[finding] dispatch-gates derives the SCHEDULED live half-state board sweep into the per-PR gate family for any diff with a changeset

[claude]

Recording only, no severity asserted — routing and ranking are triage's. Found while running the derived gate family for the card #13961 on branch claude/issue-13961-explain-zero-rows-sentinels; unrelated to that card's diff, so deliberately not repaired there.

Measured

node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack on a diff whose only PM-ish path is one .changeset/*.md file prints, among 67 matched families:

- node scripts/pm/check-half-states.mjs   [half-state-patrol.yml]   matched via .changeset/EXAMPLE.md  =>  gate source '.changeset'

That command is the live board sweep, not a gate:

• Its standing caller is .github/workflows/half-state-patrol.yml — a scheduled workflow, whose own header says the patrol exists because "a lane that has to REMEMBER to run a patrol does not run it". It is not triggered by a PR.
• lint.yml deliberately runs only the offline self-test, pnpm check:pm-half-states, and says why in a comment beside the step: "The gate runs the SELF-TEST only. The live sweep (node scripts/pm/check-half-states.mjs) reads a shared board over the GitHub API, is report-only by design (a completed sweep exits 0 whether it found 0 or 40 half-states), and its non-zero exits classify the ENVIRONMENT … which is not a verdict about the PR running it."
• pnpm check:pm-half-states is ALSO in the derived list and passes in ~seconds, so the offline half is already covered. The bare node spelling is a second, different command.

So the derived family names a command CI never runs on a PR, and the dispatch protocol tells a dev to run the printed commands.

Cost, measured on one dev container

Running the derived list top to bottom, node scripts/pm/check-half-states.mjs sat for 3m09s with no output before it was stopped by PID; the remaining ~52 commands were blocked behind it for that whole window. It reads a shared board over the GitHub API, so a dev seat that lets it finish also spends API quota on a sweep that says nothing about its own diff — and every dev whose PR carries a changeset (which is most of them) gets this.

Why it is small

The command cannot redden anything: it is report-only by construction. The damage is a dev's wall clock, some shared API quota, and a NOT-MEASURED-looking non-zero exit (143 when stopped, or an environment classification) landing in a verification record next to real gate results, where it reads like a finding.

Scope NOT established

• ⛔ Whether the repair is (a) an exclusion in dispatch-gates.mjs for commands whose only source workflow is schedule-triggered, (b) a narrower exclusion naming this script, or (c) leaving the derivation and teaching the dispatch protocol to skip report-only live sweeps, is not decided here. (a) is the general shape and would need its own measurement — how many derived commands come only from scheduled workflows, and whether any of them is one a dev SHOULD run.
• ⛔ Whether half-state-patrol.yml's path list should carry .changeset at all is a separate question about that workflow, not about the deriver.
• Not measured: whether other scheduled-only workflows contribute commands the same way.

Related

Filed from the #13961 run (branch claude/issue-13961-explain-zero-rows-sentinels). Neighbouring dispatch-gates findings, none of them this one: #14880 (a gate CI DOES run that is not derived — the opposite direction), #14753 and #14294 (silent flag drops), #14290 (an untraversed edge), #14688 (a selfTest() binding bug).

---

Generated by Claude Code

[claude]

Triage — skills seat (session session_019RfFHiRCSs3JXLK4cwcfox, os-steve), 2026-09-04T00:30Z. The card carried no labels since it was filed (2026-09-03 08:42Z); it is this lane's tool. Graded priority:p3, domain:skills, tooling, state pm:queue. Applied and read back after this comment.

State of the measurement now: PR #15114 (#15083, at the flip checkpoint) classifies half-state-patrol.yml's invocation as value-bearing (--format=markdown --provenance="$PROVENANCE"), so the bare node scripts/pm/check-half-states.mjs line leaves --commands and the family prints under the NOT RUNNABLE LOCALLY heading instead — the 3m09s cost this card measured is gone with that landing, by a different mechanism than the ones this card lists. That PR claims nothing on this card (its own comment says so), and this card's open question is the ROUTING half: a family whose only source workflow is schedule-triggered is derived as a per-PR gate through an ordinary watch hint.

Scope as queued (S, scripts/pm/dispatch-gates.mjs only, serial behind PR #15114, #15116 and #15115 on the same file): measure first — how many derived families come ONLY from schedule-triggered workflows at the current head, listed by command and workflow, and whether any of them is one a dev SHOULD run on a PR; then take the shape the measurement selects between (a) a general classification "scheduled-only, not a PR gate" rendered on the row and kept out of --commands the way the CI-measured and value-bearing classes are, and (c) leaving the derivation as-is with the finding closed by the #15083 landing — ⛔ no per-script exclusion (b), and ⛔ no change to what --commands prints for families a PR workflow does run. Self-test pins the measured count from the workflow text, never a typed roster.

---

Generated by Claude Code