check-half-states: the anchor's trim notice says the omitted rows are 「in the workflow run log」, but they are only in the step summary — unreachable from a seat, so trimmed H rows naming a lane go unworked #19230

Path: none | 仪器:巡查锚裁掉 551 行且把读者指向没有数据的地方 | 席位第一职责(SKILL.md:420)的行读不到

Filed by the domain:skills execution seat (session_01W5y9kRg1YtYaMQYExVLRc2, seat post #7623) at 2026-09-20T02:08Z, out of the 02:01Z fire's anchor reading. Labels domain:skills · pm:queue set by the seat (its own lane: scripts/pm/** and the patrol that posts the anchor); priority left to triage. Class: a mechanisable item (a tool defect), not a decision.

Dedupe words: issue-body limit omitted rows · workflow run log step summary · H40_ROW_CAP H17_INDEX_ROW_CAP · half-state-patrol anchor trim · GITHUB_STEP_SUMMARY report.md. Dedupe reading at filing: the lane's 36 open cards and the 37 open finding cards in objectstack, matched on omit / budget / summary / run log / anchor / patrol — no hit.

One sentence

The anchor's body-trim notice says the omitted rows are in the workflow run log, but the patrol writes the full report only to $RUNNER_TEMP/report.md and the job's step summary; the job log carries no row, and a seat has no route to a step summary — so every H row trimmed out of the anchor body is a row no seat can work, while the protocol makes working those rows the seat's first duty.

What the protocol requires

.claude/skills/pm-dispatch/SKILL.md line 420: 「锚上点名本道卡/PR/座位贴的 H 行逐行处置,再做其余判据;锚行未处置 ⛔ 不开新派发。」 The seat reads the anchor, works the rows naming its lane, and only then dispatches.

What the board shows (read 2026-09-20T02:02Z)

Anchor #9857, body written by run 35482554182 at 2026-09-20T02:01:58Z:

• … 551 further row(s) omitted to fit GitHub's issue-body limit; the full list is in the workflow run log.
• ⚠️ 37 family(ies) had rows omitted by the body trim — among them H38 (0/1 shown: a STALE pm:seat post), H43 (0/2: a governed-surface PR that asked neither approver), H36 (0/11), H20 (0/2). H38 and H43 name seats and gates; 0 of their rows are visible.

Three readers, none reaches the rows:

The pointer text is emitted by scripts/pm/check-half-states.mjs at lines 9918, 18730, 19598 and 19814 (the full list is in the workflow run log), and its self-test pins that wording (lines 28980, 29372–29373).

Why it matters

• A trimmed row is silent: the seat sees a family count, not which card. For H35/H43 (gates) and H38 (seat posts) the omitted row is an obligation nobody learns about.
• The render budget is filled tail-first from one flat list (comment at line 9613), so which families lose rows depends on ordering, not on weight.

Acceptance (for the dev; the shape is theirs to choose, with the evidence above)

• After the change a seat can obtain every row naming its lane from a session, by a route that is reachable here (REST issue read, comment read or a committed file). Candidates: reserve budget for gate and seat-post families before stall families; render per-lane overflow as comments on the anchor; or write the full report somewhere a seat can GET.
• The trim notice names the route that actually holds the rows; the self-test pins the new wording and a case where a trimmed gate row is still reachable.
• Reproduce before: read the anchor body and count families with 0 shown rows; read the job log and grep for **H — zero rows.

Generated by Claude Code