finding(skills): the seat's claim-time serial check reads open PRs' file lists only, so a sibling card already claimed and pushed but not yet opened as a PR is invisible to it — one git ls-remote --heads origin | grep <file stem> would have shown #18314's branch before #18844 was claimed #18964

{
  "issue": 18964,
  "status": "needs_decision",
  "branch": "claude/issue-18964-claim-time-branch-stem-scan",
  "pr": null,
  "session": "session_01Wnstp2kTth7sGXfr8fXypc",
  "premise_still_valid": false,
  "summary": "The FINDING holds; the card's proposed REMEDY does not, measured. Zone 1 asks for one runbook line telling every seat to run `git ls-remote --heads origin | grep -i FILE-STEM` (stem = basename without extension) before each claim. Measured against the two instances the card and its evidence comment name, that probe fires 0 of 2. #18314 (the card's own founding miss) declared `scripts/check-single-claim-paths.mjs` and pushed `claude/issue-18314-single-claim-paths-proxy-reexec`; the strict stem `check-single-claim-paths` matches 0 of the 1169 heads on origin (only the shorter fragment `single-claim-paths` matches). #17518 (evidence 5759534756) declared `packages/spec/src/stack.zod.ts` and pushed `claude/issue-17518-assembled-package-body-inert-json`; the stem `stack.zod` matches 0 of 1169 heads and that slug carries no fragment of the file name at all -- spec seat 2's own printed probe used `grep -i 17518`, the CARD NUMBER. Population reading over today's 21 open `claude/issue-*` PRs (changeset files excluded, since `ISSUE-SLUG.md` matches the slug by naming convention, runbook :215): the branch slug contains a changed source file's stem in 6/21 (29%) -- 15/21 would read FREE while the branch really holds the file; restricted to single-source-file PRs it is 4/6, so with the two named instances the single-file view is 4/8. Root cause: `claude/issue-N-slug` slugs are derived from the CARD TITLE, not from the files touched, so a zero is not evidence of absence. I therefore did NOT write that line into a governed, headroom-0 file: it would declare a coverage the read does not deliver, which is the card's own defect class recommitted one layer up. What DOES fire is already in the tree one word away -- SKILL.md :267 `派发前在飞检查:读本车道 pm:dispatched 卡认领评论申报的文件面,本地求交、不另发查询。` -- and the miss was CROSS-LANE: #18314 is `domain:spec`, #18844 is `domain:skills`, so the skills seat's own-lane read could not see it, while #18314's claim 5725782443 (05:50Z, 31 minutes before the 06:21Z scan) names `scripts/check-single-claim-paths.mjs` verbatim. Dropping `本车道` from SKILL.md :267 is measured to fire on BOTH named instances; it is a SKILL.md edit this card's grading forbids. Everything needed to land any option in one step is measured below, including the proven payment. Assignee was `huangyiirene` (set) at my read; I wrote no assignee.",
  "tests": "No diff, so no gate is owed and none is claimed green for a change. (1) BASE `node scripts/pm/check-skill-line-ratchet.mjs` -> exit 0, verdict `check-skill-line-ratchet: .claude/skills/pm-dispatch/references/dispatch-runbook.md is 241 lines (ceiling 241; headroom 0).` (2) PAYMENT PROBE, a temporary applied-then-restored mutation proving the arithmetic works: insert the candidate line after :37 and retire :135 -> on-disk proof `injected marker count: 1`, `retired marker count: 0`, `wc -l = 241`, mutated blob f29e0bcc2 != HEAD blob 8dd281033; `node scripts/pm/check-skill-line-ratchet.mjs` -> exit 0, same verdict `is 241 lines (ceiling 241; headroom 0)` and `widest table row is 0 bytes (pin 0; headroom 0)`. Restored with `git checkout HEAD -- PATH` under a trap; restore proven by `git hash-object` = 8dd2810336abf016c44817519e98f4f7599eb73f = the HEAD blob, with `git diff HEAD --stat` and `git status --porcelain` both empty. (3) `node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands` (no paths) -> exit 2, `this branch changes nothing against origin/main (merge base 4fba5036f) -- nothing to derive`. (4) The same with the runbook path, for whoever lands the line -> exit 0, 17 commands: check-closing-keyword-parity(+self-test), check-comment-mask-corpus, pm/check-harness-current --self-test, lint check:doc-formula-expressions, check:agent-test-spelling, check:cross-package-test-inputs, check:doc-authoring, check:driver-memory-census, check:gitlink-declared, check:nul-bytes, check:pm-governed-merges, check:pm-skill-id-lint, check:pm-skill-ratchet, check:refd-timer-probe, check:skill-frame-sync, check:watch-hint-literal; it also warns `check:pm-settings-deny-roster` keeps its roster under `.claude`, so its silence is not evidence either way. (5) Duplicate scan: an n-gram pass over all 241 lines (n = 5..20, punctuation stripped) returns exactly one cross-line repeat of 12+ characters. All exit codes captured by redirect-then-`$?`, never across a pipe. NOT MEASURED: every gate family above, reason: there is no diff to judge; `node scripts/pm/check-governed-merges.mjs --pr N`, reason: no PR exists.",
  "line_budget": "`.claude/skills/pm-dispatch/references/dispatch-runbook.md is 241 lines (ceiling 241; headroom 0)` -- unchanged, nothing spent. Per-line cap measured at 120 bytes (the file's own widest line, :61, is exactly 120). Candidate line bytes: graded wording 118 B; `即影子` variant 119 B; zero-is-not-FREE variant 111 B; all under the cap.",
  "added_line": "NONE LANDED. The verified candidate, ready to apply: - 认领前读 `git ls-remote --heads origin | grep -i STEM`:命中他卡分支 ⇒ `Blocked-by:` 不派或并卡。 -- 118 bytes in-file, where STEM is spelled in the file as an angle-bracket placeholder (rendered here as STEM so the body sanitizer cannot eat the rest of this comment). Position: new line 38, immediately after :37 `- 全量 = 会话、分支、文件面、交付物、串行约束;成员卡各留一行指针指向锚卡。` inside 〈同趟成员卡写侧打包〉. That is where it reads best of the three homes measured: :37 is the runbook's ONLY 串行约束 statement and it is the detail section for SKILL.md :499-:500 (家族派发的折叠认领 / 完整串行约束检查落链首认领), so the new read sits beside the field it feeds. The two rejected homes: :145 / :149 carry `git ls-remote` but for the takeover protocol (a different act -- reading a branch you are about to take over), and :218-:219 (在飞重叠每轮拦截) is a post-dispatch read of main's new landings, not a claim-time one. Caveat worth the seat's eye: 〈同趟成员卡写侧打包〉 is a narrow section about batch member-card packaging, so a seat claiming a single card may not read it -- the grading's `beside the serial-queue lines` and `⛔ no SKILL.md edit` leave no wider home in this file.",
  "retired_duplicate": "The pair is :135 `- 跨账号接班活性只认 GitHub 产出读数,draft-PR 时点交报正为此存在。` (89 B) and :148 `- PM 会话已死 ⛔ 不推导其 dev 已死;跨账号活性只认 GitHub 产出读数。` (90 B). They share `活性只认 GitHub 产出读数` verbatim, and that 12-character overlap is the LONGEST cross-line repeat in the whole file (n-gram pass, n = 5..20; the next longest are `首行机器判据` at :55/:56 and `以脚本头为权威` at :17/:113, each carrying distinct content). Retire :135, not :148: the rule decides an act -- whether to dispatch a second dev onto a live branch -- and :148 is in 〈接手中断的 dev〉, the section where that act happens, where it also carries the ⛔ half (`PM 会话已死 ⛔ 不推导其 dev 已死`). :135's own section keeps its operative half at :134 (`该答案与从不存在不可区分,⛔ 永不读作死亡信号`), and :135's tail (`draft-PR 时点交报正为此存在`) is a justification, not an operational criterion -- the contract it justifies is stated at :184 and in the os-dev definition, and the file's own 行文纪律 at :105 says operational text carries operational criteria only. Deleting :135 loses no rule. ⛔ This is a deletion, not a re-wrap: no content is folded into another line. No script pins either line (repo-wide grep for both sentences returns only the two file lines themselves).",
  "open_questions": [
    {
      "question": "The card's Done-when is marked `⛔ a proposal, measure`. Measured, the stem scan fires 0/2 on the instances that motivated the card and reads FREE on 15/21 of today's open claude/issue-* PRs, because a branch slug is keyed to the CARD, not to the file. Which control should close the cross-lane blind spot -- and does the runbook still get a line?",
      "options": [
        "A -- land the graded stem-scan line anyway (118 B at :38, paid by retiring :135; arithmetic and ratchet proven green above). Cost: adds 29% recall over today's 0% on this shape, at one permanently retired line in a headroom-0 governed file, and a seat that runs it and reads zero writes `Serial constraints cleared` with more confidence than it can support. It misses both founding instances.",
        "B -- drop `本车道` from SKILL.md :267 (`派发前在飞检查:读本车道 pm:dispatched 卡认领评论申报的文件面,本地求交、不另发查询。`) so the in-flight claim-surface read runs across lanes. Measured to fire on BOTH named instances: #18314's claim 5725782443 names `scripts/check-single-claim-paths.mjs` verbatim, #17518's claim 5749581295 names `packages/spec/src/stack.zod.ts` verbatim; the file surface is DECLARED, so this is exact, not heuristic. Cost: reading other lanes' pm:dispatched claim comments per claim (:267's own `本地求交、不另发查询` still applies to the intersection). Blocked here: it is a SKILL.md edit, which this card's grading forbids -- so it is a new card, or a re-grading of this one.",
        "C -- land a runbook line that states the read AND its measured limit, e.g. - 认领前读 `git ls-remote --heads origin | grep -i STEM`;零命中非 FREE:分支名随卡不随文件。 (111 B, same payment). Honest and cheap, but it closes nothing by itself: the reading that decides (the in-flight claim surface) is cross-lane only under option B."
      ],
      "recommendation": "B, with C as the runbook half if a runbook line is wanted at all; ⛔ not A alone. Along the four axes: 实际业务需求 -- the need is measured twice over (#18314/#18844 cost one dev round, #17518 was caught only because it happened to be same-lane), and only B's instrument fires on either; A's instrument is a coin flip on its own target shape (4/8 across every single-target-file case I could measure) and fired on neither founding case. 项目长远合理性 -- B tightens an existing exact read at its one wrong word; A adds a second, heuristic control beside it, which is the workaround shape Prime Directive #5/#12 rules out, and it spends a permanent line of a shrink-only ceiling to do it. 防 AI 写错 -- this is the decisive axis: A has a seat write `Serial constraints cleared` on a probe that cannot see the case 71% of the time, i.e. it declares a coverage the runtime does not deliver, which is exactly the class this card was filed about (a field written on a reading that cannot see the case) and exactly what Prime Directive #10 forbids; B makes the absence a fact, because a declared file surface either names the file or does not. 创业阶段不扩散需求 -- B is a one-word narrowing-removal on a read that already exists and already runs; A is a new recurring per-claim read for every seat plus a retired line, i.e. surface growth for 29% recall. If the seat wants this card to close inside its own grading, C is the version of A that does not overstate itself; if it wants the blind spot actually closed, B needs a card that may touch SKILL.md."
    }
  ],
  "deviations": [
    "Zone 1's line was NOT landed and no PR was opened. Reason: os-dev rule 6 (the card body is a clue, not a spec -- verify its premises against origin/main; forcing a PR onto a dead premise is the failure form). The premise falsified is the card title's own claim that one `git ls-remote | grep FILE-STEM` would have shown #18314's branch: it would not have. Evidence in `summary` and `tests`; the payment and the exact edit are measured and ready so any option lands in one step.",
    "The empty claimed branch `claude/issue-18964-claim-time-branch-stem-scan` is pushed (0 commits, at 4fba5036f) -- it was the mandatory write-route probe taken before any edit, and it is the branch the PM's Claim 5774522173 names. It carries no work; delete it when this card closes, or reuse it if an option is landed.",
    "`dispatch-gates.mjs` printed `⚠️ STALE TREE`: origin/main advanced 4fba5036f -> a251aaa19 (09:40Z) during this run, with `scripts/engine-double-contract.pinned.json` changed across that range. Immaterial here (no diff, and that file is in no family this card touches), but the derived 17-command list in `tests` was read from the 4fba5036f tree and should be re-derived by whoever lands the line.",
    "Angle-bracket placeholders are rendered as STEM / FILE-STEM / ISSUE-SLUG / PATH / N in this comment per the body-sanitizer rule; the in-file spelling uses the runbook's own angle-bracket convention and is what the byte counts measure.",
    "No write-pace refusal: `node scripts/pm/write-pace.mjs --status` at 10:01Z read `budget: 38 of 40 ... 2 left`, `verdict: the next write may go`. No tool write was needed -- no label is owed with no PR -- so nothing was spent against the hourly budget; this comment is a bare-curl write outside the tool."
  ],
  "files_changed": "none -- the working tree is byte-identical to 4fba5036f (git hash-object of the runbook = 8dd2810336abf016c44817519e98f4f7599eb73f = its HEAD blob; git status --porcelain empty).",
  "governed_merges_reading": "NOT MEASURED -- `node scripts/pm/check-governed-merges.mjs --pr N` needs a PR and none was opened. For the record, the surface is Tier S on the register (`.claude/**`), so whichever option lands needs the seat's contract-tier review of record before the queue.",
  "api_writes": "1 REST proxy write: POST /repos/objectstack-ai/objectstack/issues/18964/comments (this report). Plus one non-REST write, `git push -u origin claude/issue-18964-claim-time-branch-stem-scan` (the empty-branch route probe, exit 0). Budgeted but NOT spent: POST /pulls (no PR), POST /issues/N/labels via scripts/pm/label-write.mjs (no PR to label, and `.claude/**` publishes nothing -- confirmed: the runbook path matches no package's `files[]`, the whole `.claude/` tree being outside every published package root).",
  "mcp_calls": "0 -- no MCP GitHub tool was called, read or write. All GitHub reads were bare-curl GETs: the card, its comments, 23 open PRs and their file lists, #18314/#18844 cards, #18314's comments, and comment 5749581295.",
  "out_of_scope_findings": [
    "carrier: whoever lands this card's line · noted, not filed -- the card's `SKILL.md :367` citation is stale on this base. Measured: at the filing-date tree 89c6ec52b line 367 was - shadow 命中在飞 ⇒ `Blocked-by:` 不派;open 未认领 ⇒ 先收敛成一个派发入口。, exactly the rule the card points at; commit 9ee8e3510 removed that line, and its content now lives at :380 (查重/shadow 检查先按文件/机制查本仓与姊妹仓 open 卡,含 pm:dispatched), :381 (同文件同机制 = 同一发现...首触定级即并) and :382 (同族异缺陷 ⛔ 不并:...在飞永不并,`Blocked-by:` 不派;完工查过期). A landed line should point at :380-:382, not :367.",
    "carrier: this card · noted, not filed -- the cross-lane framing is this card's own subject (spec seat 2 already extended its radius there in 5759534756, explicitly not filing it separately), so option B above is not filed as a second card; it is the decision this report is asking for.",
    "class: none · noted, not filed -- node's global `fetch` 401s through the agent proxy: my first measurement script used `fetch` and got HTTP 401 where the identical bare-curl GET returns 200. That is exactly the already-closed #18314 / #18844 defect, so it is confirmation of a landed fix's history, not a new finding; all readings above were retaken with curl."
  ]
}

{
  "issue": 18964,
  "status": "done",
  "branch": "claude/issue-18964-in-flight-read-across-lanes",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/19792",
  "session": "session_01Wnstp2kTth7sGXfr8fXypc",
  "claim_verified": "5789531283 is the newest Claim: comment on the thread (the earlier 5774522173 was retracted by Release 5774871631) and names branch claude/issue-18964-in-flight-read-across-lanes; no second claim posted, assignee not written (it read huangyiirene, set by the PM)",
  "premise_still_valid": true,
  "summary": "Ruling 5780497628 (letter B) executed: .claude/skills/pm-dispatch/SKILL.md :266 lost the word 本车道 and nothing else moved, so the pre-dispatch in-flight read now covers every lane's pm:dispatched claim surfaces. One commit (8ead063f3) on base 2cf9db7c4, net 0 lines (815 stays 815). references/core-rules.md is untouched because it does not carry this sentence. Draft PR #19792 is open, Tier S, labelled skip-changeset, and stays draft for the seat's contract-review record and queue landing.",
  "sites": [
    {
      "file": ".claude/skills/pm-dispatch/SKILL.md:266",
      "before": "- 派发前在飞检查:读本车道 `pm:dispatched` 卡认领评论申报的文件面,本地求交、不另发查询。",
      "after": "- 派发前在飞检查:读 `pm:dispatched` 卡认领评论申报的文件面,本地求交、不另发查询。",
      "proof": "line 121 → 112 bytes; old-sentence grep -c 1 → 0, new-sentence grep -c 0 → 1; git diff --stat 1 file, +1/-1"
    },
    {
      "file": ".claude/skills/pm-dispatch/references/core-rules.md",
      "before": "unchanged",
      "after": "unchanged",
      "proof": "grep -n 本车道 gives 1 hit at :18, the round-opening mutex reading (twin of SKILL.md :80), not the in-flight sentence; grep 在飞检查 hits :68 (cross-domain path, 同域跨域皆跑), which stays true; the claim's condition (the same sentence) does not hold"
    }
  ],
  "files_changed": [
    ".claude/skills/pm-dispatch/SKILL.md"
  ],
  "line_budget": "SKILL.md 815 → 815 lines (ceiling 819, headroom 4); widest table row 342 bytes (pin 342, headroom 0), untouched; net 0 lines",
  "tests": {
    "head": "8ead063f3 (git rev-parse --short HEAD at the gate run; the only commit)",
    "derivation": "node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands → exit 0, 21 commands, 1 path; --ran with exit codes → exit 0, verdict: 21 derived famil(ies) accounted for, 21 run, 0 NOT-MEASURED (a DERIVED zero, all 21 recorded an exit code and none of them is 3)",
    "gates": [
      "pnpm check:pm-skill-ratchet :: exit 0 :: SKILL.md is 815 lines (ceiling 819; headroom 4); widest table row is 342 bytes (pin 342; headroom 0)",
      "pnpm check:pm-skill-id-lint :: exit 0 :: check-skill-id-lint: 30 file(s) clean",
      "pnpm check:skill-frame-sync :: exit 0 :: the one declared copy of the decision frame is internally coherent",
      "pnpm check:skill-frame-freshness :: exit 0 :: the decision frame in this tree is current with origin/main (named by the dispatch, outside the derivation)",
      "pnpm check:doc-authoring :: exit 0 :: 16196 customer-facing string(s) clean; sibling-package prose ids hold the baseline",
      "pnpm check:nul-bytes :: exit 0 :: scanned 9262 text file(s), no raw ASCII control bytes",
      "pnpm check:pm-governed-prose :: exit 0 :: 2 instruction surface(s) name all 6 registered governed surfaces",
      "pnpm check:pm-governed-merges :: exit 0 :: --self-test 441 assertions",
      "pnpm check:pm-half-states :: exit 0 :: self-test 4897 cases pass",
      "pnpm check:pm-expected-skips :: exit 0 :: self-test 99 cases pass",
      "pnpm check:pm-settings-deny-roster :: exit 0 :: 17 content-write tool(s) declared = enforced (roster under .claude, outside the derivation)",
      "pnpm --filter @objectstack/lint run check:doc-formula-expressions :: exit 0 :: first run exit 3 PREREQUISITE NOT MET (formula and lint unbuilt, nothing measured); built under os-verify-lock (VERDICT command-exit 0, held 183s), re-run exit 0: 9 @example(s) judged clean; 14 predicate(s) judged clean",
      "node scripts/check-closing-keyword-parity.mjs :: exit 0 :: 3 parsers agree on all 9 keywords",
      "node scripts/check-closing-keyword-parity.mjs --self-test :: exit 0 :: 40 assertions",
      "node scripts/check-comment-mask-corpus.mjs :: exit 0 :: 7015 files, 0 disagree, 0 unparseable",
      "node scripts/pm/check-governed-queue-guard.mjs --self-test :: exit 0 :: 279 cases pass",
      "node scripts/pm/check-harness-current.mjs --self-test :: exit 0 :: all 26 cases passed",
      "pnpm check:agent-test-spelling :: exit 0 :: 0 violations",
      "pnpm check:cross-package-test-inputs :: exit 0 :: OK: 29 package(s) read outside themselves, all declared",
      "pnpm check:driver-memory-census :: exit 0 :: OK",
      "pnpm check:gitlink-declared :: exit 0 :: OK, 0 gitlink(s)",
      "pnpm check:refd-timer-probe :: exit 0 :: OK",
      "pnpm check:watch-hint-literal :: exit 0 :: 71 declaration(s), all quoted literals"
    ],
    "governed": "node scripts/pm/check-governed-merges.mjs --pr 19792 :: exit 3 (EXIT_TEST_GOVERNED) :: GOVERNED, Tier S, .claude/** x1; 2 changed line(s), under the human-merge threshold",
    "package_tests": "none owed: no package is touched, so no dependency-closure build and no package test or typecheck; no check:core-rules-* script exists",
    "ci_at_report": "head 8ead063f3 read once: 29 check runs, 14 success, 12 skipped, 3 in_progress, 0 failing; Lint & Repo Gates in_progress; Test Core, Dogfood Regression Gate, Governed Surface Queue Guard success; Build Core and Temporal Conformance skipped; TypeScript Type Check not present on the head at the read. Not waited on."
  },
  "serial_check": "16 open PRs, 308 file rows, 0 unreadable, none on SKILL.md or core-rules.md (lit control: 112 .changeset/ rows); origin/main 2cf9db7c4 equals the base at open, so no merge was owed",
  "mcp_calls": "0",
  "api_writes": "3 REST writes, all through the fleet-write relay as objectstack-fleet[bot]: (1) pr_create → POST /repos/objectstack-ai/objectstack/pulls draft (run 35823071999, success; body read back byte-identical, 6470 bytes, one footer); (2) label-write labels_add → POST /repos/objectstack-ai/objectstack/issues/19792/labels skip-changeset (run 35823127939, success; read-back matches: documentation, size/xs, skip-changeset); (3) post-stamped → POST /repos/objectstack-ai/objectstack/issues/18964/comments (this report). Plus git push x2 (the empty-branch probe and the one commit), not REST writes. Zero PATCH on the PR body.",
  "deviations": [
    "The dispatch named check:skill-line-ratchet and check:skill-id-lint; the package.json names are check:pm-skill-ratchet (scripts/pm/check-skill-line-ratchet.mjs) and check:pm-skill-id-lint, and those ran. No check:core-rules-* script exists.",
    "The dispatch probe for core-rules.md (0 hits of 本车道 means untouched) measured 1 hit (:18), a different sentence; the file is untouched on the claim's own condition (the same sentence is absent).",
    "H1 held (the sentence still sat at :266). H3 held: no open PR on SKILL.md, and origin/main had moved from 628e55dfa to 2cf9db7c4 before the worktree was cut, so the branch is on the current main with no merge.",
    "Harness attribution: the harness asked for a model-named Co-Authored-By trailer and a different PR footer. The commit carries the model-free pair (Claude-Session plus Co-authored-by: Claude) and the PR body ends in the session-URL footer, per AGENTS.md and os-dev.md.",
    "Cleanup: the worktree root node_modules was removed before this post, and git worktree remove runs after it. The final handback carries this same JSON unchanged."
  ],
  "open_questions": [],
  "out_of_scope_findings": [
    "carrier: none (承接者:无) · noted in the PR Acceptance notes, not filed · SKILL.md :267 (the cross-domain exception path adds every touched domain's in-flight cards) is now covered by :266 reading every lane; still true, left as is per the ruling"
  ]
}