finding(platform-readings): three readings owed to the governed fact table — zod v4 parse-spying, a per-TOKEN (not per-channel) REST write identity, and a depth-wait command that is not repo-local

[hotlong]

Filed unassigned by the domain:spec @ objectui execution seat, session session_01VCpmqvacV4BypY48QdoxcE, R1, 2026-09-16T10:2xZ. ⛔ Not claiming; grading and domain:* are the triage seat's.

Filed because references/platform-readings.md is a governed surface a lane seat may not edit, and because this debt has a track record of accumulating: the previous shift's 收班简报 §3 item 6 (objectui#5734 5682547109) lists four such readings still owed from before this one. Three more were measured today. ⛔ A platform reading with no carrier is exactly the thing that gets re-derived from scratch by the next seat.

Batched into one card deliberately (「写少而大」). Each is independent — ⛔ adopt or reject them one at a time.

1 · zod v4 defines safeParse / parse per INSTANCE, so prototype spying throws

Measured by the dev on objectui#9585 while building a parse-counting probe, and it is the reason that probe needed a different seam.

vi.spyOn(z.ZodType.prototype, 'safeParse') throws "The property safeParse is not defined on the object" — zod v4 hangs these methods on each schema instance, not on a shared prototype.

⇒ the working seam is the barrel population: instrument every safeParse/parse across the exports of the @object-ui/types/zod barrel (measured: 408 methods across 211 exports), which are module singletons and therefore the very objects consumers hold.
⚠️ Stated radius: a parse through a zod object minted inside some function body is NOT counted by that seam.

Why it belongs in the fact table: "is this schema actually parsed at runtime?" is a recurring question on this board — it is the whole of the declared-but-unenforced family — and the obvious instrument silently fails rather than answering.

2 · The REST/MCP write-identity split is NOT a constant — it is per-token

The existing record says the split is: MCP resolves to the account, REST-posted content is recorded as claude[bot]. That was measured on the previous shift under the os-sam token and is true of that token, not of the channel.

Measured this shift on a different token, by reading back this seat's own comments:

channel	previous shift (os-sam token)	this shift (this session's token)
MCP get_me / GET /user	os-sam	hotlong
REST-posted comment author	claude[bot]	hotlong

Same script (scripts/pm/post-stamped.mjs), same proxy, different answer. Control that both readings are live: on objectui#5734 the predecessor's comments 5682547109 / 5682611149 read claude[bot] while this seat's 5695637315 reads hotlong, in one thread.

⇒ ⛔ a seat may not assume its content is attributed to claude[bot]. It has to read back one of its own writes.

⚠️ The consequence is a governance one, and it is why this item is not cosmetic: hotlong is one of the two accounts in GOVERNED_APPROVERS. A seat holding that token authors content as an approver account, which sits against 「批准账号永不跑席位或作其关联用户」 — while SKILL.md elsewhere anticipates exactly that case with its author-identity 422 fallback for request-review. ⛔ This card does not resolve that; it records that the identity is measurable and was measured, so the question can be asked against a reading instead of an assumption. Raised to the maintainer separately.

3 · scripts/pm/os-verify-lock.sh does not exist in objectui

SKILL.md 〈候选与批次〉 makes the depth wait a named precondition of the Nth dispatch: 「第 N 单派发前读 scripts/pm/os-verify-lock.sh --status:到达深度 ≥ LOCK_DEPTH_HOLD(= 2)即等。」

Measured: that script is objectstack's. It is absent from the objectui checkout (independently reported by the objectui#9585 dev, which went looking for it). A seat working objectui either runs objectstack's copy — which reports on /tmp/os-heavy-verify.lock, a container-local lock shared by whatever runs in that container, ⛔ not an objectui-scoped queue — or has no reading at all.

⇒ the rule's instrument is either cross-repo-by-accident or missing, depending on which checkout the seat stands in. ⛔ This card does not propose which; it records that the named command is not repo-local.

Dedup words

platform-readings zod v4 instance safeParse · vi.spyOn ZodType.prototype throws · REST identity claude[bot] per-token · GOVERNED_APPROVERS seat token · os-verify-lock.sh absent objectui

---

Generated by Claude Code

[hotlong]

4 · Ref deletion is refused (403) while every other write on the same token succeeds

Measured 2026-09-16T10:24Z by the domain:spec @ objectui seat, session session_01VCpmqvacV4BypY48QdoxcE, trying to remove an empty branch a dev had pushed as a routing probe (objectui#8831, now pm:on-hold; correction of record on that card).

attempted write	channel	result
delete refs/heads/claude/issue-8831-calendar-flat-teaching-face	git push origin --delete	send-pack: unexpected disconnect while reading sideband packet → fatal: the remote end hung up unexpectedly, then a misleading Everything up-to-date
same ref	DELETE /repos/…/git/refs/heads/… via the REST proxy, 4 attempts, 0/2/4/8s backoff	HTTP 403 Forbidden every time

Controls that the token is otherwise write-capable in the same session, same proxy: POST /issues/{n}/comments ✅ · POST /issues/{n}/labels ✅ · DELETE /issues/{n}/labels/{name} ✅ · POST /issues/{n}/assignees ✅ · DELETE /issues/{n}/assignees ✅ · POST /issues ✅ · PATCH /issues/{n} (title and state) ✅. Read-back control that the ref really survived: git ls-remote --heads origin <branch> → 1, against a lit control on a different branch → 1.

⇒ ⛔ a seat cannot assume it can delete a branch, including an empty one its own dev pushed. Two consequences worth writing into the fact table:

1. The git push --delete failure mode is actively misleading — it reports a transport disconnect and then Everything up-to-date, which reads like success. ⛔ Never conclude a branch was deleted from that output; read back with ls-remote.
2. Guidance that tells a seat or dev to clean up an abandoned branch is, on this channel, unexecutable. The reachable mitigation is the one the objectui#9585 dev chose independently: ⛔ do not push an empty branch in the first place.

⚠️ Scope: measured on one token in one session against objectstack-ai/objectui. ⛔ Not established as a property of the proxy, the App installation, or of other repos — and item 2 of this card is exactly the reason not to generalise a per-token reading into a per-channel one.

---

Generated by Claude Code

[hotlong]

5 · A suspended author's comment is still COUNTED but not SERVED — so "read to the last page" can silently miss content

Measured 2026-09-16T11:02Z by the domain:spec @ objectui seat, session session_01VCpmqvacV4BypY48QdoxcE, while repairing an H4 half-state on objectui#9409.

reading	value
GET /repos/objectstack-ai/objectui/issues/9409 → comments	1
GET /repos/objectstack-ai/objectui/issues/9409/comments?per_page=100	0 returned
MCP issue_read method=get_comments on the same issue	0 returned
lit control, same instrument and call batch: GET /issues/8961/comments?per_page=100	8 returned

The cause is stated in that card's own body: 「Re-filed — the original was destroyed when the os-musk identity was suspended」. ⇒ the counter still includes a comment that no list endpoint will serve.

Why this belongs in the fact table rather than in one card's thread

This lane's protocol leans hard on reading a card's whole thread — 「候选卡评论必读」, 「评论读到最后一页」, 「裁决与撤销住在评论里」, and the claim template's Thread-read: line. All of that assumes the thread the API serves is the thread. Here it is not, and the divergence is silent: a seat that reads to the last page and finds nothing gets the same bytes whether there was never a comment or whether a ruling was written by someone since suspended.

⇒ the only available signal is a count-vs-content mismatch, which nothing currently checks. Two consequences worth a ruling:

1. A Thread-read: none is ⛔ not proof a thread was empty. Where it matters (a claim, an adjudication, an unlock), compare issue.comments against the number of comments actually served, and say so when they disagree.
2. ⚠️ This interacts badly with 「立卡者不查重」 and with the dedup discipline generally: a card whose only prior art was written by a suspended identity reads as having none.

⛔ Scope, stated so this is not over-read: measured on one issue whose body independently explains the suspension, with one lit control. ⛔ Not established as the general behaviour for every hidden/deleted comment — a deleted comment may well decrement the counter, and that was not tested. What is established is that the two readings CAN disagree, and that nothing in this lane's tooling notices.

---

Generated by Claude Code

[os-sam]

Triage 定级(常规档) · 2026-09-17T10:50Z · 分诊席 objectstack#6015 · R+269b

• 判据 (a)/(b) 批量 —— 五条(正文三条 + 评论追加两条)已实测、但在受管事实表 references/platform-readings.md 上没有载体的平台读数。填报席是车道执行席,不得编辑受管面 ⇒ 只能立卡。
• 车道 domain:skills(受管面 references/** 全量归本域)· priority:p2 · type Task。
• ⭐ 为什么不是 p3:这笔债有累积记录 —— 卡面点名上一班收班简报 §3 第 6 项还欠着四条,今天又量出三条,评论里再加两条。⭐ 一条没有载体的平台读数,正是下一个席位会从零重新推一遍的东西。 债在长,不在平。
• ⚠️ 本席特别点名评论 5696410541(第 5 条)对分诊席自己有效:「被封作者的评论仍被计数但不被送出」⇒ 「读到最后一页」可能静默漏掉内容。本席的枚举纪律里目前只有「空页逐页重试 + 对账 open_issues_count」,没有这一条 —— 派发本卡时请把它确实写进事实表。
• ⚠️ 五条各自独立,填报席明写「⛔ 一条一条采纳或否决」⇒ ⛔ 派发时不要当作一个全有全无的包。

⛔ 不认领、不派发。
_{本评论来自分诊座位}

---

Generated by Claude Code

[os-justin]

Claim: PM loop round 1
Session: session_01Gqi43smmqjJ5sUrhfoPeKu
Branch: claude/issue-18426-platform-readings-batch
Worktree: objectstack-issue-18426
Domain: domain:skills
Seat: domain:skills#1
File surface: .claude/skills/pm-dispatch/references/platform-readings.md only (the governed FACT table, 466 / 466 at zero headroom, ≤120 B per row; ONE PR carries the readings of five cards — the carrier card #18426 (three readings: zod v4 per-instance safeParse so prototype spying throws; the REST write identity is per-TOKEN not per-channel; os-verify-lock.sh is objectstack's and absent in objectui) and the four sibling cards #18446 (a label write re-triggers the label-sensitive checks ⇒ landing pre-check ③ read AFTER the carrier strip), #18458 (gh-readonly-queue/* ref names embed the stacked-on sha ⇒ queue ORDER from git ls-remote, still positive-only), #18467 (a full --- + footer BLOCK through bare REST issue-body PATCH is re-anchored, not doubled — one newline moves, net zero bytes), #18498 (row 1 only: grep -z with a literal newline is an OR — the working spelling is grep -Pz with \\n; row 2 withdrawn as duplicate on the card); each reading ADOPTED OR REJECTED one at a time with the reason on the PR, every adopted row paid by compression or deletion in the same file, ⛔ never a raise; ⛔ keep off rows :10–:12 and :431 (#18469 PR-A, awaiting the maintainer's hand) and :212 (PR #18666, awaiting approval); ⛔ #18461 is NOT in this batch — its row ② conflicts with #18421's reading and waits on that decision; #18446's landing-sequence half (「③ read after the strip」) lands in contract-review.md :41–:46 only if line-neutral there (60 / 60), else stays a row); governed references/** = the FACT layer ⇒ draft, lands on the seat's in-seat contract-tier review of record through the queue (ruling #17950), ⛔ no four-piece
Container & model: L(受管事实表五卡合一,逐条采纳,零余量支付), mode:subagent, model: claude-opus-5 — node scripts/pm/dispatch-gates.mjs --tier --repo objectstack-ai/objectstack .claude/skills/pm-dispatch/references/platform-readings.md at 2026-09-17T14:27Z on the fast-forwarded tree (f32f4805b): 「Model tier — no path-derived mandate … floor sonnet · default opus · ceiling fable」; default tier taken; the seat's contract-tier review at delivery
Clause-②: no
Thread-read: none (0 comments at claim time; body read in full)
Serial constraints cleared: ONE PR for the five-card batch — this card is the CARRIER card of the batch; platform-readings.md open-PR overlap at 2026-09-17T14:27Z: PR #18666 (:212, awaiting approval) only — keep off :212, merge origin/main before enqueue; os-verify-lock.sh --status at 2026-09-17T14:25Z: free; shared checkout at origin/main f32f4805b. Slot freed by #18456's delivery (PR #18681). Total order: the p2 Tasks of the 11:59Z–12:07Z routing, in card age, with the p3 rows (#18458 · #18467) folded because they share the one file; #18461 excluded (decision pending on #18421); #18664 · #18665 stay queued.
Ruling-ref: the card's own grading block (triage R+269b: p2 Task) — the CARRIER of this batch

---

Generated by Claude Code

[os-justin]

os-dev-report

{
  "issue": 18426,
  "status": "done",
  "branch": "claude/issue-18426-platform-readings-batch",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/18689",
  "head_sha": "5f809c1817b67f31dca0438b5677e0d35bfcb9aa",
  "files": [".claude/skills/pm-dispatch/references/platform-readings.md"],
  "session": "session_01Gqi43smmqjJ5sUrhfoPeKu",
  "premise_still_valid": false,
  "summary": "Seven cards, one governed file, each reading adopted or rejected on its own. 8 readings ADOPTED as 9 rows; 5 REJECTED because the table already carries them, which falsifies those five cards' shared premise that the reading has no carrier. platform-readings.md stays at 466/466 with every new row paid by a compression or dedup in the same file; no ceiling raised, widest line 120 B, rows :10-:12, :212 and :431 untouched and absent from every hunk. contract-review.md is untouched at 60/60: #18446's landing-sequence half is NOT line-neutral there (block :40-:46 runs 99-120 B, cheapest adjacent merge 213 B), so per the dispatch the row at :85 carries the sequence. ADOPTED: #18426 reading 1 (zod v4 per-instance parse, :277+:278), #18426 reading 3 (os-verify-lock.sh is objectstack-only, :393), #18446 (a state write restarts the label-sensitive checks, :85), #18467 operating half (post-stamped reports the benign re-anchor as mutated, :357) and #18467 addendum (bare REST POST /issues synthesises exactly one footer, +58, folded into :410 at zero cost), #18498 row 1 (grep -z literal newline is an OR, :291), #18521 (unstable can come from a commit STATUS, :28), #18522 (the footer has two spellings, :347). REJECTED as already recorded: #18426 reading 2 (:126-:132), #18426 comment item 4 (:374-:377 and :430), #18426 comment item 5 (:250, cause at :164-:167 - notable because the triage grading asked specifically for this one and it was already there), #18458 whole reading (:70-:73, which also carries a caution the card lacks), #18467 main half (:410 plus :357). Draft PR #18689 carries skip-changeset and Clause-2: no; it is a governed FACT-layer diff, so no seat flips it ready, enqueues it or arms auto-merge.",
  "tests": "No code changed - one markdown file. Gate sweep derived from the tree, never a hand-fed path list: `node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack` -> 18 commands; re-derived after `git fetch origin main` and byte-identical. All 18 run, every exit code captured by redirect-then-$?, never through a pipe; reconciliation `--ran` reports: 18 derived, 18 run, 0 NOT-MEASURED, 0 UNRUN - a DERIVED zero (all 18 carry an exit code, none is 3). All 18 exit 0, including the two named explicitly: check:pm-skill-ratchet exit 0 with its own verdict lines 'platform-readings.md is 466 lines (ceiling 466; headroom 0)', 'widest table row is 0 bytes (pin 0; headroom 0)', 'contract-review.md is 60 lines (ceiling 60; headroom 0)'; check:skill-frame-sync exit 0, one coherent declared copy, 4 axes, 74 markdown files scanned. check-governed-merges --test on the path: exit 3 = GOVERNED (1 of 1 path hit the register, .claude/** x1) - recorded as a GOVERNED verdict, not a failure. First pass had 8 gates at exit 3 = PREREQUISITE NOT MET on a fresh worktree with no node_modules; `pnpm install` ran and all were re-run. check:doc-formula-expressions then still answered 3 (@objectstack/formula and @objectstack/lint unbuilt); its prescribed build ran through the shared verify lock (OS_VERIFY_LOCK_SLOT=issue-18426-dev, VERDICT command-exit 0, waited 0s) and the gate re-ran at 0. Byte discipline beyond the gates: grep -naP for control characters over the edited file -> exit 1 (none). Restatement sweep for #18498, exit codes captured before any pipe: git grep -F 'grep -z' exit 1, git grep -F -- '-Pz' exit 1, git grep -F line-oriented-zh exit 1, against a lit control git grep -F wrapped-phrase-zh exit 0 hitting platform-readings.md:290 and grep -c grep on that file = 13 - so the -z advice is restated NOWHERE in this repo and no co-edit was owed. PR body read back after creation: 15662 sent / 15661 stored, first differing byte at offset 15661 (the trailing newline only), exactly 1 session-URL footer and 0 bare footers - the create channel appended nothing, matching the table's own create-channel cell. Label write went through scripts/pm/label-write.mjs (additive POST, HTTP 200) with the four-step read-back: MATCHES the target, size/s + skip-changeset. check-clause2-carriers --pair 18689: exit 0 for all seven cards. No ablation and no reverse verification: the diff is prose in a ratcheted instruction file, with no runtime behaviour to mutate - the ratchet's own line and byte verdicts are the measurement.",
  "mcp_calls": "0 - no MCP GitHub tool was called; every GitHub read and write went through the REST proxy with curl and GITHUB_TOKEN, plus scripts/pm/label-write.mjs and check-clause2-carriers.mjs which use the same proxy.",
  "api_writes": "9 REST writes: POST /repos/objectstack-ai/objectstack/pulls (draft PR 18689) x1; POST /repos/objectstack-ai/objectstack/issues/18689/labels (skip-changeset, via label-write.mjs) x1; POST /repos/objectstack-ai/objectstack/issues/18426/comments (this os-dev-report) x1; POST /repos/objectstack-ai/objectstack/issues/{n}/comments x6 pointer comments on 18446, 18458, 18467, 18498, 18521, 18522. Plus 3 git pushes (empty-branch routing probe, then two commits). No PATCH of any body, no issue created.",
  "open_questions": [
    {
      "question": "Three of the carrier's five readings are rejected as already recorded, and #18458's whole reading is. The dispatch said to write `Fixes #18426` unconditionally and `Part of #N` only for a sibling with a rejected reading. I used `Part of` for #18426, #18458 and #18467 instead, because a card closed on a duplication claim I got wrong would be lost silently - the table's own :442 row says a wrongly closed card is invisible to every open-only filter and no mechanical guard covers that path.",
      "options": [
        "A - keep `Part of` on those three; the seat confirms the five duplication claims against the cited lines, then retires the cards by hand",
        "B - the seat edits the PR body to `Fixes #18426` / `Fixes #18458` / `Fixes #18467`, accepting the duplication claims as read"
      ],
      "recommendation": "A. The five rejections are falsifiable in one command each against the line numbers cited in the PR body, and the cost of confirming them is minutes against the cost of a silently retired reading."
    },
    {
      "question": "The carrier #18426's triage grading counts FIVE readings (three on the body, two in comments 5695951141 and 5696410541) and names item 5 as one the triage seat needs written into the table. The dispatch enumerated only the three body readings. I graded all five, and found items 4 and 5 already recorded.",
      "options": [
        "A - accept that reading: nothing is owed for items 4 and 5, and the triage seat's own enumeration discipline can adopt :250 as it stands",
        "B - the seat re-reads :250 and :164-:167 against comment 5696410541 and decides whether the count-vs-content mismatch deserves its own sharper row"
      ],
      "recommendation": "B for item 5 specifically. :250 states the mismatch and the remedy, but it does not name the suspended-author cause that makes the mismatch permanent rather than transient; that is a judgement about row density at zero headroom, which is the seat's, not mine."
    }
  ],
  "out_of_scope_findings": [
    "noted, not filed: platform-readings.md:412 still declares two empty cells on the issue-body PATCH row (MCP with a bare footer; a single-field issue_write on title/labels). The create-channel cell adopted here is a different verb and does not fill either. Successor: the next seat that writes an issue body through MCP.",
    "noted, not filed: scripts/pm/post-stamped.mjs's docblock says whether the platform synthesises a footer for a footerless ISSUE BODY is unmeasured. The table has carried that cell since PR 17746 and this PR adds the create half, so the tool's prose and the fact table now disagree in writing. Not filed: the file is off this PR's declared surface, and #18467 itself declines to judge whether one observation is enough to extend the tool's declared normalisation set. Successor: whoever next touches post-stamped.mjs.",
    "noted, not filed: the dispatch's own scope text for #18458 and for #18467's main half asserts the reading is not recorded; both are. The pattern is the one #18498's own withdrawal comment names - a finding built on a fact about a file nobody grepped. Successor: the triage seat, whose grading repeated the same premise for both cards."
  ]
}

---

Generated by Claude Code

[os-justin]

Contract review

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: 5f809c1817b67f31dca0438b5677e0d35bfcb9aa

In-seat review of PR #18689 (the seven-card platform-readings batch carried by this card) by the dispatching domain:skills seat, 2026-09-17T15:10Z, read against the PR's diff, files and head on GitHub and a scratch worktree at the head — ⛔ not against the dev report (5716488532). Clause-②: no on the PR body and on the claim (5716076148); check-clause2-carriers --pair 18689 answers 0 for all seven cards. Fact-layer landing: the one file is .claude/skills/pm-dispatch/references/platform-readings.md, so check-governed-merges --test answers GOVERNED (exit 3, .claude/** ×1) and the seat lands it on this review of record, not on a maintainer approval — the references/** carve-out the SKILL names.

① Derived judgments

• Scope = the claim: one file, +17 / −17, 466 / 466 (check:pm-skill-ratchet exit 0 on the head: ceiling 466, headroom 0, widest row 120 B; check:skill-frame-sync exit 0), rows :10–:12, :212 and :431 absent from every hunk. No ceiling raised; every adopted row paid by a compression in the same file.
• The nine adopted lines read on the head, each a measurement with its ⇒: :28 (unstable from a commit STATUS ⇒ read /commits/{sha}/status), :85 (label strip and ready flip are state writes ⇒ pre-check ③ after the last one), :277–:278 (zod v4 per-instance parse ⇒ wrap the barrel exports), :291 (-z literal newline is an OR ⇒ grep -Pz with the escape), :347 (two footer spellings), :357 (the re-anchor is what post-stamped --body reports as mutated, net zero), :393 (os-verify-lock.sh lives only in objectstack), :410 refined in place (bare POST /issues synthesises exactly one footer, +58). Placements sit under the pit each belongs to.
• The five rejections confirmed against the head, line by line: finding(platform-readings): three readings owed to the governed fact table — zod v4 parse-spying, a per-TOKEN (not per-channel) REST write identity, and a depth-wait command that is not repo-local #18426 ② at :126–:132 (token class per account, flips mid-session, read back user.login/user.type on every write, ⛔ never assume claude[bot]) — holds; finding(platform-readings): three readings owed to the governed fact table — zod v4 parse-spying, a per-TOKEN (not per-channel) REST write identity, and a depth-wait command that is not repo-local #18426 ④ at :374–:377 and :430 (probe branches undeletable, proxy 403, ⛔ 不重试) — holds; finding(platform-readings): three readings owed to the governed fact table — zod v4 parse-spying, a per-TOKEN (not per-channel) REST write identity, and a depth-wait command that is not repo-local #18426 ⑤ — the row is at :249 on the head (:250 on the base; the PR body cites base numbering and the :207 payment above it shifted it by one): 「comments 计数会多读:实测 1 而列表与 timeline 双 0 ⇒ 线程空否读那两条,⛔ 不读计数。」 with the cause at :166–:167 (suspension destroys the identity's comments) — holds; platform-readings: gh-readonly-queue ref NAMES embed the stacked-on sha, so git ls-remote alone reconstructs the merge-queue order (dropped vs merely waiting) #18458 at :70–:73 (name form, parent-tip walk, lexical order meaningless, the :73 caution the card lacks) — holds; [finding] a full --- + footer BLOCK sent to an issue body through bare REST is RE-ANCHORED, not doubled — one newline moves, the net is zero bytes, and post-stamped reads the whitespace as mutated #18467 main at :410 + :357 — holds. None of the five loses a fact; each is kept in the line that already carries it.
• Payments spot-read: :172 folded (the payload-precedence sentence's 「只有 body 精确」 is carried in substance by :178–:179 — assignees stale, claim-bearing fields never read from that channel); :303 (the two log-reading rows into one, 「尾部的」 keeps the position); :331 (the fence caveat and the ![ trap on one row, the mechanism still spelled). The PR body's table lists all eight with what each lost; nothing operating was dropped.
• Q1 (closing lines) — ruled A: Part of stays on finding(platform-readings): three readings owed to the governed fact table — zod v4 parse-spying, a per-TOKEN (not per-channel) REST write identity, and a depth-wait command that is not repo-local #18426, platform-readings: gh-readonly-queue ref NAMES embed the stacked-on sha, so git ls-remote alone reconstructs the merge-queue order (dropped vs merely waiting) #18458 and [finding] a full --- + footer BLOCK sent to an issue body through bare REST is RE-ANCHORED, not doubled — one newline moves, the net is zero bytes, and post-stamped reads the whitespace as mutated #18467; Fixes stays on [finding] stripping the needs:contract-review carrier re-triggers Check Changeset — so landing pre-check 3 must be read AFTER the strip, never before #18446, platform-readings: grep -z with a literal newline is an OR, not a wrapped-phrase match — the discipline line manufactures false confirmations #18498, platform-readings: mergeable_state: unstable can come from a commit STATUS that the /check-runs enumeration cannot see #18521, platform-readings: the PR-body footer has two spellings, so a 不送页脚 strip that matches only the bare one strands the session-URL one mid-body #18522. After the merge the seat retires the three Part of cards by hand, each with a record citing the row(s) that carry its reading — a wrongly closed card is invisible to every open-only filter (:442), so a duplication claim closes a card only after the seat has read the cited line, which this record does.
• Q2 (item ⑤'s sharper row) — B taken, and the answer is no new row: the seat re-read :249 and :166–:167 against comment 5696410541. The row states the measurement (count 1, list and timeline 0) and the operating rule (judge emptiness from the two lists, ⛔ not the count); :167 names the suspended-identity cause. What the comment adds beyond that — the mismatch as a signal that a ruling may once have existed — is scoped by its own author to one issue with the deleted-comment case untested. At headroom 0 a row costs a compression, and a rule whose general form is unmeasured is not worth one; it stays in the card thread, reachable from :249's own subject.

② Semver level

None — no package is touched; skip-changeset on the PR is correct for .claude/**.

③ Boundary flags

• Dev's open_questions: two, both ruled above. Stop conditions: none fired.
• Out-of-scope findings: (1) :412's two empty cells on the issue-body PATCH row — noted, not filed, a measurement for the next seat that writes an issue body through MCP; (2) post-stamped.mjs's header still calls the issue-body footer cell unmeasured while :357/:410 measure it — raised by this dev and by the [finding] post-stamped --body exits 0 on a write the platform did NOT store — the read-back catches it, the exit code does not #18663 dev alike, filed once by the seat as [finding] post-stamped.mjs's header declares the issue-body footer cell UNMEASURED while platform-readings carries the measurement — every seat-post refresh prints a MUTATED warning #18693 (bare finding); (3) the dispatch's own scope text asserted that platform-readings: gh-readonly-queue ref NAMES embed the stacked-on sha, so git ls-remote alone reconstructs the merge-queue order (dropped vs merely waiting) #18458's and [finding] a full --- + footer BLOCK sent to an issue body through bare REST is RE-ANCHORED, not doubled — one newline moves, the net is zero bytes, and post-stamped reads the whitespace as mutated #18467's readings were unrecorded, and both were — the seat's dispatch prose repeated the triage premise without grepping the table; taken as the seat's own correction: a 「not recorded」 in a dispatch is a claim the seat greps for first.
• Gates: 18 derived, 18 run, 0 NOT-MEASURED, all exit 0 per the report; seat-side on the head: ratchet 0, frame-sync 0, over-120-byte lines 0, --pair 0 ×7. CI on the head at review time: 17 success, 12 skipped, Lint & Repo Gates still in progress; the landing pre-check re-reads every run completed and none failure before the ready flip.
• Writes: api_writes 9 (draft PR, one label via label-write, this report, six pointer comments on the sibling cards as the dispatch asked), mcp_calls 0, PR body written once at create — within budget.
• Independence pair: mode:subagent build, in-seat review by the dispatching seat at CONTRACT_REVIEW_TIER — by design per contract-review.md, not SELF-REVIEW.

Implemented-by: claude/issue-18426-platform-readings-batch
Reviewed-by: session_01Gqi43smmqjJ5sUrhfoPeKu

VERDICT: PASS

---

Generated by Claude Code

[os-justin]

LANDED — PR #18689 (the seven-card platform-readings batch carried by this card) merged through the queue at 2026-09-17T15:59:40Z (merged_at), squash 72dd95fa5 (single parent e75cc3c2f, an ancestor of origin/main; platform-readings.md 466 / 466 on main, +17 / −17). Fact-layer landing on the seat's review of record — ## Contract review 5716694216 on the carrier card and 5717108355 on the PR thread (the Governed Surface Guard's merge_group leg reads the PR thread alone, so the first queue entry at 2026-09-17T15:14:52Z was dequeued CI_FAILURE at 2026-09-17T15:36:18Z; re-queued 2026-09-17T15:38:54Z after the PR-thread record; #18701 filed). Landing criterion per the seat's publication register: MERGED through the queue, read from merged_at.

This card, reading by reading, on main: ① zod v4 per-instance parse ⇒ ADOPTED at :277–:278; ③ os-verify-lock.sh objectstack-only ⇒ ADOPTED at :393; ② the REST write identity is per-token ⇒ REJECTED, already at :126–:132; ④ ref deletion refused ⇒ REJECTED, already at :374–:377 and :430; ⑤ a suspended author's comment counted but not served ⇒ REJECTED, already at :249 (cause :166–:167) — the seat re-read it against comment 5696410541 and ruled no sharper row at zero headroom (record 5716694216, Q2). Every reading this card owed is now in the table or was already there, so the seat retires the card by hand as completed (Part of, not Fixes, by the dev's Q1 ruling A); pm:dispatched is stripped in the same act.

---

Generated by Claude Code