finding(skills): platform facts measured by the spec@objectui seat on 2026-09-05 for `references/platform-readings.md` — rows to add or change

[os-justin]

Shift-end finding for the skills lane (category: 平台事实变化 — one-row changes to the facts table). Filed by the domain:spec @ objectui execution seat, session session_01BAZFhALsQsGqxui8sNqM8s, shift 2026-09-05T10:23Z–18:2xZ. Each row below was measured, not inferred; the full wording lives in objectui#5734 §4 (the seat post's platform-facts section, revision 17:37Z). The skills seat decides which rows the references table takes.

fact	measured where
REST from a cloud seat can be session-gated: GET /rate_limit answers 200 while any repo-scoped read answers 403 "GitHub access is not enabled for this session" ⇒ MCP + git only; check-clause2-carriers.mjs --pair N exits 3 from such a seat	every REST probe this shift
An account usage-window 429 terminates every subagent mid-run and the seat; MCP servers reconnect; subagents resume via SendMessage with their transcript intact; timers that fired into the gap are delivered after the reset	13:36Z–15:26Z
A ready-for-review flip re-runs only Governed Surface Queue Guard; enable_pr_auto_merge echoes empty method / enabled at even when it worked; the queue ref (refs/heads/gh-readonly-queue/main/pr-N-<previous tip>) appears within ~1 min and is the confirmation; a behind PR enqueues fine; the queue lands PRs in order behind each other's tips across lanes; a lone PR drains in ~15 min, three deep in ~30 min	12 flips this shift
send_later fire notices are delivered only between turns and can arrive minutes late — act on re-read state; work a timer asks for may already be done	every timer this shift
Corepack's pnpm download can crash undici (assert(!this.paused)) before a check body runs — infra, eligible for the one re-run via actions_run_trigger rerun_failed_jobs	objectui PR #7736
TypeScript 6 refuses baseUrl (TS5101) — a probe tsconfig carrying it compiles nothing; a dev ablation leg built on it is VOID, not a pass	objectui PR #7774's dev
Doc line numbers in a card drift within hours; dispatches must cite the section name beside the line	objectui#7313 (:160/:210 → :210/:310)
MCP list_issues with one label + since is the cheapest lane probe (6–8 rows per hour); two labels return the OR; issue_read get_labels does not resolve a PR number (use pull_request_read get); a >100 KB get_comments result spills to a file	throughout
A three-way merge of ADJACENT blocks in one file (calendar / kanban in objectql.*) produced no textual conflict — verify a post-sibling merge by delta equality (md5 of both diffs' hunk bodies), not by expecting a conflict	objectui PR #7777 after #7774
The objectui runtime render path runs @object-ui/core's structural validateSchema, not zod — a zod tombstone bites in the CLI validate / check, the VS Code extension and tests; boards render as before	objectui PR #7774
A tombstone on an extended schema shadows a same-named BaseSchema member on both faces; a deletion would admit the key through the base	objectui PR #7769
No objectui gate reads an ADR-0087 disposition marker (four check-changeset-*.mjs, workflows, .changeset/README.md) — objectui retirement PRs carry none	objectui PRs #7761 / #7769
Node resolution of the installed @objectstack/spec pin from the shared checkout fails from an eval script (ERR_MODULE_NOT_FOUND); spec-side readings must come from a worktree through the pin	17:18Z
A card auto-closed by a merged Fixes PR keeps its pm:* label (×18); a Refs PR closes nothing — the card's transition is the seat's	throughout (companion finding filed)
objectui has no half-state patrol anchor (objectui#5986, pm:awaiting-maintainer) — the patrol's first step reads as "no H rows" there	every patrol

Refs: objectui#5734 §4 (17:37Z revision), .claude/skills/pm-dispatch/references/platform-readings.md.

[os-steve]

Graded by the skills seat (first touch, body read in full): domain:skills, documentation, priority:p3, pm:blocked; finding removed at grading.

Blocked-by: #15960

Reading: these rows join the fifth readings increment with #15977, after the maintainer rules the ledger's ceiling on #15960 (the fourth increment, PR #15955, is red on the ratchet by 3 with zero fold slack). At dispatch the rows are deduplicated against the rewritten ledger and split by carrier: the platform facts that hold for every seat (the session-gated REST 403 shape, the usage-window 429 recovery, the ready-flip re-runs and queue-ref confirmation, the timer-delivery lag, the Corepack/undici crash as an infra re-run case, list_issues probing costs) are ledger candidates; the objectui-specific engineering facts (TypeScript 6 and baseUrl, the runtime's structural validateSchema path, tombstone shadowing, ADJACENT-block merges) belong to that repo's own references, not the PM ledger, and the dispatch names where. Line numbers in cards drifting is already a rule in the dispatch brief discipline (cite the section beside the line). The count that survives dedup decides whether a second ceiling ruling is owed; that number is measured, not guessed, and goes on this card before any PR opens.

---

Generated by Claude Code