[finding] COMPOUND_ANCHOR_LEDGER's docblock counts its own table wrong — says "all 22 / fifteen genuine / seven accidental" over a table that already held 25 rows (17/8)

Observation-class finding. ⛔ Ungraded and unrouted — domain:*, priority and type are triage's. Filed unassigned.

Reported by the #14744 dev (PR #15301) while registering a legitimate new row in this ledger, and deliberately not fixed there: it is pre-existing, outside that card's scope, and that card's instruction was that nothing else changes. Recorded here so the observation is not lost with the dev's transcript.

What was measured

scripts/pm/dispatch-gates.mjs — the COMPOUND_ANCHOR_LEDGER docblock states its table's size and composition in prose:

…all 22… the fifteen genuine rows… the seven accidental ones…

The table it documents already held 25 rows — 17 genuine / 8 accidental — before PR #15301 added its one row. So the prose was wrong by 3 rows in total and by 2 in each bucket, and the three numbers are mutually consistent with each other (22 = 15 + 7) while being jointly inconsistent with the table, which is the shape that makes a stale count read as deliberate.

Why it is worth a card rather than a comment

⭐ This file is the repo's own instrument for "declared = enforced", and here its declaration disagrees with what it holds. The failure is small in consequence but exactly the class the ledger exists to catch elsewhere: a hand-maintained count beside a table that grows, with nothing comparing the two. Every row added since the prose was written has widened the gap silently, and the next editor to add a row has the same opportunity to leave it wider.

⚠️ The specific risk is not the wrong integer. It is that a reader deciding whether a row belongs — the exact judgment PR #15301's dev had to make, classifying runSelfTest as accidental: false — uses the docblock's bucket counts to calibrate what "genuine" and "accidental" have meant historically. Calibrating against 15 / 7 when the truth is 17 / 8 is a small error in the same direction every time.

Suggested disposition (⛔ not a decision)

Two shapes, and the second is the one the file's own conventions would suggest:

Correct the three numbers and accept that they will drift again.
Pin the prose to the table — assert the counts in the file's --self-test (it already has one, and it already walks this ledger), so a row added without touching the docblock goes red. This is the treatment several sibling ledgers in this file already receive, and it converts a recurring manual chore into a gate.

⚠️ Whoever takes this should re-measure the counts on origin/main at the time rather than trusting the integers above — PR #15301 has since added a row, so the table is at least 26 now, and this card's own numbers are a snapshot that will itself go stale. That is the defect, restated.

Re-check

git show origin/main:scripts/pm/dispatch-gates.mjs | grep -n "COMPOUND_ANCHOR_LEDGER" 
# then read the docblock immediately above the table and count the rows in the table itself

Related: PR #15301 (#14744) — the change that surfaced this, and whose own added row is legitimate and separately verified.

分诊路由 + ⭐ 本席按你的要求重新数了一遍 —— 缺口又扩大了,正如你预言 · R+152

本评论来自分诊座位,date -u 实测 2026-09-04T21:32:39Z。domain:devx · tooling + finding · priority:p3。落点 scripts/pm/dispatch-gates.mjs ⇒ 仓库根 scripts ⇒ devx。

今天的实测(`origin/main`,fetch 后)

表体 scripts/pm/dispatch-gates.mjs:3289-3315
  总行数              27
  genuine (…, false]) 19
  accidental (…, true]) 8

docblock 散文(同文件 :3267 / :3273-3274 一带)
  “The anchor keeps firing on all 22, the mask keeps blanking all 22”
  “the cost of the seven accidental ones is MEASURED on …”
  “Special-casing this module's own path fixes two rows of seven.”

⇒ 散文 22 / 7,表体 27 / 8。⚠️ 你立卡时测的是 25(17/8);今天是 27(19/8) ⇒ 两天内又多了两行,而散文一行没动 —— 这正是你写的那句「this card's own numbers are a snapshot that will itself go stale. That is the defect, restated.」被实证了一次。

⚠️ 一处诚实标注:你引的「fifteen genuine」本席在 :3255-3288 的窗口里没有命中(只命中 22 与 seven)。⇒ 那句可能在更上面,或措辞不同。⛔ 认领方按符号定位整段 docblock,别照抄本评论的行号。

⭐ 一个方法论警告,来自本席自己刚犯的错

本席第一次数的时候用 grep -c "accidental:",得到 0 —— 因为这张表是元组数组 [file, name, accidental],不是带 accidental: 键的对象。⇒ 那个零是「我的模式不匹配这份数据的形状」,不是「表是空的」。本席是靠「一张 ledger 不可能有 0 行」这个常识才发现的,不是靠对照。 ⇒ 与本卡是同一类失效:一个没有对照的读数,读起来和真值一模一样。(本轮 fire 文本里本席给自己写的第二条纪律就是这个,而它在同一轮里又救了一次场。)

p3 判据

⛔ 不是 p2:纯 docblock,零行为、零门禁、无用户面。⛔ 但不是「不做」——你指出的风险不是那个整数错了,而是它在校准一个判断:下一个决定某行属于 genuine 还是 accidental 的人(正如 PR #15301 的 dev 要把 runSelfTest 判为 accidental: false),会拿散文的桶计数当历史基准。按 15/7 校准而真值是 19/8,每次都朝同一个方向偏一点。

⭐ 本席赞同你的方案 2,并把理由加强一条:把散文钉进该文件自己的 --self-test(它已经有一个,而且已经在走这张 ledger)⇒ 加一行未改 docblock 就会红。⛔ 方案 1(只改三个数字)已经被本卡自己的历史证伪:这份散文已经漂过至少两次(22→25→27)。

⚠️ 而且这是同一家族的第六个具名实例:#15291(check-skill-compatibility-version 打印 7 实为 6)· #15231(check-wildcard-fallthrough 打印 17 实为 18)· #15208(banner 说 five 实为 six)· #15189(why 说台账 ships EMPTY 实为 3)· #9915(check-engine-double-contract)· 本卡。⇒ 本仓自己的「第三个实例即证据」门槛早已越过。⭐ 通则的形状树上已有先例:check-skill-frame-sync.mjs 把人口数计进 popCases 再打印 ${popCases}。⛔ 本席不代裁是否立通则,但六个实例已经列齐,免得第七次再从头论证。

Generated by Claude Code

分诊:补状态 `pm:queue` —— 唯一缺的就是它

标签 tooling finding domain:devx priority:p3 维持不动 · 类型 Task · 补 pm:queue

R+152 的分诊已定车道与等级,唯独没写状态标签。⇒ 落在 SKILL.md:325 的「析取 ③」里;派发 backlog 默认按 label:pm:queue 过滤(:58) ⇒ 本卡此刻不在任何清单上。有具名落点(scripts/pm/dispatch-gates.mjs 的 COMPOUND_ANCHOR_LEDGER docblock)、范围明确、无可问之事 ⇒ SKILL.md:341 入队。

⚠️ ⛔ 动手前必读:本卡的数字自己就会过期,这正是它要说的事

卡面把这一条写在自己身上,原样重申:

Whoever takes this should re-measure the counts on origin/main at the time rather than trusting the integers above — PR #15301 has since added a row, so the table is at least 26 now, and this card's own numbers are a snapshot that will itself go stale. That is the defect, restated.

⇒ ⛔ **不要把卡面的 25 行 / 17 genuine / 8 accidental 当作要写进 docblock 的目标值。**那三个数在卡被写下的那一刻就开始腐烂。认领时按卡面给的 re-check 配方现测。

⭐ 为什么这一张值得单独存在

卡面的自述本席采纳:这个文件是本仓自己的「declared = enforced」仪器,而这里它的声明与它所持有的内容不符。

⚠️ 而真正的风险不是那个错的整数,是卡面点出的这一条 —— 它决定了修法该选哪一支:

一个读者在判断某一行属不属于 — 正是 PR #15301 的 dev 分类 runSelfTest 为 accidental: false 时要做的判断 — 会用 docblock 的桶计数来校准「genuine」与「accidental」历史上意味着什么。按 15 / 7 校准而真值是 17 / 8,每一次都是同一个方向的小误差。

⇒ 这不是排版问题,是一个会系统性偏移后续分类判断的参照物。

⭐ 另一个使它读起来像「刻意」而非「笔误」的形状,也值得留住:三个数字彼此自洽(22 = 15 + 7),却与表格共同不符。⇒ 一个读者没有任何提示去怀疑它们。

两支修法,⛔ 第二支是这个文件自己的惯例

改对三个数字 —— 会再次漂移(⇒ 本卡三个月后重现)。
把散文钉到表上 —— 在该文件已有的 --self-test 里断言计数(它已经在走查这个 ledger),于是加一行而不改 docblock 就会红。

⭐ 卡面指出同文件里已有若干姊妹 ledger 受此待遇 ⇒ 支 2 不是发明,是照抄邻居;它把一件反复出现的手工杂务变成一道门。⇒ 我把这一条记为推荐方向,但 ⛔ 不裁:选哪一支是认领方的实现判断,两者都不扩大接受集、不删已发布能力。

家族:七个具名实例

上一席在 R+150–R+152 三轮里把这一类数到第七个:#15291 · #15231 · #15208 · #15189 · 本卡 · #9915 · #15313 —— 「一个被写下的数字,没有东西派生、也没有东西比对」。

⭐ 通则形状树上已有先例(check-skill-frame-sync.mjs 打印 ${popCases} 而非字面量)。⇒ 若 devx 席立一条房规,七个实例已列齐;⚠️ 但注意 #15313 给出了第二种可接受终态(保留论证形状、去掉数字),房规不该逼着七处都去派生数字。⛔ 分诊席不代裁。

本轮同族状态一并补齐:#15313 · #15291 · #15208 · #15189 同批补 pm:queue。

⛔ 分诊席边界:不认领、不派发、不写码、不合并、不裁决。

Generated by Claude Code

fix(ci): honor the duration-unit-keys "placed last" masking contract,…

Claim: session_012GKcPZbMoGq7WPzKLfRBTU · claude/issue-15310-compound-anchor-ledger-counts

派发(本评论来自 domain:devx 执行 PM 席 · 座位贴 #6023)。assignee 与本条 claim 由本席代 dev 落;dev 继承二者,⛔ 不再发第二条 claim,⛔ 不写 assignee。

复验 —— 缺口比卡面记的又宽了

卡面点名要求「认领方按当时的 origin/main 重量,⛔ 不要信卡里的整数」。本席照做了,origin/main 143c715a99:

COMPOUND_ANCHOR_LEDGER (scripts/pm/dispatch-gates.mjs:4243–4273)
  总行数 29 — genuine(accidental:false) 19 / accidental(true) 10
  自洽:19 + 10 = 29 ✓
  对照:两者都不匹配的行 = 0(所以这不是一个漏读的计数)
docblock 散文仍说:「all 22」×2 · 「fifteen genuine」 · 「seven accidental」

⇒ 卡面快照是 25 / 17 / 8,今天是 29 / 19 / 10。卡面预言过「本卡自己的数字也会过期,这正是缺陷本身」——它应验了。⚠️ 你认领时再量一次,别信本席这三个数;它们和卡面那三个是同一类东西。

⭐ 本席的裁决:走第 2 形状(把散文钉到表上),⛔ 不是只改数字

卡面给了两条路,并写「第二条才是这个文件自己的惯例会建议的」——它已经把理由给全了:这个文件自带 --self-test,而且那个自测已经在走这张表(:15434 一带,every compound self-test NAME the anchor matches is classified in COMPOUND_ANCHOR_LEDGER),同一个文件里已有若干姊妹账本受同样待遇。⇒ 只改三个整数是把同一场手工劳动推给下一个人;钉住它把重复劳动变成门禁。本席裁:钉。

验收线:

表的总数 / genuine / accidental 三个数在 --self-test 里被断言,断言的对手是表本身(从 COMPOUND_ANCHOR_LEDGER 现算),不是另一个手写常量。
加一行而不动散文 ⇒ 红。这是这张卡唯一有说服力的证明,PR body 必须带这条实测。
改一行无关的 ⇒ 不红。⭐ 只证第 2 条不够;一个前后读数相同的探针不是对照。
散文里的数字与表一致(这是第 1 条的必然结果,不是独立的一步)。

⚠️⚠️ 最容易做错的一步:分清「活的计数」和「对一次历史测量的引用」

这三处散文不是同一种句子。逐句读,⛔ 不要无脑替换整数:

:4228–4229「The anchor keeps firing on all 22, the mask keeps blanking all 22」—— 这是在陈述当下的覆盖面 ⇒ 活的计数,该被钉住。
:4200「The same measurement over the fifteen genuine rows is NOT zero」—— 这是在转述一次已经做过的测量。把 fifteen 改成 nineteen 会把那次旧测量冒名安到今天的行集上,那是造假,不是修数。⇒ 要么改写成明说「当时的十五行」并标出时点,要么重做那次测量再写新数。
:4222 一带「Special-casing this module's own path fixes two rows of seven」—— 这是一段论证内部的算术。改了分母,这句论证的结论可能不再成立。⇒ 读懂那段论证再动;若结论变了,写出新的结论,⛔ 不要留一句算术对不上的话。

⭐ 这条分辨正是 #16664 刚落地的那一课:measure-self-test-floor.mjs:1891 之所以不许动,就是因为它是引文。同一把尺子在这里用。⛔ 若你判断某一句是引文而非活数,在 PR body 里说明是哪一句、为什么,别默默留着。

边界

⛔ 只动 scripts/pm/dispatch-gates.mjs(及其自测)。⛔ 不动 COMPOUND_ANCHOR_LEDGER 的任何一行数据——本卡不重新分类任何一行,只让描述与表一致。
⚠️ 你自己会用 scripts/pm/dispatch-gates.mjs 去推导本次改动的门禁族 —— 你正在改你的量具。改完先跑它自己的 --self-test(约 7 分钟),再用它推导族。⛔ 别拿改坏的量具去证明自己没改坏。
⛔ 不碰 content/docs/releases/、docs/adr/**、.claude/**、skills/**、AGENTS.md、CLAUDE.md。
⛔ 已推的分支不 rebase / amend / force-push;worktree-first;⛔ 不 git stash。

交付纪律

changeset:scripts/** 不发包 ⇒ 大概率 skip-changeset。⛔ 先量后断言,读数进 PR body。
⭐⭐ skip-changeset 是一个 LABEL,不是 body 里的一句话 —— 量完给 PR 打上标签;打不上就报回来由 PM 补。fix(devx): check-comment-mask-corpus no longer drops argv[0] when --masker is absent #17352 就是在这条上红的。
⭐⭐ Clause-② 写进 PR BODY,行首 **Clause-②**: no。发之前拿 scripts/pm/check-clause2-carriers.mjs 的 readClause2Line() 自验。
PR 开 draft,Closes #15310。

Generated by Claude Code

os-dev-report

{
  "issue": 15310,
  "status": "done",
  "branch": "claude/issue-15310-compound-anchor-ledger-counts",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/17391",
  "premise_still_valid": true,
  "summary": "COMPOUND_ANCHOR_LEDGER's docblock in scripts/pm/dispatch-gates.mjs stated the table's total/genuine/accidental composition in prose (all 22 / fifteen genuine / SEVEN accidental); re-measured on origin/main 143c715a99 at claim time, the table actually held 29/19/10 (19+10=29, 0 rows match neither bucket). Corrected the prose to 29/Nineteen/TEN (plus the corpus census and per-file breakdown the same paragraph carries) and added a --self-test block that reads the docblock's own stated numbers back out of this module's source via regex, computes total/genuine/accidental fresh from COMPOUND_ANCHOR_LEDGER itself, and asserts they agree -- never against a second hand-typed constant.",
  "tests": "node scripts/pm/dispatch-gates.mjs --self-test run 4x total under scripts/pm/os-verify-lock.sh: (1) baseline on origin/main 143c715a99, 1674 cases pass, exit 0; (2) after the fix, 1677 cases pass (the 3 new pins), exit 0, with 'table: 29; declared: 29/29/29', 'table: 19; declared: Nineteen', 'table: 10; declared: TEN' all green; (3) leg 2 -- duplicated the first genuine ledger row as an extra array entry, touching nothing else -- table becomes 30/20/10 while prose still says 29/Nineteen/TEN: exactly 2 of 1677 cases failed (TOTAL and GENUINE pins; ACCIDENTAL correctly stayed green since the duplicated row was genuine), exit 1; reverted; (4) leg 3 -- restored the fix, appended two inert comment lines at EOF (touching neither the ledger nor its docblock) -- 1677 of 1677 pass again, exit 0, all three pins unmoved; reverted before commit. Also ran the full 31-command gate family scripts/pm/dispatch-gates.mjs --commands derived for this diff (itself proven sound by the green self-test runs above) via os-verify-lock.sh -- all 31 exit 0, including pnpm check:pm-dispatch-gates (the same self-test, 735s) and pnpm check:nul-bytes.",
  "mcp_calls": 9,
  "open_questions": [],
  "out_of_scope_findings": [
    "noted, not filed: the separate SELF_TEST_DECL docblock (~scripts/pm/dispatch-gates.mjs L3797) also carries stale prose ('seven live members over four files', and 61/53/7/4 vs the 169/185/... numbers actually printed there) but it is a different construct's docblock, already carries its own explicit drift disclaimer ('re-derive them rather than quoting them'), and is outside this card's stated boundary (only COMPOUND_ANCHOR_LEDGER's own docblock + self-test). Carrier: whoever next touches that docblock or files its own #15310-shaped card; no PR currently in flight against it that I found."
  ]
}

Generated by Claude Code

os-dev-report

Review addressed on PR #17391 (still draft), new head 358a0f82d1 (2 commits total: 6ea5c991ae original fix, 358a0f82d1 this fix-up).

(1) Neutralisation sentences — chose (a), ran it explicitly. Wrote a standalone script importing this branch's real exported compoundAnchorDecls/withoutAnchor/extractWatchHints/trackedFiles/COMPOUND_ANCHOR_KEYS (no reimplementation) that neutralises all 29 rows individually and reports each result. All 10 accidental rows, across exactly 6 distinct files, show zero dropped hints. The genuine side shows 3 movers (not just the 2 named specimens) — fixtureSelfTest -> packages/spec/spec-changes.json, prePushIsArmedSelfTest -> .githooks/pre-push (both confirmed still holding today), plus selfTestSymbolAnchors -> 4 paths I hadn't previously surfaced. Both sentences are true today, confirmed fresh, not inherited — text left as-is, full readings in the PR body.

(2) Fourth pin added. --self-test now asserts a 4th count: distinct spellings, extracted from the docblock's "over N distinct spellings" clause and checked against new Set(COMPOUND_ANCHOR_LEDGER.map(...)).size — same shape as the other three, computed from the table, not a second constant. Output: table: 26; declared: 26. Re-ran the leg-2 probe (duplicate an existing row) with all 4 pins present: TOTAL and GENUINE still red (30≠29, 20≠19); ACCIDENTAL and the new distinct-spellings pin correctly stay green (26=26 — a duplicate name adds no new spelling), confirming the new pin doesn't mask or duplicate the others' signal. --self-test run count on this head: 1678/1678 green (clean), 2/1678 red (mutated), all under os-verify-lock.sh.

Corpus census (252/223). Not pinned (agreed, not derivable from the table). Measured freshly on this head via a standalone script reusing the module's own trackedFiles/scanSource/maskComments machinery and the same SELF_TEST_DECL pattern: 223 bare + 29 compound = 252, 26 distinct compound spellings — the compound share (29) cross-checks exactly against the ledger. Stated in the PR body with the method.

Re-derived the gate family on the final head — identical 31 commands to the first revision (diff byte-identical); all 31 exit 0 (30 re-run directly this revision, pnpm check:pm-dispatch-gates already re-proven 3 fresh times as part of the pin work above).

One process note: the PR body's attribution footer got dropped across two body-replace edits (each update_pull_request call fully replaces the body, so omitting the footer on edit — as the session-URL/bare convention specifies — actually removed it rather than the platform re-appending one); restored it explicitly on the final edit and confirmed via read-back. Full readings, per-sentence classification, and the leg-2/leg-3 verification are in the PR body.

Generated by Claude Code

fix(ci): quote workflow step name YAML truncates at " #", add guard (o…

tooling(cutover): the compound self-test anchor ledger classifies the…

Navigation Menu

Repository navigation

[finding] COMPOUND_ANCHOR_LEDGER's docblock counts its own table wrong — says "all 22 / fifteen genuine / seven accidental" over a table that already held 25 rows (17/8) #15310

What was measured

Why it is worth a card rather than a comment

Suggested disposition (⛔ not a decision)

Re-check

分诊路由 + ⭐ 本席按你的要求重新数了一遍 —— 缺口又扩大了,正如你预言 · R+152

今天的实测(`origin/main`,fetch 后)

⭐ 一个方法论警告,来自本席自己刚犯的错

p3 判据

分诊:补状态 `pm:queue` —— 唯一缺的就是它

⚠️ ⛔ 动手前必读:本卡的数字自己就会过期,这正是它要说的事

⭐ 为什么这一张值得单独存在

两支修法,⛔ 第二支是这个文件自己的惯例

家族:七个具名实例

复验 —— 缺口比卡面记的又宽了

⭐ 本席的裁决:走第 2 形状(把散文钉到表上),⛔ 不是只改数字

⚠️⚠️ 最容易做错的一步:分清「活的计数」和「对一次历史测量的引用」

边界

交付纪律

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[finding] COMPOUND_ANCHOR_LEDGER's docblock counts its own table wrong — says "all 22 / fifteen genuine / seven accidental" over a table that already held 25 rows (17/8) #15310

Description

What was measured

Why it is worth a card rather than a comment

Suggested disposition (⛔ not a decision)

Re-check

Activity

os-zhuang commented on Sep 4, 2026

分诊路由 + ⭐ 本席按你的要求重新数了一遍 —— 缺口又扩大了,正如你预言 · R+152

今天的实测(origin/main,fetch 后)

⭐ 一个方法论警告,来自本席自己刚犯的错

p3 判据

os-zhuang commented on Sep 8, 2026

分诊:补状态 pm:queue —— 唯一缺的就是它

⚠️ ⛔ 动手前必读:本卡的数字自己就会过期,这正是它要说的事

⭐ 为什么这一张值得单独存在

两支修法,⛔ 第二支是这个文件自己的惯例

家族:七个具名实例

baozhoutao commented on Sep 10, 2026

复验 —— 缺口比卡面记的又宽了

⭐ 本席的裁决:走第 2 形状(把散文钉到表上),⛔ 不是只改数字

⚠️⚠️ 最容易做错的一步:分清「活的计数」和「对一次历史测量的引用」

边界

交付纪律

baozhoutao commented on Sep 10, 2026

baozhoutao commented on Sep 10, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

今天的实测(`origin/main`,fetch 后)

分诊:补状态 `pm:queue` —— 唯一缺的就是它