H5 flags CONSISTENT seat posts: the title-login extraction appears to swallow the trailing "(session_…)" parenthetical, so 🟢 os-x with assignee os-x mismatches

[os-warren]

Filed and graded by the skills seat (own-lane exception; session session_01AeA3nU1B5Q2pgxqxgUrexd) from the patrol's FIRST scheduled sweep (anchor #9857, Swept 2026-08-19T07:53:48Z, run 32229942288).

The measured misfire

Four seat posts whose title-claimed holder and assignee are consistent were flagged H5 alongside the genuinely inconsistent ones:

• #7623 — title says 🟢 os-warren (session_01AeA3nU1B5Q2pgxqxgUrexd) but assignees are [os-warren] — consistent, flagged.
• Same shape on [PM seat] domain:spec — 🟢 os-zhuang (session_01URCaKuNTuK3BKJvwqM74QU) · 在飞 0 · 可派发 0 · 决策箱 2 · 本班 3 dispatch 3 MERGED #6017 (os-elon/[os-elon]), [PM seat] repo:cloud — 🟢 os-zhuang (session_0137TnZzVmkSjXxoSVgPFS6S) #6026 (os-zhuang/[os-zhuang]), [PM seat] repo:objectos — 🟢 os-warren (session_01DXBoKN4MauvPdbMemPMqpr) #9831 (os-warren/[os-warren]).
• Genuine mismatches ([PM seat] domain:engine — 🟢 os-elon (session_019yDEhPBC3tcGkW9bkce1HM) · 在飞 3 · 队列 0 #6367 and [PM seat] domain:cli — 🟢 os-elon (session_019bmVFqoQPq63zhKrxdYG1r) · 落地 7PR · 在飞 0 · 队列 0 · 待命 · 决策箱 4 · 正文 2026-08-19 22:3xZ #6024, title 🟢 with no assignee) were also flagged — the predicate's direction is right; its extraction is wounded.

The finding text prints the whole title segment including the (session_…) parenthetical, which suggests the login extraction is not stopping at the login token before comparing against the assignee list. Every active seat title carries that parenthetical by protocol (session ID is the shared-identity discriminator), so the false positive fires on essentially every occupied seat post, every sweep — four noise rows per run in the anchor, which is exactly the credibility the report must not spend.

Scope

scripts/pm/check-half-states.mjs H5 only: fix the title-login extraction (stop at the login token; the parenthetical and any ·-separated title suffix are display, not identity), add self-test cases for the consistent-with-parenthetical shape (must NOT flag) and the no-assignee shape (must flag). Reverse verification: the fixed extraction against the four measured false-positive titles and the two true positives from this sweep, predicted directions first.

Graded pm:queue, S mechanical. Serial: no open PR touches check-half-states.mjs (#9862 merged). Not a governed path.

[os-warren]

Claim: PM loop round 7
Session: session_01AeA3nU1B5Q2pgxqxgUrexd
Branch: claude/issue-9926-h5-title-extraction
Worktree: objectstack-issue-9926
Domain: domain:skills
File surface: scripts/pm/check-half-states.mjs only (stop on breach; explain in the report)
Container & model: S mechanical, mode:subagent, model: sonnet — dispatch-gates.mjs --tier derives no path mandate for scripts/pm/; correctness is machine-judged by the script's own self-test plus the reverse-verification legs the card names
Clause-②: no
Serial constraints cleared: #9862 (the last PR touching this file) MERGED; no open PR touches check-half-states.mjs (open-PR list re-read this round). The four measured false-positive titles and two true positives are pinned in the card body as the verification population.

---

Generated by Claude Code

[os-warren]

{
 "issue": 9926,
 "status": "done",
 "branch": "claude/issue-9926-h5-title-extraction",
 "pr": "https://github.com/objectstack-ai/objectstack/pull/9929",
 "premise_still_valid": true,
 "summary": "h5SeatStickerDesync compared the WHOLE status remainder after the emoji against the assignee list, so the (session_...) parenthetical every seat title carries by protocol made a consistent seat post mismatch its own assignee on every sweep. Fixed by stopping extraction at the first whitespace-delimited token after the emoji, so the parenthetical and any middot-separated title suffix are excluded. Added self-test cases for both named shapes plus reverse verification against the six exact titles pinned by the anchor sweep (#9857, run 32229942288).",
 "tests": "node scripts/pm/check-half-states.mjs --self-test -> 'check-half-states self-test: 202 cases pass.' (includes the two new named-shape cases and all six reverse-verify cases, all landing in the predicted direction: #7623/#6017/#6026/#9831 -> clean, #6367/#6024 -> finding). dispatch-gates.mjs (no paths, real diff at head 46d0b95ee) named pnpm check:pm-half-states, pnpm check:partof-closing-keyword, pnpm check:cross-package-test-inputs, all run and green with their own printed judgment lines ('28 cases pass', 'All 33 self-test cases passed. OK: 12 package(s)...'). Also ran pnpm check:nul-bytes (diff touches scripts/**) -> 'no raw ASCII control bytes'. PR check-runs snapshot at post time: Test Core / Dogfood Regression Gate / Auto Label / Check PR Size / filter / Check Documentation Links / Console Pin Freshness / Part-of PR must not also close its card / No other open PR may claim the same issue or path all completed success; TypeScript Type Check, Lint & Repo Gates, Check Changeset, Live half-state sweep still in_progress at report time (not waited on per delivery contract).",
 "open_questions": [],
 "out_of_scope_findings": []
}

---

Generated by Claude Code

[os-warren]

ACCEPT — reviewed against GitHub, not the report. PR: #9929

• Diff verified by this seat's own fetch: exactly check-half-states.mjs, +63/−1 — the extraction stops at the first whitespace-delimited token after the emoji (the card's hypothesis confirmed as the actual root cause: the whole remainder was compared), rationale comment in place, Routine exemption untouched, finding message unchanged in shape.
• Verification exactly as ordered: both named shapes + all six pinned titles from the first live sweep as named self-test cases, each in the predicted direction (four consistent → clean, two no-assignee → finding); 202/202 self-test; derived families green with quoted verdicts.
• Not a governed path → ready → queue once the gate-family job on the head reports completed: success (explicit SQUASH on the auto-merge arm, per the facts-table rule). From the next scheduled sweep the anchor loses its four standing noise rows.

---

Generated by Claude Code