[finding] platform-readings.md lines 47–49 say undraft is MCP-only and a red window can only wait — the proxy's `ccr` REST route flips a draft on the seat's own credential (the sibling of #17820)

[claude]

Filed by the skills seat (session session_01MCLBsUgfykL74aU716rzVK, GitHub os-sales) at 2026-09-12T09:17Z, promoting the #17820 dev's noted, not filed (report 5644946032) so the reading does not live only in a report. Class (b): a reference table's declared fact no longer matches the platform.

The three lines

.claude/skills/pm-dispatch/references/platform-readings.md on origin/main f830fa21, lines 47–49 (the PR-lifecycle write section):

47  - undraft 可用路径只有 MCP 这条:传 `draft: false` 落地。
48  - 裸 GraphQL 在会话内被拒,它建议的改用 REST 对 undraft 是错的。
49  - 裸 REST `PATCH /pulls/{n}` 传 `draft: false` 回 200 而无操作 ⇒ 接口性质,池 0 只能等重置。

What the platform does today (measured twice on 2026-09-12, both with read-back)

The session proxy refuses POST /graphql and its refusal names REST routes on the seat's own credential; POST /repos/{owner}/{repo}/pulls/{n}/ccr/ready_for_review flips a draft (read back draft: false, timeline ready_for_review) and PUT …/pulls/{n}/ccr/auto_merge arms auto-merge (timeline added_to_merge_queue). Evidence: #11742 comments 5644551740 (cli seat, 07:46Z, PR #17812) and 5644830981 (this seat, 08:45Z, PR #17816). The sibling table rest-channel.md is corrected by PR #17823 (#17820, at the human terminal).

⇒ line 47 is false (MCP is not the only path); line 48 is false in its second half (the REST route the refusal suggests DOES work for undraft — the ccr one, not the naked PATCH); line 49's first half is still true (the naked PATCH is a no-op) but 「池 0 只能等重置」 is false — the ccr route is the exit. Lines 50 / 58 / 60 (the enable_pr_auto_merge rows) stay true of the MCP channel and are ⛔ not this card's.

Work item

Rewrite lines 47–49 in place so they state today's fact, dated 2026-09-12, with the naked-PATCH no-op reading kept as the ⛔ line: undraft has two channels (ccr REST on the seat credential, MCP update_pull_request); the GraphQL refusal's suggested ccr route works, the naked PATCH does not; a red window on the MCP identity is not a wait. ⛔ In place at the same line count — the file is at its ceiling (449/449); no raise, no new row. ⛔ No ruling on the timeline-actor question (#11742). The 「不可迁移」 wording in rest-channel.md is PR #17823's, ⛔ not touched here.

What this card is NOT

• ⛔ Not [Decision] The shared GitHub identity's GraphQL quota is being burned to 2× — MCP writes go through GraphQL, so seats are silently write-blocked while reads keep working #11742 (the shared identity's quota, a decision on hold) and ⛔ not 共享身份的限流纪律不存在：一次限流信号约束的是「身份」不是「客户端」，而规矩只说了不要重试 —— 2026-09-10 全 fleet 停摆事故 #17374 (rate-limit discipline). ⛔ Not [finding] rest-channel.md rows 1–2 say the draft flip and auto-merge are GraphQL-only with no fallback — the proxy's ccr REST routes now do both on the seat's own credential (measured twice on 2026-09-12) #17820 (the rest-channel.md half, already at the human terminal). Dedupe: the 32 open domain:skills cards grepped for undraft and for platform-readings ∧ (ccr ∨ GraphQL) — hits [finding] rest-channel.md rows 1–2 say the draft flip and auto-merge are GraphQL-only with no fallback — the proxy's ccr REST routes now do both on the seat's own credential (measured twice on 2026-09-12) #17820 (the sibling), [PM decision] skills lane batch 9 — platform-readings.md third increment: raise the ceiling by the measured +39 (358 → 397); and the MCP write-pool quota sentence collision #15275 (a ceiling-raise decision on hold), PM corpus audit (maintainer-directed): classify every instruction line — keep / duplicate / demote / delete — then rewrite and re-lock every ratchet #13597 (the corpus audit anchor); control: 6 open lane cards mention platform-readings, so the corpus was read. [finding] The "pair every zero with a control" rule does not say the control must be scoped to the same thing as the claim — three misses in one shift, every one returning a reassuring answer #17569's rider is a different fact (git grep pathspec).

Refs

#17820 / PR #17823 (the rest-channel.md half) · #11742 (5644551740, 5644830981) · platform-readings.md lines 47–49 · scripts/check-skill-line-ratchet.mjs CEILINGS (449).

---

Generated by Claude Code

[claude]

Triage (skills-lane finding self-triage — the lane's standing exception; filed by this seat from the #17820 dev's report, dedupe run before grading): admitted — class (b): platform-readings.md lines 47–49 declare undraft MCP-only and a red window a wait; the platform answered otherwise twice on 2026-09-12 with read-back (#11742: 5644551740, 5644830981). finding dropped; pm:queue · Task · priority:p3 — the same line as #17754 and #17820: a wrong fact-table row costs a seat a wrong red-window plan. Direction: three lines rewritten in place at the same line count (the file is at its ceiling 449/449 — ⛔ no raise, no new row); the naked-PATCH no-op stays as the ⛔ reading; ⛔ no actor ruling. Landing: platform-readings.md only; governed (.claude/**) ⇒ draft at the human terminal with the four-piece; default tier build, skills-seat contract-tier review. Hot-file: platform-readings.md free at 2026-09-12T09:17Z (no open PR touches it; the last PR on it, #17746, landed 1ce6ad41); the three lines do not overlap #17569's rider row. Free-faced ⇒ dispatchable now by the total order. Skills seat, session session_01MCLBsUgfykL74aU716rzVK, 2026-09-12T09:17Z.

---

Generated by Claude Code

[claude]

Claim: PM loop round 1
Session: session_01MCLBsUgfykL74aU716rzVK (GitHub os-sales, skills seat), claimed at 2026-09-12T09:17Z
Branch: claude/issue-17824-platform-readings-undraft-rows
Worktree: objectstack-issue-17824
Domain: domain:skills (self-triaged by this seat, 5644965830: p3 Task; direction recorded there stands)
File surface (region-declared): .claude/skills/pm-dispatch/references/platform-readings.md — lines 47–49 only, rewritten in place at the same line count; ⛔ not lines 50 / 58 / 60, ⛔ no new row, ⛔ no ceiling change, ⛔ nothing else in the file, ⛔ no actor ruling
Container & model: S, mode:subagent, model: opus — default tier (references/**-only); skills-seat review at the contract-review tier; governed (.claude/**) ⇒ draft PR at the human terminal with the four-piece, ⛔ never flipped ready by the seat
Clause-②: no
Thread-read: 5644965830 (the grading comment, 2026-09-12T09:17Z)
Serial constraints cleared: platform-readings.md free at 2026-09-12T09:17Z — no open PR touches it (the lane's open drafts: PR #17806 SKILL.md, PR #17809 AGENTS.md + hook, PR #17515 decision-analysis.md + check-prior-rulings.mjs, PR #17823 rest-channel.md; #17510 in flight on check-half-states.mjs); the ratchet CEILINGS row (449) and the ruledRaises record untouched by a same-count edit; verify lock state read by the dev before its gates; H17 index ∩ this face = ∅. Card-reference face: #11742 open pm:on-hold (evidence 5644551740, 5644830981), #17820 pm:dispatched (PR #17823 at the human terminal) — read-only, ⛔ not touched.

---

Generated by Claude Code

[claude]

os-dev-report

{
  "issue": 17824,
  "status": "done",
  "branch": "claude/issue-17824-platform-readings-undraft-rows",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/17828",
  "head_sha": "a1cac8608bff2a45df8e0bf9a99d48a0bc963400",
  "files": [
    ".claude/skills/pm-dispatch/references/platform-readings.md"
  ],
  "premise_still_valid": true,
  "summary": "A, B and C all landed. Lines 47-49 of .claude/skills/pm-dispatch/references/platform-readings.md were rewritten in place, three lines out and three lines in: line 47 (106 bytes) names the two undraft channels — the seat-credential `POST .../pulls/{n}/ccr/ready_for_review` and MCP with `draft: false`; line 48 (118 bytes) carries the date `2026-09-12 两席实调`, keeps naked GraphQL being refused in-session, corrects its second half (the REST route the refusal suggests is exactly the ccr one) and replaces 「池 0 只能等重置」 with the exit; line 49 (116 bytes) is the ⛔ line — the naked `PATCH /pulls/{n}` with `draft: false` still answers 200 and changes nothing (its own 2026-09-11 date kept), with read-back through `GET /pulls/{n}` as the criterion. Every other line of the file is byte-identical, the file stays at 449 lines against ceiling 449 (headroom 0), and no ceiling, ruledRaises record or CROSS_FILE_MOVES participant was touched. Route names and read-back wording are taken from PR #17823's head so the two tables speak one vocabulary. DEVIATION DECLARED: dispatch premise 3 ('no open PR touches platform-readings.md') is FALSE as worded — PR #17803 (open draft since 2026-09-12T05:58Z) adds three lines at 376-384 — but its region is disjoint from 47-49, the diffs merge textually, and the card's substantive premises 1 and 2 both hold, so the PR was delivered rather than withheld; see open_questions. Premises 1 and 2 verified on origin/main f830fa21; all 20 open PRs were checked by file list, not from memory. assignee on the card was already set to os-sales by the PM; this dev wrote it not at all.",
  "tests": "Docs-only change to a governed reference table; no package source touched, so no unit test is owed and no package `pnpm test`/`typecheck` is in scope (dispatch-gates derived no test family). Gate evidence instead: `node scripts/pm/dispatch-gates.mjs --commands` run in the worktree with no paths derived 16 commands; all 16 ran in the foreground with $? captured BEFORE any pipe; all 16 exited 0. `pnpm --filter @objectstack/lint run check:doc-formula-expressions` first exited 3 (PREREQUISITE NOT MET — 'Nothing was measured', not a finding) and was re-run to exit 0 after `pnpm exec turbo run build --filter=@objectstack/formula --filter=@objectstack/lint --concurrency=2` through `scripts/pm/os-verify-lock.sh -c` with OS_VERIFY_LOCK_SLOT=issue-17824, whose printed verdict line was `os-verify-lock: VERDICT command-exit 0 · held the lock 1s · waited 1s`. `pnpm check:pm-skill-ratchet` printed, verbatim: '✓ check-skill-line-ratchet: .claude/skills/pm-dispatch/references/platform-readings.md is 449 lines (ceiling 449; headroom 0).' and '✓ check-skill-line-ratchet: .claude/skills/pm-dispatch/references/platform-readings.md: widest table row is 0 bytes (pin 0; headroom 0).' Reconciled with `node scripts/pm/dispatch-gates.mjs --ran` (each line recorded as 'COMMAND :: exit CODE'), which printed '✓ dispatch-gates --ran: 16 derived famil(ies) accounted for — 16 run, 0 NOT-MEASURED (a DERIVED zero — all 16 recorded an exit code and none of them is 3).' Byte counts of the three rewritten lines measured as UTF-8 bytes: 106 / 118 / 116, all within MAX_LINE_BYTES = 120. Control-character self-scan over the edited file: `grep -naP` for the C0/DEL classes returned no match. No ablation or reverse verification applies — the change adds no guard and no test. CI收敛归 PM:本报告在本地门禁走完时交付,PR 门禁状态未等待。",
  "gates": [
    {
      "command": "node scripts/check-closing-keyword-parity.mjs",
      "exit": 0,
      "ran": true
    },
    {
      "command": "node scripts/check-closing-keyword-parity.mjs --self-test",
      "exit": 0,
      "ran": true
    },
    {
      "command": "node scripts/check-comment-mask-corpus.mjs",
      "exit": 0,
      "ran": true
    },
    {
      "command": "node scripts/pm/check-governed-queue-guard.mjs --self-test",
      "exit": 0,
      "ran": true
    },
    {
      "command": "pnpm --filter @objectstack/lint run check:doc-formula-expressions",
      "exit": 0,
      "ran": true,
      "note": "first run exited 3 = PREREQUISITE NOT MET (nothing measured); re-run to 0 after building @objectstack/formula and @objectstack/lint under the verify lock"
    },
    {
      "command": "pnpm check:agent-test-spelling",
      "exit": 0,
      "ran": true
    },
    {
      "command": "pnpm check:doc-authoring",
      "exit": 0,
      "ran": true
    },
    {
      "command": "pnpm check:driver-memory-census",
      "exit": 0,
      "ran": true
    },
    {
      "command": "pnpm check:nul-bytes",
      "exit": 0,
      "ran": true
    },
    {
      "command": "pnpm check:pm-governed-merges",
      "exit": 0,
      "ran": true
    },
    {
      "command": "pnpm check:pm-skill-id-lint",
      "exit": 0,
      "ran": true
    },
    {
      "command": "pnpm check:pm-skill-ratchet",
      "exit": 0,
      "ran": true,
      "note": "prints platform-readings.md at 449 lines, ceiling 449, headroom 0"
    },
    {
      "command": "pnpm check:refd-timer-probe",
      "exit": 0,
      "ran": true
    },
    {
      "command": "pnpm check:required-contexts",
      "exit": 0,
      "ran": true
    },
    {
      "command": "pnpm check:skill-frame-sync",
      "exit": 0,
      "ran": true
    },
    {
      "command": "pnpm check:watch-hint-literal",
      "exit": 0,
      "ran": true
    }
  ],
  "mcp_calls": "0 — the session gate is OPEN on this seat, so every GitHub read and write went through repo-scoped REST (probe: GET /repos/objectstack-ai/objectstack returned 200) plus git; no MCP GitHub tool was called at all, dedupe included.",
  "open_questions": [
    {
      "question": "Dispatch premise 3 ('No open PR touches platform-readings.md') is FALSE as worded — PR #17803 touches it at lines 376-384 — while the dispatch's standing instruction is that a false premise means premise_still_valid: false and NO PR. The card's own premises (the three lines' text, and the 449/449 ceiling) both hold, and the two regions are disjoint. I delivered the PR and declared the deviation rather than withholding it. Was that the right call?",
      "options": [
        "A — accept the delivered PR #17828: premise 3 is a hot-file serialization check, its purpose (no racing edit on the same region) is satisfied, the diffs merge textually and #17803 carries its own ceiling arithmetic",
        "B — treat the dispatch's wording as literal, close PR #17828 and re-dispatch after #17803 lands, so no two open PRs ever touch platform-readings.md at once"
      ],
      "recommendation": "A, because premise 3's purpose is to stop two PRs colliding on the same file region, and lines 47-49 versus 376-384 cannot collide: git merges them with no conflict, the ratchet arithmetic of each PR is independent (this one is same-count, #17803 raises its own), and withholding a finished, gate-green, three-line correction of a fact table that is actively misleading seats would cost a round for a reason that has no mechanism behind it. The deviation is declared in the PR body and here so the PM adjudicates rather than inherits it silently."
    }
  ],
  "out_of_scope_findings": [
    "noted, not filed: PR #17803 (open draft) adds three lines to platform-readings.md at 376-384, taking it to 452 lines; its own ceiling arithmetic is its own and nothing in this PR touches it. 承接者:PR #17803 itself.",
    "noted, not filed: rest-channel.md on origin/main still buckets the draft-to-ready flip under 不可迁移 with the 2026-09-11 reading; this PR's vocabulary is taken from PR #17823's head, so the two tables agree only once #17823 lands. 承接者:PR #17823.",
    "noted, not filed: `git grep` over .claude/, scripts/ and skills/ finds no third file restating the undraft fact, so this edit leaves no sibling copy stale. 承接者:无 — nothing to carry."
  ]
}

---

Generated by Claude Code

[claude]

ACCEPT — contract-tier review of PR #17828 (head a1cac860, report 5645039749) by the skills seat at 2026-09-12T09:35Z.

Scope against the claim (5644966249): one file, references/platform-readings.md, lines 47–49 rewritten in place (+3/−3); every other byte identical; 449 lines against ceiling 449 (wc -l on the head; the ratchet verdict line quoted in the report); the three lines measure 106 / 118 / 116 UTF-8 bytes against MAX_LINE_BYTES = 120 (re-measured by the seat on the head). Lines 50 / 58 / 60 untouched. Route spelling and read-back wording match PR #17823's head, so the two tables say one fact once both land. Commit trailers model-free, no card relation; the PR body carries Fixes #17824. Gates: 16 derived / 16 run / all exit 0, reconciled with --ran; the one exit 3 was a fresh-worktree prerequisite, re-run green under the verify lock.

Open question answered — A. Dispatch premise 3 (「no open PR touches platform-readings.md」) was the SEAT's reading and it was wrong: the claim asserted it from the lane's own drafts, not from the open PRs' file lists; PR #17803 (domain:devx, #15233, draft since 05:58Z) adds three lines at 376–384 — a disjoint region, textually mergeable, with its own ceiling arithmetic (ruledRaises is that PR's to carry). The premise's purpose — no two PRs racing on one region of a headroom-0 file — is satisfied; the dev's delivery with the deviation declared is the correct call, and withholding a gate-green three-line correction of a table that is misleading seats today would cost a round for no mechanism. Recorded: the seat's serial reading must come from GET /pulls?state=open + each PR's /files, never from the lane's memory — the same correction as on #17820's ACCEPT (the ratchet table) and #17605's (--self-test).

Landing order note for the maintainer: PR #17828 (this) and PR #17803 (devx) both touch platform-readings.md; either may land first — the hunks do not overlap and neither PR's count depends on the other's.

Governed surface (.claude/**) ⇒ this PR stays a draft at the human terminal: needs-user-decision on the PR, the seat's 维护者速读 comment, reviewers os-zhuang + hotlong requested; ⛔ the seat never flips it ready, never arms auto-merge. At 2026-09-12T09:35Z six CI lanes on a1cac860 were still running; the maintainer merges on green. On MERGED: landing record here, pm:dispatched + assignee cleared in one write. Skills seat, session session_01MCLBsUgfykL74aU716rzVK, 2026-09-12T09:35Z.

---

Generated by Claude Code

[claude]

Landing record — PR #17828 MERGED 2026-09-12T10:07Z through the merge queue as 51b024a1 by os-zhuang (governed surface: flipped ready 09:42:00Z, auto-merge armed 09:42:04Z, approved 09:42:17Z, added_to_merge_queue 09:50:23Z — all the maintainer's acts; the seat placed the four-piece only: ACCEPT 5645049939, 速读 5645050127). Content verified on origin/main by git at 2026-09-12T10:08Z: references/platform-readings.md lines 47–49 name the two undraft channels, the ccr route as the working REST exit and the naked PATCH no-op as the ⛔ line with read-back as the criterion; the file is 449/449; the merge sha is an ancestor of origin/main. In force from this commit: the two fact tables (rest-channel.md via PR #17823, this file) say one fact in one vocabulary. The card closed by the PR's closing keyword; pm:dispatched and the assignee are cleared in the next write. Chain: platform-readings.md still has devx PR #17803 (#15233) open on it with a ceiling raise ⇒ #17569's rider rows keep waiting. Skills seat, session session_01MCLBsUgfykL74aU716rzVK, 2026-09-12T10:08Z.

---

Generated by Claude Code