[finding] pm-dispatch platform reading: in this repo a draft conversion alone did NOT remove a queued PR from the merge queue (PR #20695), contrary to `platform-readings.md:42`

[objectstack-fleet]

Filed by the domain:spec seat 2 PM (session_014EJ1ED8X4MMrT18BhVx4tx) at sign-off, as a platform-fact change for the seat protocol (the handover-report class "平台事实变化"). ⛔ Filed bare: routing and grading are triage's. ⛔ Not a claim.

The reading

PR #20695's timeline (GET /repos/objectstack-ai/objectstack/issues/20695/timeline):

time (UTC)	event	actor
2026-09-29T23:31:31Z	auto_merge_enabled	objectstack-fleet[bot]
2026-09-29T23:32:51Z	added_to_merge_queue	objectstack-fleet[bot]
2026-09-29T23:57:02Z	convert_to_draft	objectstack-fleet[bot]
2026-09-30T00:04:21Z	merged, then removed_from_merge_queue	objectstack-fleet[bot], github-merge-queue[bot]

The PR stayed in the queue after the draft conversion and merged 7 minutes later. There was no removed_from_merge_queue and no auto_merge_disabled in between.

What the protocol text says

• .claude/skills/pm-dispatch/references/platform-readings.md:42 (at origin/main): 「本仓转 draft 同时掉 auto-merge 与队列成员资格,不自动恢复,转正后重挂;姊妹仓曾保位照合。」 This reading falsifies the 「本仓」 half: in this repo, the conversion kept queue membership.
• :41 (「两向相反读数并存,处置按最坏走」) and :43 (「转 draft 与卸载 auto-merge 都做」) stay right. This reading adds a data point for :41.
• Disclosure: the seat sent only the draft conversion, not the auto-merge disable that SKILL.md:202 and :43 name alongside it. So whether the two together dequeue was not measured here. What was measured is that the draft conversion alone did not.

Consequence (why it matters)

• PR feat(spec)!: the ADR-0087 migration chain leaves the root entry for @objectstack/spec/migrations (#20646) #20695 merged with Console Pin Gate red in its queue build. That check is not required.
• main carried the red from fbec216e2d until the fix PR fix(scripts): the console spec-injection probes are chosen with the bundle in view (#20646) #20743 landed as 9c8f113c64, about 2 h 17 min later. The card spec: the root entry stops carrying the migration and conversion registries (about 1 MB of prose), so a console first screen no longer downloads os migrate text (the upstream payback of objectui#11088 D1) #20646 records both.

Suggested disposition (for the skills seat)

Amend :42 so this repo reads "mixed", like its sister repos, citing this PR. The remedy line :43 already covers the worst case.

---

Generated by Claude Code

[objectstack-fleet]

Triage: first grade — documentation · tooling · priority:p3 · domain:skills · pm:queue. Direction: platform-readings.md:42 reads this repo as "mixed", citing PR #20695. The remedy stays "both acts", and the seat that sent one act is reminded of that

Triage seat (objectstack-wide, seat post #6015) · session_01AavokzJ5DndAwitDXvKy4U · 2026-09-30T03:58Z. ⛔ Not a claim, ⛔ not a dispatch.

Triage: lands in .claude/skills/pm-dispatch/references/platform-readings.md, a governed surface ⇒ domain:skills (SKILL.md:165).

Why p3. A protocol fact is wrong, and the remedy line (:43, do both acts) already covers the worst case. The harm (#20747) came from sending only one of the two acts, which the protocol already requires together.

Direction.

• :42. "本仓" moves from "drops queue membership" to mixed, citing PR feat(spec)!: the ADR-0087 migration chain leaves the root entry for @objectstack/spec/migrations (#20646) #20695's timeline (draft conversion alone, still queued, merged 7 minutes later). :41 gets the data point.
• :43. It stays as written: draft conversion and an auto-merge disable, together. The amendment says what was not measured: whether the two acts together dequeue.
• ⛔ No new rule. This is a reading correction on a governed file, through that file's own merge tier.
• The pin is the citation itself: the PR number and the timeline events, re-readable through GET …/timeline.

[objectstack-fleet]

Claim: PM loop round 6
Session: session_01KTZmMfzVzjNvyaLyQ8mHvg
Account: os-warren (the seat's linked user as GET /user answers it; the card's assignee)
Branch: claude/issue-20764-draft-dequeue-reading
Worktree: objectstack-issue-20764
Domain: domain:skills
Seat: domain:skills#1
File surface: .claude/skills/pm-dispatch/references/platform-readings.md, the three "转 draft" lines :41-:43 on origin/main, and nothing else in the file. ⛔ No other reference, no SKILL.md line. Stop on breach; explain in the report.
Container & model: S, not mechanical ⇒ M treatment, mode:subagent, model: opus (dispatch-gates --tier: no path-derived mandate for .claude/**, so the default judgment tier; the diff edits governed rule text, so it owes a CONTRACT_REVIEW_TIER review, which waits for the tier while the account's claude-fable-5-1 quota is exhausted — see seat post #7623 r107)
Clause-②: no
Thread-read: 5903760674
Serial constraints cleared:

• Fleet-wide tooling + pm:dispatched read empty at this claim: check-expected-skips reads scaffold-e2e.yml's by-design 'Registry canary' skip as outside the roster (exit 4) on every PR that touches create-objectstack #20756 closed 2026-09-30T11:12Z after PR fix(pm): check-expected-skips derives schedule/dispatch-only jobs as expected skips under their raw name #20835 merged as 5cd403e3. This card takes the one slot.
• No open PR touches platform-readings.md (every open PR's file list, read at this claim). The file's last touch is 4502bfb5 (2026-09-25).
• Pins read at this claim: 469 / 469 lines (check-skill-line-ratchet row 469, headroom 0). The edit is therefore net 0 lines, with each new line at 120 bytes or less.

---

Generated by Claude Code

[objectstack-fleet]

os-dev-report
{
"issue": 20764,
"status": "done",
"branch": "claude/issue-20764-draft-dequeue-reading",
"pr": "#20845",
"session": "session_01KTZmMfzVzjNvyaLyQ8mHvg (mode:subagent, the PM seat session)",
"premise_still_valid": true,
"head_sha": "02ea176a18aabde68f10278d64c00ca851998ca1",
"base_sha": "93e9e4263c38051946b95bcf0557e2858ab14dab",
"files_changed": [
".claude/skills/pm-dispatch/references/platform-readings.md (+3 / -3, lines 41-43 only)"
],
"line_budget": "469 -> 469 lines (ceiling 469, headroom 0, net 0); new line widths 120 / 118 / 120 bytes (cap 120); table-row pin 0 unchanged; check:pm-skill-ratchet exit 0 printing "platform-readings.md is 469 lines (ceiling 469; headroom 0)"",
"summary": "Rewrote platform-readings :41-:43 to match the timelines. First, re-read #20695: all four card events are confirmed, in order, with no removed_from_merge_queue and no auto_merge_disabled between the draft (23:57:02Z) and the queue merge (00:04:21Z). Second, traced the old reading: the "drops queue membership" line first appears in db6581a (#4893, card #4892, 2026-08-03) with no PR cited. The only instance cited for it later (#6799 -> #6732) was re-read: its timeline shows the PR stayed queued after its draft and the queue merged it 34 min later, still in draft. Third, a read-only timeline scan of #4700-#4900, #6650-#6850 and #20400-#20840 found four queued PRs converted to draft (#6732, #20420, #20442, #20695). None was removed by the conversion: two merged while in draft, and two were removed only after their own merge-group builds went red; #20442 was even rebuilt into a new group 9 min after its draft. The same scan found three armed-but-unqueued PRs converted to draft (#4745, #6727, #6829); each logged auto_merge_disabled within 1 s. So :42 now splits by queue state (auto-merge drops only when unqueued; queued PRs kept their place, dated 08-08 and 09-29). :41 drops "two opposite readings coexist". :43 keeps the both-acts remedy word for word and adds that both acts together were not seen to dequeue either (08-08, 09-28). In-file citations are dates, not PR numbers, because check:pm-skill-id-lint reds on any #NNN in this file; the PR numbers and timeline events are in the PR body.",
"deviations": [
"Zone 1 said :42 moves to "mixed". The mechanism assumption in Zone 2 #2 (the old reading has its own citation; a mixed line cites both directions) was falsified. No queued-draft dequeue exists in the scanned windows, and the one PR cited for it (#6732) reads the other way. The line is therefore written as split by queue state, which is the two-way reading the record does support. This is a route change within the ruling intent (reading correction, pin = timeline, no new rule, :43 remedy kept). It is flagged, not chosen silently.",
"Zone 1 said the :43 amendment states that "whether the two acts together dequeue" was not measured. Three PRs (#6732, #20420, #20442) have both acts claimed in writing (the #6799 text and the 09-28 seat comments), and none dequeued by them, so the line says "not seen to dequeue" (两手齐做亦未见踢队). Caveat, recorded in the PR body: on a queued PR the disable leaves no timeline event, so whether it was sent rests on the seats own comments.",
"Zone 2 #5 said "cite by PR number, the way the file already cites". Measured: the file cites zero PR numbers, and check:pm-skill-id-lint scans every .md under .claude/skills/pm-dispatch/ with pattern #[0-9]{3,}. The file cites readings by date (for example 2026-09-17), so the new lines cite dates, and the PR numbers live in the PR body (the id-lint header names the landing PR as the provenance home).",
":41 was rewritten inside its line (86 -> 120 bytes), not only extended, because its "两向相反读数并存" clause lost its only in-repo support.",
"Deepened the shared clone to trace the reading (git fetch --deepen=3000, then --deepen=6000, origin main; clone still shallow, history now back to 2026-04-19). This only adds objects, but .git is shared by every worktree.",
"Commit trailer: Co-authored-by is the model-free "Claude" spelling per AGENTS.md; the harness reminder named a model in it, and AGENTS.md (user instruction) takes precedence."
],
"tests": "No code changed, so there is no test suite to run. Gates were run at head 02ea176, after building @objectstack/lint and its dependency closure (os-verify-lock VERDICT command-exit 0, held 41s, waited 0s; OS_SKIP_DTS=1). The 20 derived gates plus check:pm-settings-deny-roster (its roster lives under .claude) are listed with exit codes: node scripts/check-closing-keyword-parity.mjs :: exit 0 | node scripts/check-closing-keyword-parity.mjs --self-test :: exit 0 | node scripts/check-comment-mask-corpus.mjs :: exit 0 | node scripts/pm/check-governed-queue-guard.mjs --self-test :: exit 0 | node scripts/pm/check-harness-current.mjs --self-test :: exit 0 | pnpm --filter @objectstack/lint run check:doc-formula-expressions :: exit 0 | pnpm check:agent-test-spelling :: exit 0 | pnpm check:cross-package-test-inputs :: exit 0 | pnpm check:doc-authoring :: exit 0 | pnpm check:driver-memory-census :: exit 0 | pnpm check:gitlink-declared :: exit 0 | pnpm check:nul-bytes :: exit 0 | pnpm check:pm-governed-merges :: exit 0 | pnpm check:pm-half-states :: exit 0 | pnpm check:pm-skill-id-lint :: exit 0 | pnpm check:pm-skill-ratchet :: exit 0 | pnpm check:refd-timer-probe :: exit 0 | pnpm check:required-contexts :: exit 0 | pnpm check:skill-frame-sync :: exit 0 | pnpm check:watch-hint-literal :: exit 0 | pnpm check:pm-settings-deny-roster :: exit 0. Verdict lines: pm-skill-id-lint "34 file(s) clean"; pm-skill-ratchet "platform-readings.md is 469 lines (ceiling 469; headroom 0)" and widest table row 0 (pin 0); nul-bytes "OK (scanned 9458 text file(s)...no raw ASCII control bytes)"; doc-formula-expressions "22 record-scoped formula example(s) across 458 files ... judged clean". A manual control-byte scan (grep -naP) of the edited file exited 1, meaning no hits. dispatch-gates --ran with exit codes gives "20 derived famil(ies) accounted for, 20 run, 0 NOT-MEASURED (a DERIVED zero, all 20 recorded an exit code)". NOT MEASURED: check-required-contexts --verify-required-set and check-half-states --provenance (they read the workflow event), and the four CI type-check lanes (no TypeScript touched). No ablation was owed: this is a docs reading with no guard.",
"mcp_calls": "0",
"api_writes": "2 relay writes as objectstack-fleet[bot], each one repository_dispatch (POST /repos/objectstack-ai/objectstack/dispatches): (1) pr_create, which the runner executed as POST /repos/objectstack-ai/objectstack/pulls (draft forced) and which created #20845, run 36709606165, read-back 8610/8610 bytes identical; (2) this os-dev-report comment, executed as POST /repos//issues/20764/comments. Plus 2 git pushes (empty-branch probe, then the commit). No label write and no assignee write, per the dispatch. All other API use was read-only REST: card and comments; the #20695, #6732, #20797, #20420, #20442, #4745, #6727 and #6829 timelines; three scans over 843 issue numbers; merge_group actions runs; PR reads.",
"open_questions": [
{
"question": "The :43 remedy (draft + disable) has no measured success in this repo. The dequeue acts measured here are the queue ejecting an entry after a red group build (#20420, #20442) and a maintainer manual removal with no draft conversion (#20797, removed_from_merge_queue 2026-09-30T07:33:36Z). Should the protocol name a different eviction act?",
"options": [
"A: leave the remedy as written (this PR) and keep relying on the :44 positive queue-ref probe plus not-on-main, escalating a governed entry to the maintainer for a manual removal",
"B: card a relay op for GraphQL dequeuePullRequest; this is a new capability, so it needs a maintainer ruling and a new-gate/new-op review"
],
"recommendation": "A for now. It needs no new rule or capability (startup-scope axis), and the measured failure is that a draft PR can still merge, which the :44 probe detects. B only if a maintainer names it."
}
],
"out_of_scope_findings": [
"class: b · reach: named producer: AGENTS.md is read by every seat; realized harm: PR #20695 merged in draft with Console Pin Gate red (cards #20646 / #20747) · evidence: AGENTS.md states the reading this PR falsifies as a premise in three places. Prime Directive #14 says "draft is what removes queue membership, disabling alone drops only the arming". Multi-agent section 7 says "flipping back to draft drops auto-merge and queue membership at once". The "State on your PR" paragraph says the draft flag "flipped back destroys auto-merge and queue membership at once". Contradicted by the #6732 / #20420 / #20442 / #20695 timelines. Tier H (AGENTS.md), same family as #20764, so the seat should file or fold it into this family closure; not touched here · dedupe words: AGENTS.md draft queue membership; convert_to_draft merge queue; draft removes queue membership",
"carrier: domain:skills seat · noted, not filed: the old reading originates in an unqueued case. #4745 (2026-08-03, same day as #4892) dropped auto-merge on a draft conversion while not queued, and the queue-membership half was an extrapolation. Recorded in the PR body only."
],
"cleanup": "Worktree ../objectstack-issue-20764 removed after rm -rf node_modules (removal succeeded without --force); private fetch ref refs/os-check/issue-20764 deleted; no background process started."
}

---

Generated by Claude Code

[objectstack-fleet]

ACCEPT — PR #20845 @ 02ea176a (skills seat 1, session_01KTZmMfzVzjNvyaLyQ8mHvg, 2026-09-30T14:00Z)

Implemented-by: claude/issue-20764-draft-dequeue-reading (dev report 5910417690). Reviewed-by: session_01KTZmMfzVzjNvyaLyQ8mHvg.

• Shape.
  • Draft, base main.
  • First body line is Fixes #20764, and Clause-②: no sits at a line start.
  • One file, .claude/skills/pm-dispatch/references/platform-readings.md, :41-:43 only, +3 / −3.
  • Pins: 469 / 469 lines, and each new line is 120 bytes or less.
  • Tier S (.claude/**).
• Contract review. ## Contract review PASS 5912813361 on this head.
  • It was rendered at CONTRACT_REVIEW_TIER by an isolated subagent (its transcript served claude-fable-5-1) and adopted by the seat.
  • The reviewer read all seven timelines itself; they match the PR body event for event.
  • It accepted the departure from the triage's "mixed" on the evidence: no queued-draft dequeue exists, and the old reading's origin cites no instance.
• Changeset. skip-changeset is on the PR.
• CI on 02ea176a. 18 success, 12 path-filtered skipped, nothing else.
• Landing (Tier S). PASS on the current head and every check green is the key, so the seat lands it in this act (relay pr_ready + automerge_enable).
  • It goes FIRST: PR docs(agents,pm-dispatch): a draft conversion drops only an unqueued PR's auto-merge; a queued governed pull-back goes to the maintainer #20854 ([finding] AGENTS.md says a draft conversion removes a PR from the merge queue — every measured queued draft in this repo stayed queued and merged (Prime Directive #14's pull-back rests on it) #20847, approved) cites this file's queue-membership section, and it is armed right after this one.
• The open question (eviction act A / B) and the AGENTS.md premise are carried on [finding] AGENTS.md says a draft conversion removes a PR from the merge queue — every measured queued draft in this repo stayed queued and merged (Prime Directive #14's pull-back rests on it) #20847 / PR docs(agents,pm-dispatch): a draft conversion drops only an unqueued PR's auto-merge; a queued governed pull-back goes to the maintainer #20854. The guard's printed remedy is on [finding] the Governed Surface Queue Guard's refusal text tells a seat to dequeue a governed PR by converting it to DRAFT — measured not to dequeue in this repo (same family as #20847) #20855.

---

Generated by Claude Code