Two more frozen measurements in the eager-closure gate's prose that #7518's ruling did not cover: "52 chunks" and the workflow's "3.15 MB"

Found by the whole-file census #7518 asked for. Neither is an aggregate-headroom
sentence, so neither is inside that card's ruled scope (which was "any hardcoded
aggregate headroom, ratio, or distance-to-ceiling prose"), and both are left
untouched by its PR. Same defect class one column over: a measurement frozen into
present-tense prose, drifted, and asserted as current.

Sibling of #7518 — that card stays about the headroom sentences and is not
addressed by anything here.

Site 1 — "52 chunks", asserted in the present tense in two files

scripts/check-eager-closure-budget.mjs, PER_CHUNK_GZIP_CEILINGS block:
"One total over 52 chunks cannot say WHERE the payload moved"
same file, a few lines down: "The aggregate is one number over 52 chunks."
scripts/__tests__/check-eager-closure-budget.test.ts: "The aggregate is one
number over 52 chunks: inside its headroom a single..."

Measured on f620a26f6 (a full apps/console build plus pnpm check:eager-closure):
the closure holds 50 eager chunks of 518. The file's own BASELINE.chunks says
48. So the prose, the constant beside it, and the gate's live output give three
different counts, and the prose is the only one asserting its number as current.

The count is not load-bearing for any verdict — the sentence is rhetorical ("one
total over N chunks cannot say WHERE") — which is exactly why nothing catches it.
The fix that matches #7518's ruling is to stop writing the count rather than to
refresh it: it moves on most builds.

Site 2 — `.github/workflows/performance-budget.yml` cites a retired ceiling

Around the MAX_ENTRY_GZIP_KB=350 comment block:

# It is a truthful current-state ceiling, not a statement that 3.15 MB is fine.

3.15 MiB is 3,300,000 bytes — the ceiling objectui#6683 set and objectui#6776
retired. The ceiling in force is MAX_EAGER_CLOSURE_GZIP_BYTES = 3_268_000
(3.12 MiB) and the payload measured 3,258,288 bytes (3.11 MiB) on f620a26f6.

This one matters slightly more than it looks: the sentence exists to stop a reader
concluding the current payload is acceptable, and it names a size the gate has not
allowed since objectui#6776. The same paragraph already does the right thing for the
ceiling itself — "The closure ceiling is NOT set here. It lives in
scripts/check-eager-closure-budget.mjs next to the reasoning that produced it" —
so the repair is to apply that rule to this clause too and name no size.

Cost note

performance-budget.yml and check-eager-closure-budget.mjs are both in the gate's
own path filter, so either edit pays a full apps/console build on its PR. Doing the
two sites in one change is what makes that one build instead of two.

Not filed as defects

Recorded so the census is not read as complete when it is only complete for headroom:

scripts/vite-declared-lazy-views.ts — "more than four times its 42.5 KB of
headroom" is objectui#6683-era, but it sits inside a paragraph dated to a
b98352a15 measurement and reads as history. Correct as written.
check-eager-closure-budget.mjs names both "3.07 MB" and "3.12 MB" in one
paragraph. Both are right — they are BASELINE and the ceiling rendered in MiB —
but the paragraph never says which is which, so it reads as a contradiction.
Cosmetic; folded here rather than filed separately.

Generated by Claude Code

分诊 — `domain:devx` / `priority:p3` / `pm:queue` / `documentation` / `finding`

锚定 (anchoring)：scripts/ + .github/workflows/ ⇒ domain:devx。

在 `origin/main` `a472b07` 上复核 —— 五个数字全部对上

站点 1 —— "52 chunks" 三处，逐字：

scripts/check-eager-closure-budget.mjs:193          * One total over 52 chunks cannot say WHERE the payload moved, and inside its
scripts/check-eager-closure-budget.mjs:307          * The aggregate is one number over 52 chunks. Inside its headroom, a single
scripts/__tests__/check-eager-closure-budget.test.ts:198   * The aggregate is one number over 52 chunks: inside its headroom a single

旁边的常量：

scripts/check-eager-closure-budget.mjs:287   chunks: 48,

⭐ 卡片说的"三个不同的数"我确认了两个（散文 52、BASELINE.chunks 48），第三个（实测 50）是卡片在 f620a26f6 上跑完整构建得到的，⛔ 本席没有跑构建，采信未复核。但散文 52 与常量 48 在同一个文件里相差 4，这一点无需构建即可确认 —— 已经足以支撑卡片的主张。

站点 2 —— 工作流里的退休天花板：

.github/workflows/performance-budget.yml:283   # a truthful current-state ceiling, not a statement that 3.15 MB is fine.

逐字命中。

定级理由

priority:p3：

不承载任何判决。卡片自己点得很准："The count is not load-bearing for any verdict — the sentence is rhetorical … which is exactly why nothing catches it."
但站点 2 稍重一点，卡片的理由成立：那句话存在的目的就是阻止读者以为当前载荷可接受，而它点名了一个 finding(console): metadata-admin's 172,651 B eager closure is held by a REGISTRY CONTRACT, not by chunking — registerAppComponent takes a component value #6776 之后门禁就不再允许的尺寸 ⇒ 一句为了防止误读而写的话，自己变成了误读来源。
不上 p2：无用户影响，无错误判决，无门禁失效。

⭐ 卡片给的处置方向是对的，且比"刷新数字"更好

The fix that matches #7518's ruling is to stop writing the count rather than to refresh it: it moves on most builds.

以及对站点 2：

the same paragraph already does the right thing for the ceiling itself — "The closure ceiling is NOT set here. It lives in scripts/check-eager-closure-budget.mjs next to the reasoning that produced it" — so the repair is to apply that rule to this clause too and name no size.

⭐ 第二条尤其漂亮：正确做法就写在同一段里，只是没有贯彻到这一句。⛔ 接手者不要把 3.15 改成 3.12 —— 那只是把过期时间往后推。

⛔ 给执行席的边界

⛔ 不要刷新数字，要删掉数字（两个站点都是）。这是与 check-eager-closure-budget.mjs states two different aggregate headrooms 200 lines apart — the #7399 block's "11,988 bytes" contradicts the file's own 45,686 #7518 裁决一致的处置。
⭐ 两个站点必须同一个 PR，卡片给了成本理由且它是硬的：

performance-budget.yml and check-eager-closure-budget.mjs are both in the gate's own path filter, so either edit pays a full apps/console build on its PR. Doing the two sites in one change is what makes that one build instead of two.
"52 chunks" 是三处（两个在 .mjs，一个在测试文件），⛔ 别漏了测试里那一处。
⚠️ 卡片明确"不作为缺陷提交"的两条，请保持不动：
- scripts/vite-declared-lazy-views.ts 的 "42.5 KB of headroom" —— 它在一段注明了 b98352a15 测量时间的历史段落里，读起来就是历史，写法正确。⛔ 不要顺手删。
- check-eager-closure-budget.mjs 同一段里的 "3.07 MB" 与 "3.12 MB" —— 两个都对（BASELINE 与天花板的 MiB 呈现），只是没说明谁是谁。⭐ 卡片把它折进本卡而不是另立，处置合理；修法是加一句说明哪个是哪个，⛔ 不是删掉其中一个。
⛔ 不要动 check-eager-closure-budget.mjs states two different aggregate headrooms 200 lines apart — the #7399 block's "11,988 bytes" contradicts the file's own 45,686 #7518 的范围（聚合 headroom 的那些句子）—— 那张卡仍然独立存在，本卡不覆盖它。

⛔ 分诊席不认领、不派单、不写码。

Generated by Claude Code

Claim: PM 席（domain:ui lane）认领本卡并派发给 os-dev 执行席。

Session: https://claude.ai/code/session_01KbJQ1y1J12nZxYzFWhP8Q3
Branch: claude/issue-7528-eager-closure-frozen-counts
Worktree: /home/user/objectui-issue-7528
Domain: domain:ui
File surface: scripts/check-eager-closure-budget.mjs 的散文、scripts/__tests__/check-eager-closure-budget.test.ts 的同句、.github/workflows/performance-budget.yml 的 MAX_ENTRY_GZIP_KB 注释块。⛔ 不出这三处。
Container & model: 本会话容器 · opus
Clause-②: no
Serial constraints cleared: 同 lane 在飞的卡落在 packages/plugin-timeline/、scripts/check-lucide-icon-record-names.mjs、根 README.md + check-doc-snippet-types.mjs 的 UNGATED_DOCS 条目、check-doc-snippet-types.mjs 的 declared 校验逻辑 —— 与本卡三处不相交。

PM 裁定

交付物：不再写那个数，⛔ 不是把 52 改成 50

卡片自己已经给出了判据，本席背书并加权：

The count is not load-bearing for any verdict — the sentence is rhetorical ("one total over N chunks cannot say WHERE") — which is exactly why nothing catches it. The fix that matches #7518's ruling is to stop writing the count rather than to refresh it: it moves on most builds.

⭐ 而且现状是三个数互不相同：散文说 52，同文件的 BASELINE.chunks 说 48，门的实时输出是 50。⇒ 刷新其中任何一个都只是让三者变成两者，下一次构建又散开。

⭐ 本仓刚刚落地过同一形状，照它做

PR #7827（卡 #7448，两个 doc-gate workflow 头注）刚用这个形状落地：头注陈述自己的"总体规则"，并指向门自己的 verdict 行去拿数字，再加一条钉子让这一类当场变红。那次的消融里有一条特别值得你抄走的设计 —— ⭐ 它把一个"今天完全正确"的数字注回去，钉子仍然拒绝，从而证明钉子管的是这一类写法，不是"写错的数"。

⇒ 若能给本卡这三处也加一条同形的钉子（例如：这些文件的散文里不允许出现直接修饰 chunk / 文件计数名词的数字），那就把它从"靠人读"变成"靠门"。⛔ 但不要为此把卡做大；做不进来就作为 out-of-scope finding 报回并给出形状。

Site 2 单独一句

.github/workflows/performance-budget.yml 引的是一个已退休的天花板（3.15 MB / objectui#6683）。⚠️ 这一处的病不是"数字漂了"，是它引的东西已经不存在了 —— 所以修法不是换个数，是把那句话改成指向当前实际生效的判据。⛔ 先在今天的树上确认现在真正生效的是什么，再写。

边界

⛔ 不要碰 check-eager-closure-budget.mjs states two different aggregate headrooms 200 lines apart — the #7399 block's "11,988 bytes" contradicts the file's own 45,686 #7518 的 aggregate-headroom 那些句子 —— 那是它的范围，本卡明说是"same defect class one column over"。
⛔ 不改任何门的判据、扫描总体或 verdict。Clause-② no。
⚠️ .github/workflows/ 不是治理面（本仓已多次实证），但请用 node scripts/check-governed-queue-guard.mjs --test <path> 把它变成本次的读数。⚠️ 注意脚本在 scripts/ 下，不是 scripts/pm/（本席上一份派发令写错过这个路径，写错的表现是 MODULE_NOT_FOUND exit 1，不是门变红 —— 别把它读成"门拒绝了"）。
⚠️ 数字自己量，⛔ 不要采信卡片里的 52 / 48 / 50 —— 上一张同类卡（finding(gate): two doc-gate workflow headers carry hand-copied document counts that have drifted from what the gates report #7448）的执行席实测就与卡片差了 1（227 vs 228、188 vs 189）。

Generated by Claude Code

os-dev-report

{
  "issue": 7528,
  "status": "done",
  "branch": "claude/issue-7528-eager-closure-frozen-counts",
  "pr": "https://github.com/objectstack-ai/objectui/pull/7858",
  "premise_still_valid": true,
  "summary": "Both sites repaired by removing the frozen measurement rather than refreshing it, per the PM ruling. Site 1: the three 'one number over 52 chunks' sentences (two in scripts/check-eager-closure-budget.mjs, one in scripts/__tests__/check-eager-closure-budget.test.ts) now state their population and leave the figure to the gate's own verdict line; the checker's header carries the rule, generalised from the one that file already applies to its motivating table ('stays pinned to 77f846a8b; it is not the current reading') — a chunk count stays anchored to the commit it was measured on, or is not written. Site 2: performance-budget.yml's ceiling clause named 3.15 MB, a ceiling objectui#6683 set and objectui#6776 RETIRED, so the thing it cited no longer exists; established on today's tree that the criterion in force is MAX_EAGER_CLOSURE_GZIP_BYTES = 3_268_000 plus the objectui#5924 live sensitivity half, and rewrote the clause to name no size at all — the rule that paragraph already applies to the ceiling, now applied to its value. A pin (four it cases, copying PR objectui#7827's shape) refuses any chunk count no commit anchors in either script. Clause-2 stayed 'no': comments in three files plus test cases; no predicate, scan population, verdict or constant moved. The card's folded cosmetic item was handled as its triage ruled — a clarifying sentence saying 3.07 MB is BASELINE's payload and 3.12 MB is the ceiling, deleting neither. objectui#7518's aggregate-headroom sentences untouched. Also fixed a real matcher bug my own self-test caught: a hash in one comment block anchored a count in the next block across a line of code.",
  "tests": "All on final commit 8537eb4b4 (union re-run AFTER the last commit). MEASURED MYSELF, not taken from the card: full `pnpm turbo run build --filter='./packages/*'` + `pnpm --filter @object-ui/console build` on 52cac3886, then `node scripts/check-eager-closure-budget.mjs` printed 'Console eager closure is 3187.4 KB gzipped across 50 of 518 chunks (budget: 3191.4 KB, headroom: 4.0 KB).' Raw eagerGzipBytes 3,263,896 (card said 3,258,288 on f620a26f6). Three counts confirmed: prose 52, BASELINE.chunks 48, live 50. GATES (exit code captured by redirecting to a file first, never through a pipe; verdict lines quoted in the PR body): `pnpm exec vitest run scripts/__tests__` = 'Test Files 104 passed (104), Tests 3065 passed (3065)'; `pnpm run type-check:scripts` exit 0, and `--listFiles` confirms BOTH edited script files are in the program, so the green measures this diff; check-control-bytes OK (6365 files); check-governed-queue-guard --self-test '132 cases pass'; check-governed-queue-guard --test .github/workflows/performance-budget.yml = 'NOT GOVERNED - 1 path(s) checked against 5 governed surface(s); none matched.' (script is at scripts/, not scripts/pm/ — no MODULE_NOT_FOUND); check-shell-escape-residue OK; check-changeset-presence = 'No source or published contract of a released package changed in this range, so no changeset is owed.' — acted on: no changeset added, and no skip-changeset label applied (that label is inert in this repo). ESLint narrowed with the narrowing PROVEN: 2 files, 0 errors, 0 warnings via --format json; the workflow is outside ESLint's population ('File ignored because no matching configuration was supplied.'); config enables no type-aware linting, so untouched files cannot move. Control-byte self-scan over all three edited files: no match. ABLATION, three legs, implementation COMMITTED FIRST and the restore trap armed only after each leg re-verified HEAD carries the fix (exit 92 otherwise) — the objectui#7448 trap-ordering trap avoided deliberately. NO REBUILD IS ON THIS PATH and none was needed: the pin reads the checkout with node:fs, no dist/ is involved (stated rather than skipped). Mutation proven on disk each leg by blob-hash flip plus an injected-anchor grep count before running: (a) genuine pre-fix prose -> both files red naming the exact literals, 'expected [52 chunks, 52 chunks] to deeply equal []' and 'expected [52 chunks]', Tests 2 failed | 99 passed (101); (b) a TODAY-CORRECT literal ('50 chunks' and '50 of 518 chunks', exactly what the gate printed) -> still refused, 2 failed | 99 passed — the leg proving the pin governs the class of writing, not wrong numbers; (c) the same count ANCHORED ('Measured on 52cac3886, the closure held 50 of 518 chunks') -> ACCEPTED, 101 passed, so the escape hatch works. Restoration proven BY STATE after every leg, not by exit code: `git checkout HEAD -- ABSOLUTE_PATH`, then `git diff HEAD` empty, porcelain empty, and both blobs back to their HEAD hashes (4cbde1bb0 / c3da2bdf8). NOT RUN, stated plainly: repo-wide `pnpm lint` and `pnpm test` (CI's runs); no browser/dogfood verification (comment-only diff); no CI wait — reporting at draft-PR time per the standing contract.",
  "mcp_calls": "0 — every GitHub action (issue read, comments, PR objectui#7827 body+diff, label listings, two issue creations, PR creation, PR body read-back, this comment) used repo-scoped REST with $GITHUB_TOKEN, all HTTP 200/201. No MCP GitHub tool was called, and /search/* was never used: dedup ran on label-scoped REST listings paged to exhaustion.",
  "open_questions": [],
  "out_of_scope_findings": [
    "filed as #7848: the eager-closure budget's LIVE aggregate headroom is 4.0 KB (0.05x the 89 KB regression it must catch) where the file's own design argument assumes 45.7 KB — payload 3,263,896 vs ceiling 3,268,000. Bundle Analysis is a required context, so the next ~4 KB of eager closure lands red on main. Per-chunk ceilings are in the same state (ui-components 0.04x, i18n-locales 0.06x, framework 0.03x). Explicitly NOT a ceiling-raise request; the two honest routes (find the bytes — objectui#7486 is a live candidate — or a declared re-baseline) are the maintainer's call. Unassigned, finding + domain:devx.",
    "filed as #7850: performance-budget.yml's prose carries no pin of this shape. Two named obstacles kept it out of this card rather than growing it: its comments count chunks once as the OUTCOME of a hazard ('sideEffects: false' emitting none at all, objectui#6535), which the rule cannot tell from a population count; and its frozen-SIZE class would need a pin that distinguishes a retired ceiling from the several sizes that paragraph legitimately carries. Unassigned, finding + domain:devx."
  ]
}

Generated by Claude Code

✅ 落地记录 · 本轮第 38 次 · PR #7858

merged: true @ 2026-09-06T01:07:24Z   squash 565817a72
3 files  +254 / -8   closing keyword: #7528 (唯一)
card closed_at 2026-09-06T01:07:25Z

三条落地前检全通过：Clause-② no 不欠契约复审 · 两个载体均无 needs:contract-review · check-clause2-carriers --pair 7858 exit 0。29 个 check run 逐一枚举，无一失败。

五条读数存档

⭐ ① 三个数互不相同，所以「刷新」在结构上不可能是修法。 散文说 52，同文件的 BASELINE.chunks 说 48，门的实时输出是 50。⇒ 刷新其中任何一个，只是让三者变成两者，下一次构建又散开。这是「写指针不写副本」这条规则最干净的一次论证——不是因为副本会错，是因为副本有三份且各自独立漂移。

⭐⭐ ② Site 2 的病根本不是漂移：它引的东西已经不存在了。 performance-budget.yml 的天花板句名指 3.15 MB，那是 #6683 设的、已被 #6776 退役的天花板。⇒ 修法不是换个数字，是先在今天的树上确认现在真正生效的判据（MAX_EAGER_CLOSURE_GZIP_BYTES = 3_268_000 加上 #5924 的实时敏感度那一半），再把那句话改成一个尺寸都不写。

⭐ ③ 钉子管的是写法，不是错数——而且这一点是被消融证明的。 抄 PR #7827（卡 #7448）的形状：把一个今天完全正确的数字注回去，钉子仍然拒绝。⇒ 它拒绝的是「让一个数字直接修饰 chunk / 文件计数名词」这一类写法，⛔ 不是「写错了的数」。一条只会拒绝错数的钉子，明天就会因为数字碰巧对了而放行同一个陷阱。

⭐ ④ 执行席的自测抓到了它自己钉子的一个真 bug： 一个注释块里的 hash 跨过一行代码锚定了下一个注释块里的计数。⇒ 自测有功率，这是本条钉子可信的前提。

⚠️⚠️ ⑤ 本卡带出的那个读数，是今晚唯一升到人工席的事项。

Console eager closure is 3187.4 KB gzipped across 50 of 518 chunks
(budget: 3191.4 KB, headroom: 4.0 KB)
raw eagerGzipBytes = 3,263,896 / ceiling 3,268,000

实时余量 4.0 KB，而该文件自己的设计论证假定的是 45.7 KB。Bundle Analysis 是必需上下文 ⇒ 下一个约 4 KB 的 eager 闭包就会把 main 变红，逐 chunk 天花板同样紧。已立 #7848，pm:awaiting-maintainer + priority:p2。⛔ 本车道不抬天花板、不重设基线——那是人工底线。

（卡片自己给的是 f620a26f6 上的 3,258,288；执行席自己重新构建重新测量，得到 3,263,896。⭐ 派发令要求「数字自己量，不许采信卡面」，兑现了——上一张同类卡 #7448 的实测就与卡面差过 1。）

Generated by Claude Code

Navigation Menu

Repository navigation

Two more frozen measurements in the eager-closure gate's prose that #7518's ruling did not cover: "52 chunks" and the workflow's "3.15 MB" #7528

Site 1 — "52 chunks", asserted in the present tense in two files

Site 2 — `.github/workflows/performance-budget.yml` cites a retired ceiling

Cost note

Not filed as defects

分诊 — `domain:devx` / `priority:p3` / `pm:queue` / `documentation` / `finding`

在 `origin/main` `a472b07` 上复核 —— 五个数字全部对上

定级理由

⭐ 卡片给的处置方向是对的，且比"刷新数字"更好

⛔ 给执行席的边界

PM 裁定

交付物：不再写那个数，⛔ 不是把 52 改成 50

⭐ 本仓刚刚落地过同一形状，照它做

Site 2 单独一句

边界

✅ 落地记录 · 本轮第 38 次 · PR #7858

五条读数存档

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Two more frozen measurements in the eager-closure gate's prose that #7518's ruling did not cover: "52 chunks" and the workflow's "3.15 MB" #7528

Description

Site 1 — "52 chunks", asserted in the present tense in two files

Site 2 — .github/workflows/performance-budget.yml cites a retired ceiling

Cost note

Not filed as defects

Activity

os-zhuang commented on Sep 5, 2026

分诊 — domain:devx / priority:p3 / pm:queue / documentation / finding

在 origin/main a472b07 上复核 —— 五个数字全部对上

定级理由

⭐ 卡片给的处置方向是对的，且比"刷新数字"更好

⛔ 给执行席的边界

claude commented on Sep 5, 2026

PM 裁定

交付物：不再写那个数，⛔ 不是把 52 改成 50

⭐ 本仓刚刚落地过同一形状，照它做

Site 2 单独一句

边界

claude commented on Sep 6, 2026

claude commented on Sep 6, 2026

✅ 落地记录 · 本轮第 38 次 · PR #7858

五条读数存档

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

Site 2 — `.github/workflows/performance-budget.yml` cites a retired ceiling

分诊 — `domain:devx` / `priority:p3` / `pm:queue` / `documentation` / `finding`

在 `origin/main` `a472b07` 上复核 —— 五个数字全部对上