finding(ci): objectui's half-state patrol RUNS but its findings reach nobody — the run summary is unreadable from a seat, and the one exit code that is readable does not mean "clean" #9842

The half-state sweep runs on this repo four times a day and completes, but its findings are delivered ONLY to the workflow run summary — and the run summary is not reachable from an agent seat's API surface. So every execution seat's first judgement of every fire is a leg it cannot read.

⛔ This is ⛔ not a bug report against half-state-patrol.yml: running without an anchor is an explicitly supported configuration here (objectui#8740), and the workflow says so in its own notice. This card is about the consequence for this repo specifically, and it has a one-action remedy the workflow itself documents.

Measured, 2026-09-18T09:07Z

⇒ the sweep is alive, and what it found is unreadable from here. The rendered report exists only in $GITHUB_STEP_SUMMARY, which the API does not return for an Actions job.

⚠️ And the one number that IS readable does not mean what it looks like

check-half-states exited 0 is reachable through the job log. ⛔ It does not mean the board is clean. scripts/pm/check-half-states.mjs's own header says so, at its line 65 and line 80:

Why report-only, and why the exit code is ALWAYS 0 on a completed sweep

… a completed sweep exits 0 whether it found 0 or 40 violations

⇒ the exit code discriminates completed versus did-not-run, and nothing more. (Exit 2 = usage error; exit 3 = prerequisite not met.)

⚠️ Filed with the near-miss attached, because it is the point of the card: the filing seat read exited 0 and formed the belief «no half-states on the board» before reading that header. The header is what stopped it. A number that is reachable while the thing it seems to report is not is exactly the shape that produces a confident wrong reading.

What the invisibility has cost, measured on this board today

This seat found three half-states by hand during one shift, none of them via the patrol:

• objectui#9061 — pm:dispatched with no assignee since 2026-09-13, five days, because PR objectui#9138 merged with no Fixes line. It was in that state when the 07:45Z sweep ran, and ⛔ there is no way from here to tell whether the sweep flagged it.
• objectui#9782 and objectui#9710 — both auto-closed completed while still carrying pm:dispatched and an assignee (the Fixes auto-close never strips either). Both were cleaned by hand within minutes, but only because a seat happened to be watching the merge.

The remedy, quoted from the workflow rather than invented

.github/workflows/half-state-patrol.yml:257-261:

TO ADOPT (a sibling repo): change NOTHING here — set the repository variable, or set nothing and read the run summaries (objectui#8740).

and :263-265:

The anchor deliberately carries tracking and NO domain:* label: tracking is in the sweeper's own H13_EXEMPT_LABELS, so the anchor can never appear as a finding in the sweep it hosts.

Two steps: open a tracking-labelled issue in this repo carrying no domain:* label, then set the repository variable HALF_STATE_ANCHOR_ISSUE to its number (Settings → Secrets and variables → Actions → Variables).

⚠️ Step 2 is a maintainer action. A repository variable is outside an agent seat's API surface, so ⛔ no seat can do it and ⛔ no seat should be asked to. Step 1 a seat can do, but ⛔ doing it alone would leave a tracking issue nothing points at — a new half-state — so both belong in one act.

⚠️ Checked before proposing: this repo has 8 open tracking issues and ⛔ none is a patrol anchor (six carry a domain:* label, which the workflow's own text disqualifies). Control, same query shape, same run: pm:seat returns 5 rows, so the 8 is a positive result.

What would make this not worth doing

If the maintainer's intent is that objectui's half-states are healed by seats noticing them at the moment of landing — which is what happened to two of the three above — then the accepted trade the workflow already names («it notifies nobody, by accepted trade») stands and this card closes. ⇒ the call is whose eyes the findings are for, ⛔ not whether the sweep works.

Dedupe words

• half-state patrol anchor objectui unset
• HALF_STATE_ANCHOR_ISSUE repository variable
• run summary not reachable from API
• check-half-states exited 0 does not mean clean
• patrol findings notify nobody

Related

objectui#8740 (the card that made the no-anchor configuration supported) · objectui#9061 / objectui#9782 / objectui#9710 (the three half-states found by hand today)

_{filed by the domain:ui#2 execution seat · session_018HrVaotisyhgmot9o2MLRq · ⛔ this seat does ⛔ not grade or route: no priority:* and no domain:* set here · readings taken 2026-09-18T09:07Z}

Generated by Claude Code