. MAX_FILE_BYTES untouched; #1135's assertions not implemented; the card's option 2 (90%-of-cap advisory) not ridden along - reported below instead. Reference sweep: 21 in-repo pointers repointed, each asserted to match exactly once before rewriting, plus 7 formerly-internal cross-references ('the check above', 'its own rule below') that became cross-file at the split. The four fenced trees (src/pages, src/actions, src/flows, src/translations) were NOT touched and contain zero docs-drift references, so the fence cost the sweep nothing.",
  "tests": "pnpm verify (validate -> typecheck -> lint -> lint:i18n-gate -> hygiene -> hygiene:tokens -> build -> test) on 43daabcf, under the container's shared verify lock, exit code read from the wrapper's own verdict line: 'os-verify-lock: VERDICT command-exit 0 - held the lock 118s (1m58s) - waited 0s'. The gate that was red at 107KB printed its own verdict: '  checkmark no source file over 100KB' / '  checkmark copyright header at the top of every .ts file' / 'checkmark source hygiene clean'. NOTE for the reader of the raw log: lines further down reading 'X source token ratchet failed' are FIXTURE OUTPUT inside test/source-token-ratchet.test.ts, not the gate - the real hygiene:tokens run printed 'checkmark source token ratchet clean' above the test step. MOVE PROOF, three independent ways: (1) whole suite, same command, local both times - before (origin/main 6ed7b8dd): 'Test Files  124 passed (124)' / 'Tests  2900 passed | 1 skipped (2901)'; after (43daabcf): 'Test Files  131 passed (131)' / 'Tests  2900 passed | 1 skipped (2901)'. (2) family only - before 'Test Files  1 passed (1)' / 'Tests  76 passed (76)'; after 'Test Files  8 passed (8)' / 'Tests  76 passed (76)'. (3) name-level: vitest run --reporter=verbose over the family before and after, 'describe > it' paths sorted and diffed - 76 names each side, sets IDENTICAL. Structural proof that nothing was dropped: the eleven moved line ranges tile [11, 2018] exactly once each and their concatenation diffs byte-for-byte clean against the original body. NOT-MEASURED check: tsc --noEmit --listFiles puts all eight files in the program (tsconfig include is test/**/* with no *.test.ts exclusion), so 'typecheck clean' really covers them. BYTES / HEADROOM / %CAP per resulting file: docs-drift 10,778 / 91,622 / 10.5%; docs-src-tree-paths 10,035 / 92,365 / 9.8%; docs-runnable-samples 10,641 / 91,759 / 10.4%; docs-dashboard-tiles 11,760 / 90,640 / 11.5%; docs-locale-callouts 6,366 / 96,034 / 6.2%; docs-retired-personas 10,764 / 91,636 / 10.5%; docs-metadata-counts 14,724 / 87,676 / 14.4%; docs-declared-versions 31,963 / 70,437 / 31.2%. Fullest file is 31.2% of cap (was 97.4%); sum 107,031 B vs 99,746 B before, the 7.3KB being eight copyright/import blocks plus the SPLIT BY FAMILY table and per-file family notes. Changeset: .changeset/docs-drift-split-by-family.md with empty frontmatter (the sanctioned 'releases nothing' declaration changeset-check.yml documents); no skip-changeset label applied. CI on the PR had not converged at report time - reporting at draft-PR time per the standing rule.",
  "open_questions": [],
  "out_of_scope_findings": [
    "NOT FILED (os-dev seats get HTTP 403 from api.github.com) - PM to file: the card's option 2, a 90%-of-cap advisory in scripts/check-source-hygiene.mjs. My conclusion after doing the split: this PR fixes the INSTANCE, not the CLASS. No resulting file will re-fill soon (fullest is 31.2%), but the cap is still silent at 99% and red at 101%, and this is the third time the condition was found at the moment it bit (#814 at 817 B headroom, #635 at ~1.5KB, this card at 2,654 B). The advisory is what would have named all three while the fix was cheap. Same shape as the token ratchet's re-anchor nag; it is a capability change to a different file and a different design, so it wants its own card.",
    "NOT FILED (403) - PM to judge: with docs-drift.test.ts down to 10,778 B, the repo's largest first-party source file is now test/flow-scheduled.test.ts at 71,456 B (69.8% of the 100KB cap), followed by test/sharing-coverage.test.ts at 66,068 B (64.5%). Neither is urgent - 30KB+ of room each - and this is an observation, not a defect. It is the concrete thing a 90% advisory would watch next.",
    "Dispatch correction, no card needed: the PM claim's seam list has twelve describes; the file carried thirteen. The omitted one is 'docs/STATUS.md states the current repository (#1011)' at line 1810 - the card's own prose does name it (the '#1011 STATUS transcript rule'), so only the line-numbered listing was short. It went into docs-declared-versions.test.ts.",
    "Considered and deliberately NOT edited, flagged so the next sweep does not re-litigate it: content/docs/whats-new{,.zh-Hans,.zh-Hant}.mdx say 'the repository's docs-drift guard' with no file path, so nothing there points at a moved family; a reader who follows the name lands on the SPLIT BY FAMILY table in the kept file. Same for CHANGELOG.md and the three existing .changeset/*.md entries - dated records of what a past change did, true when written, and none names a path that moved. Rewriting release history in a code PR would also cross the standing 'release notes are written centrally' rule."
  ]
}

Branch claude/issue-1196-docs-drift-split @ 43daabcf, draft PR #1285 (base main, first body line Fixes #1196). Both worktrees (hotcrm-issue-1196 and the origin/main comparison tree used for the before-numbers) were removed without --force.

Generated by Claude Code

PM review — ACCEPT. PR #1285. Holding for `Playwright` to converge, then landing.

Reviewed against the repo and against GitHub, not against the report.

Checklist: draft ✓ · base main @ 09f83fef (already on the post-#1284 head, no rebase owed) ✓ · first body line Fixes #1196 ✓ · changeset newly added, empty frontmatter — sanctioned "releases nothing" form, Check Changeset green ✓ · 23 files, all inside the declared surface ✓ (see below) · Build and Test (22.x) confirmed by NAME as completed: success ✓ · path fork: no docs/adr/**, no .claude/**, no skills/**, no AGENTS.md / CLAUDE.md ⇒ not governed, loop-mergeable ✓

⏳ Not yet landed: Playwright is in_progress, which is why mergeable_state reads unstable. My entry criterion is every check green, not the required subset — a non-required red is either a real defect or a broken gate, and both are mine to handle before the queue takes it. Landing when it converges.

The four numbers this card exists for — measured myself on the branch head, not read off the report

test/docs-drift.test.ts              10,778   10.5%   ← where #1135 lands
test/docs-src-tree-paths.test.ts     10,035    9.8%
test/docs-runnable-samples.test.ts   10,641   10.4%
test/docs-dashboard-tiles.test.ts    11,760   11.5%
test/docs-locale-callouts.test.ts     6,366    6.2%
test/docs-retired-personas.test.ts   10,764   10.5%
test/docs-metadata-counts.test.ts    14,724   14.4%
test/docs-declared-versions.test.ts  31,963   31.2%   ← fullest

97.4% → fullest 31.2%. Every figure matches the report exactly. And the file #1135 actually lands in has 91,622 bytes of room, which is the point of the card rather than a side effect.

Also verified myself, because they were the two things I forbade:

git diff origin/main FETCH_HEAD -- scripts/check-source-hygiene.mjs → empty. MAX_FILE_BYTES untouched.
The four fenced trees (src/pages, src/actions, src/flows, src/translations) → zero files touched. The fence held, and the dev measured that those trees contain no docs-drift references at all — so it cost the sweep nothing and left nothing stranded for A lead flagged duplicate_status: suspected looks clean on its record page, and Convert Lead never mentions it — the duplicate converts into a second account, contact and opportunity #1207.

On the surface question, since 23 files looks wide

It is not a breach. My claim declared "test/docs-drift.test.ts + the sibling files you split it into, + comments that reference the moved describes" and fenced only the four src/ trees. objectstack.config.ts, docs/STATUS.md and the three content/docs/analytics/dashboards*.mdx are each exactly that — a pointer to a family that moved. In-surface by the wording I wrote.

The move proof is the best part

Three independent ways, and the third is the one that matters:

Whole suite, same command, local both sides: 124 → 131 files (+7 siblings), 2900 tests either side.
Family only: 1 → 8 files, 76 tests either side.
⭐ Name-level: vitest run --reporter=verbose, describe > it paths sorted and diffed — 76 names each side, sets identical. Counts can match while names drift; only this one rules that out.

Plus a structural proof that nothing was dropped: the eleven lifted line ranges tile [11, 2018] exactly once each, and their concatenation is byte-for-byte identical to the original body. That converts "it's a move, trust me" into arithmetic.

And the tsc --noEmit --listFiles check again — all eight files are in the typecheck program, so "typecheck clean" is a statement about them rather than around them. Same discipline as #1169's dev, independently.

The #931 half — this is what I asked for and it was done properly

21 pointers repointed, each asserted to match exactly once before rewriting (not a blind sed), plus 7 cross-references that were internal to the old file and became cross-file the moment it split — "the check above", "its own rule below", "the flow rules at the top of this file". Those seven are the ones a careless sweep leaves behind, and they are precisely how #931 happened.

Two comments quoting the old size as the reason a rule shipped elsewhere were rewritten rather than left: the subject argument stands on its own, the arithmetic one no longer holds. That is the right call — a stale "that file is 97KB" would have become tomorrow's #1188.

⭐ And the deliberate non-edits are argued, not skipped: CHANGELOG.md and the existing .changeset/*.md are dated records true when written and name no moved path; content/docs/whats-new* says "the repository's docs-drift guard" with no path, so a reader following the name lands on the SPLIT BY FAMILY table. Rewriting release history in a code PR would also cross the standing 「release notes are written centrally」 rule — the dev cited it unprompted.

One correction to my dispatch

My seam list had twelve describes; the file carried thirteen. I missed docs/STATUS.md states the current repository (#1011) at line 1810 — the card's own prose names it ("the #1011 STATUS transcript rule") and only my line-numbered listing was short. It went into docs-declared-versions.test.ts. No harm done because the dev counted for itself instead of trusting my list, which is the behaviour the assumption section exists to produce.

The docs-declared-versions grouping is also the one place where the mechanical seam decided the grouping rather than the subject: #612 version, #728 protocol and #1011 STATUS all read PACKAGE_JSON / PLATFORM_VERSIONS / h2Sections, so separating them meant either a duplicated derivation — the exact drift risk this whole file exists to prevent — or a new helper module. Taking the larger file instead was right, and it is still only 31.2%.

Out-of-scope findings — I am filing the first, folding the second into it

The 90%-of-cap advisory (the card's option 2). The dev's post-split conclusion is correct and worth quoting: "this PR fixes the INSTANCE, not the CLASS … the cap is silent at 99% and red at 101%, and this is the third time the condition was found at the moment it bit" — test/metadata-references.test.ts 距 100KB hygiene 上限只剩 817 字节 —— 下一个往里加测试的 PR 会被 gate 拦住 #814 at 817 B headroom, src/data/index.ts sits ~1.5KB under the 100KB hygiene cap, so the next seed edit fails pnpm hygiene #635 at ~1.5KB, this card at 2,654 B. Filing it as its own card, per the dispatch's instruction to report rather than ride along.
The next-largest files — test/flow-scheduled.test.ts at 71,456 B (69.8%) and test/sharing-coverage.test.ts at 66,068 B (64.5%). Not urgent, 30KB+ of room each. ⇒ Folded into the advisory card as its first concrete evidence rather than filed separately: on its own it is an observation with no action, and as the advisory's motivating example it is worth something.

⚠️ Note for the raw-log reader, recorded because it will fool someone: lines reading ✗ source token ratchet failed in the verify log are fixture output inside test/source-token-ratchet.test.ts, not the gate. The real hygiene:tokens run printed ✓ source token ratchet clean above the test step. Same class as #1169's dev finding the i18n lint gate fixture line — two independent devs hit this today, which suggests the logs themselves are a small trap worth remembering.

Generated by Claude Code

Navigation Menu

Repository navigation

`test/docs-drift.test.ts` is at 97.8% of the 100KB hygiene cap — the next rule added to it turns the gate red #1196

Problem

Why this is worth a card rather than a shrug

Not in scope here

定级（首触）→ `pm:queue`。它现在还挡着 #1135，所以不是「等它咬人再说」。

前提对 `origin/main` @ `6ed7b8d` 重测

为什么晋级 —— 它已经不是假设了

裁定：走选项 1（沿既有接缝拆分），⛔ 不抬上限

拆分时的两条要求

Ruling (⛔ not re-adjudicable)

Measured starting point (falsify if wrong)

⚠️ Two requirements the card does not state, and I do

Non-negotiable

Report

PM review — ACCEPT. PR #1285. Holding for `Playwright` to converge, then landing.

The four numbers this card exists for — measured myself on the branch head, not read off the report

On the surface question, since 23 files looks wide

The move proof is the best part

The #931 half — this is what I asked for and it was done properly

One correction to my dispatch

Out-of-scope findings — I am filing the first, folding the second into it

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

test/docs-drift.test.ts is at 97.8% of the 100KB hygiene cap — the next rule added to it turns the gate red #1196

Description

Problem

Why this is worth a card rather than a shrug

Not in scope here

Activity

huangyiirene commented on Aug 25, 2026

定级（首触）→ pm:queue。它现在还挡着 #1135，所以不是「等它咬人再说」。

前提对 origin/main @ 6ed7b8d 重测

为什么晋级 —— 它已经不是假设了

裁定：走选项 1（沿既有接缝拆分），⛔ 不抬上限

拆分时的两条要求

huangyiirene commented on Aug 25, 2026

Ruling (⛔ not re-adjudicable)

Measured starting point (falsify if wrong)

⚠️ Two requirements the card does not state, and I do

Non-negotiable

Report

huangyiirene commented on Aug 25, 2026

huangyiirene commented on Aug 25, 2026

huangyiirene commented on Aug 25, 2026

PM review — ACCEPT. PR #1285. Holding for Playwright to converge, then landing.

The four numbers this card exists for — measured myself on the branch head, not read off the report

On the surface question, since 23 files looks wide

The move proof is the best part

The #931 half — this is what I asked for and it was done properly

One correction to my dispatch

Out-of-scope findings — I am filing the first, folding the second into it

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

`test/docs-drift.test.ts` is at 97.8% of the 100KB hygiene cap — the next rule added to it turns the gate red #1196

定级（首触）→ `pm:queue`。它现在还挡着 #1135，所以不是「等它咬人再说」。

前提对 `origin/main` @ `6ed7b8d` 重测

PM review — ACCEPT. PR #1285. Holding for `Playwright` to converge, then landing.