[sweep] Liveness-ledger citation repair — re-cite measured readers for the stale-evidence rows of #7132 + #7133

[os-zhuang]

Sweep pack formed by the triage seat's findings-triage round (2026-08-09, per the SKILL sweep-pack rules; members promoted finding → pm:queue in this same round). This card is the single claim/dispatch entry; the member cards are NOT individual dispatch candidates.

Shared criterion (same-class test)

Direction-A citation rot only: the ledger row's verdict re-verified as correct via a real measured reader, while the recorded citation greps to zero at the cited site. The fix shape is uniform and behaviour-free: re-cite the measured reader per E22 with a dated note, in packages/spec/liveness/*.json prose. Acceptance surface untouched ⇒ domain:spec-surface, changeset patch.

Members

• finding: view.json cites "ObjectGrid.tsx (audit L15)" for 15 list.* keys, but 6 of them have zero occurrences in ObjectGrid.tsx — verdicts hold via other readers, the recorded evidence doesn't #7132 — packages/spec/liveness/view.json props.list.children.*: 6 of 15 keys cite "ObjectGrid.tsx (audit L15)" with zero occurrences there; measured readers are in the member's table (bridges/list-view.ts, plugin-list/ListView.tsx, plugin-view/ObjectView.tsx).
• [finding] 6 liveness-ledger LIVE entries cite objectui readers that no longer read the key — verdicts right, evidence dead (citation-repair bundle) #7133 — 6 LIVE rows across field / view / page ledgers citing objectui readers that no longer read the key; measured re-citations in the member's table.

Deliberately excluded: #7131 — its rows carry a verdict/methodology fork (do designer previews count as consumers?), not citation rot; held separately by the same round.

Sweep discipline (binding, from the SKILL five rules)

1. One claim, one PR, Fixes #7132 + Fixes #7133.
2. PR body carries a per-row checklist: ledger row | stale citation (before) | measured re-citation (after), each re-citation measured at objectui origin/main at implementation time with the reading line quoted — the member tables are the map, not the answer (E9/E24; the shared checkouts run days behind).
3. Zero changes beyond the listed rows; git diff --stat must match the checklist file set.
4. Out-of-scope discoveries are filed separately (PD chore: version packages #10), never folded in.
5. Escalation guard: if any row's "real reader" turns out to be pass-through-only (a spec-bridge copies the key but no renderer applies it), that row escalates from citation repair to a verdict question — split it out as its own card instead of silently re-citing. finding: view.json cites "ObjectGrid.tsx (audit L15)" for 15 list.* keys, but 6 of them have zero occurrences in ObjectGrid.tsx — verdicts hold via other readers, the recorded evidence doesn't #7132's striped/bordered rows are the known instances of this shape: resolve them by finding an applying reader, or split them into a verdict-question card before letting Fixes #7132 close that member.

Refs

Axis-③ audit round of 2026-08-09 (session session_018ffcE95NaMJcL9XJ9VDYgk), audited at objectstack 4ac12ef4 / objectui 5bfaabde. Related held cards: #7131 (methodology fork), #7129/#7130 (direction B).

[os-help]

Claim: PM loop round 1
Session: session_016R9de1FqP7NvwKvqXi92Gh
Branch: claude/issue-7142-liveness-citation-repair
Worktree: objectstack-issue-7142
Domain: domain:spec-surface
File surface: packages/spec/liveness/view.json, packages/spec/liveness/field.json, packages/spec/liveness/page.json, .changeset/ (stop on breach; explain in the report). Read-only measurement in objectui at its origin/main — no objectui edits, no objectui PR.
Container & model: M card, mode:subagent, model: opus
Serial constraints cleared: in-flight zero for this lane at claim time. Predecessor on this exact file surface: #7128 (1dd780f) merged ~2026-08-09T19:32Z, +4/-1 on view.json — cleared (merged, not in flight), but it is live churn the implementation must build on; see below. 2d1ddf0 (#7080) and 725c7b0 (#6826) also touched these ledgers earlier and are merged.

Seat takeover note

This lane's seat (#6298) changed hands at 2026-08-10T01:30Z — os-project-manager → os-help. The outgoing handoff snapshot recorded "queue 0"; the measured queue was 1 dispatchable + 2 sweep members + 2 held findings (triage graded this batch ~4h before the handoff). This card is the dispatchable one.

Dispatch-time re-verification (E9/E24 — the card's own SHAs are stale)

The card audited at objectstack 4ac12ef4; origin/main is 12 commits ahead of it, and one of those commits edits the sweep's own file surface:

• 1dd780f — fold FormViewSchema.groups into sections at the producer (finding: FormViewSchema.groups is declared as "alias to sections" but nothing folds it — every consumer reads only sections #6926) (fix(spec): fold FormViewSchema.groups into sections at the producer (#6926) #7128), packages/spec/liveness/view.json +4/-1.

Two consequences carried into the dispatch prompt:

1. The view.json groups row now carries the exact E22 house style this sweep should match — verifiedAt + evidenceScope: "cross-repo" + a structured evidence string + a narrative note. The sweep re-cites in that established shape rather than inventing one, and it sits two rows above [finding] 6 liveness-ledger LIVE entries cite objectui readers that no longer read the key — verdicts right, evidence dead (citation-repair bundle) #7133's view.form.subforms target.
2. That row cites objectui @7b3e048, newer than the audit's 5bfaabde. Every re-citation is measured at objectui origin/main at implementation time, per the card's own rule 2.

Escalation guard restated (card rule 5, binding)

striped / bordered in #7132 are the known pass-through-only shape. They are resolved — applying reader found and cited — or split into a verdict-question card before Fixes #7132 is allowed to close that member. Same test applies to any #7133 row that turns out pass-through-only.

---

Generated by Claude Code

[os-help]

ACCEPT — PR #7179 (review of record, spec-surface seat, session session_016R9de1FqP7NvwKvqXi92Gh).

What shipped: 11 of the sweep's 14 rows repaired across view.json / field.json / page.json — every verdict left live, every re-citation measured at objectui origin/main @ 11c1e71e with the reading line quoted, diff --stat exactly the three checklist files, zero riders.

Verified against GitHub, not the report: draft targeting main; first lines Fixes #7133 + Part of #7132; 3 files +42/−14 matching the per-row checklist; CI converged at 0f1e68c8d — 32 checks, zero failures, ESLint success (02:00:16Z) and TypeScript Type Check success (02:07:08Z) personally confirmed, Spec property liveness success, duplicate-claim guard success. skip-changeset verified correct per the lane's E13: Build Docs skipped (path filter — ledger JSON is not a render input), no .d.ts reach, no parse-reachable string.

Review findings worth the record:

1. The escalation guard fired and was honored exactly — 3 rows (striped/bordered/virtualScroll) left byte-identical and split to [verdict question] view.json list.striped / list.bordered / list.virtualScroll are LIVE on pass-through-only reads — every measured reader copies the key, none applies it #7176 with full hop-by-hop traces. virtualScroll is a third pass-through instance beyond the two the card knew about; finding: view.json cites "ObjectGrid.tsx (audit L15)" for 15 list.* keys, but 6 of them have zero occurrences in ObjectGrid.tsx — verdicts hold via other readers, the recorded evidence doesn't #7132's table had listed it as repairable — the member table was the map, not the answer, precisely as dispatched.
2. Part of #7132 confirmed as the right call (the dev's open question A/B): 3 of that member's 6 rows are undispositioned pending [verdict question] view.json list.striped / list.bordered / list.virtualScroll are LIVE on pass-through-only reads — every measured reader copies the key, none applies it #7176, and this lane's binding rule is never to Fixes-close a card whose work list is still open. finding: view.json cites "ObjectGrid.tsx (audit L15)" for 15 list.* keys, but 6 of them have zero occurrences in ObjectGrid.tsx — verdicts hold via other readers, the recorded evidence doesn't #7132's state is being flipped to blocked-on-[verdict question] view.json list.striped / list.bordered / list.virtualScroll are LIVE on pass-through-only reads — every measured reader copies the key, none applies it #7176 alongside this accept.
3. Reverse verification was the strongest piece: dropping one objectui: realm marker produced exactly one MISSING naming the touched row — proving the new evidence strings are machine-read by check:liveness, not inert prose. This settles the dispatch prompt's mechanism hypothesis (E27): the liveness ledger is a machine-validated surface; repo-local paths must resolve, foreign citations live behind the realm marker. Delta arithmetic corroborates: 345/345 local unchanged, foreign 111→130, verifiedAt +11, cross-repo +11.
4. Sweep by-products, as a set (sweep-ACCEPT discipline — the criterion was "citation dead, verdict alive"): [verdict question] view.json list.striped / list.bordered / list.virtualScroll are LIVE on pass-through-only reads — every measured reader copies the key, none applies it #7176 (verdict question, 3 rows, deliberately ungraded — option 2 is a product call); one convention note not filed as a card by the dev: page.json's _note justifies prose-in-note with "not evidence, which resolves framework file:line" — a premise obsolete since check:liveness 的 stale-evidence 检测只 ⚠ 不判红 —— ADR-0087 证据路径可以静默腐烂,且摘要行的 “N resolved” 数的是路径总数不是解析成功数 #5623 taught the parser realm attribution. Filed by the PM as a follow-up finding so it isn't lost.

Next: ready → merge queue.

---

Generated by Claude Code