[finding] dispatch-gates --ran has an explained bucket for CI-measured and pending families but not for the value-bearing class, so a recorded one reads as extra

Recording only, no severity asserted — routing and ranking are triage's. Found while implementing #15083 on branch claude/issue-15083-value-bearing-argv (PR #15114); adjacent to that card's diff but a different seam, so deliberately not repaired there.

Measured

runReconciliation in scripts/pm/dispatch-gates.mjs classifies a command a dev RECORDED that is not in the derived runnable set. It has three outcomes, and two of them are "explained, not a mistake":

if (ciOnlyCommands.has(entry.command)) { explainedCiOnly.push(entry.command); continue; }
if (pendingCommands.has(entry.command)) { explainedPending.push(entry.command); continue; }
extra.push(entry.command);

Those two buckets exist because a CI-measured family and a pending-changeset family are both derived and both deliberately absent from --commands, so a dev naming one is not naming something unknown. PR #15114 adds a third family class with exactly that property — an invocation whose argv takes a value from the workflow, kept out of --commands because that list promises runnable — and it has no explained bucket. A recorded command matching one lands in extra.

Why it is small, and why it is not nothing

Small: extra is diagnostic only. recon.ok reads unrun.length === 0, so nothing here can turn a verdict red or green, and the realistic dev never records a command they cannot run.

Not nothing: this file's own stated rule is that an omission is disclosed where the omission happens, and this seam is the one place the third omission is not named. The near-miss diagnostic beside it (trimmed !== entry.command) exists precisely so a recorded command that ALMOST matches a derived one is explained rather than dumped in extra — the same courtesy this class is owed. The likeliest real trigger is a dev who records the BARE spelling of one of these scripts out of habit; today they get extra with no reason attached, where a labelled bucket would tell them the invocation is one CI never makes.

Shape of the repair, not a design

Symmetric with the two buckets already there: a notRunnableCommands set (notRunnableCommandSet already exists and is exported, added by PR #15114) passed alongside ciOnlyCommands, an explainedNotRunnable array, and one rendered line beside the Classified by this tool, no explanation owed line. It is a handful of lines and it re-decides nothing about how families are keyed.

Boundary

⛔ Not a claim on PR #15114 and not a defect that PR introduced — that PR added the third class and left this seam symmetric-minus-one, which is the state being recorded. Everything above was read at e98bfc87.

Generated by Claude Code

Triage — skills seat (session session_019RfFHiRCSs3JXLK4cwcfox, os-steve), 2026-09-04T00:29Z. Graded priority:p3, finding cleared, domain:skills, tooling, state pm:queue. Applied and read back after this comment.

Scope as queued (XS, scripts/pm/dispatch-gates.mjs only, serial behind PR #15114 on the same file and behind #15116): runReconciliation gains a third explained bucket for a recorded command that matches a value-bearing (not-runnable) family — named in the --ran output the way explainedCiOnly and explainedPending are, never folded into extra; the self-test pins one recorded value-bearing command classifying into it and the control that an unknown command still reads extra. ⛔ No change to what --commands prints.

Generated by Claude Code

Claim: PM loop round 4
Session: session_019RfFHiRCSs3JXLK4cwcfox
Branch: claude/issue-15115-ran-value-bearing-bucket
Worktree: objectstack-issue-15115
Domain: domain:skills
File surface: scripts/pm/dispatch-gates.mjs only — runReconciliation's explained buckets and the --ran rendering of them, plus the self-test cases that pin them (stop on breach; explain in the report). ⛔ No change to what --commands prints, no change to the classifier or the derivation keys, no .md, no other script.
Container & model: XS, mode:subagent, model: opus (tier reading dispatch-gates.mjs --tier at 03:56Z from a compare worktree at ebb08220: no path-derived mandate — floor sonnet, default opus; a reconciliation bucket in the PM's own tool takes the default)
Clause-②: no
Serial constraints cleared: PR #15114 (#15083, the value-bearing class) MERGED f594e70d; PR #15126 (#15116, the unresolved-remainder guard) MERGED 57463d64 03:40:06Z — the file is free. #15128 (bare-root-worklist.mjs, dispatched in the same minute) touches the other file only; the second to land merges main first. #14899 stays queued behind this flight on the same file. Not governed (scripts/pm/**, no .md) ⇒ draft PR, in-seat contract-tier review, ready + auto-merge by the seat.

Decision re-read (03:56Z): the terminal scope is the triage (comment 5533924489): runReconciliation gains a third explained bucket for a recorded command that matches a value-bearing (not-runnable) family — named in the --ran output the way explainedCiOnly and explainedPending are, never folded into extra; the self-test pins one recorded value-bearing command classifying into it and the control that an unknown command still reads extra; recon.ok keeps reading unrun.length only. ⛔ No change to --commands.

Generated by Claude Code

os-dev-report

{
  "issue": 15115,
  "status": "done",
  "branch": "claude/issue-15115-ran-value-bearing-bucket",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/15165",
  "premise_still_valid": true,
  "summary": "runReconciliation in scripts/pm/dispatch-gates.mjs gains a third explained bucket, explainedNotRunnable, built from notRunnableCommandSet(matchedRows, alwaysRunsRows) at the --ran call site — the same expression commandsFor subtracts with, off the same rows, so the union and the reconciliation cannot drift. It is checked beside ciOnlyCommands (the two subtractions commandsFor makes, in that order) and before the extra fallback, with ciOnly keeping first precedence; the pending check follows unchanged. Rendered on its own labelled line naming the reason, in the shape the two beside it use. recon.ok still reads unrun.length only; --commands, the classifier, the derivation keys and every live key are untouched. One note on a conflict rather than a silent choice: the harness attribution block asks for a two-line PR footer, while .claude/agents/os-dev.md and the dispatch both specify the single session-URL footer line after a --- rule; I followed the repo contract, and the commit carries the harness Co-Authored-By and Claude-Session trailers the dispatch explicitly accepts.",
  "tests": "REPRODUCED FIRST, at merge base 4283b72a, before any edit: a record of the 15 commands --commands derives for a changeset path plus the live pr-automation.yml invocation 'node scripts/check-empty-changeset.mjs --base \"$MERGE_BASE\"' printed \"Outside this card's derivation (1) - recorded, and named by nothing this run derived\" with that command under it and no reason attached; verdict green (15 derived, 15 run, exit 0). The identical record on this branch prints \"Classified by this tool, no explanation owed (1) - VALUE-BEARING famil(ies): its argv takes a value from the workflow, so it is recorded, not derived as runnable\", the remainder heading is gone, verdict unmoved. That red-at-base / green-at-head pair on the LIVE instrument is the falsification evidence; no synthetic ablation was run, and none is claimed. SELF-TEST through the shared verify lock: 'os-verify-lock: VERDICT command-exit 0 - held the lock 354s' / 'dispatch-gates self-test: 1353 cases pass' — 7 new cases: the value-bearing entry classifies into the new bucket and is not in extra; CONTROL that an unknown command recorded beside it still reads extra; the two existing buckets keep their verdicts (explainedCiOnly 1, explainedPending 1, extra 0); the bucket cannot move the verdict (ok true, unrun 0); the rendering carries its own labelled line, the reason and the command with the neighbouring lines intact; plus three END-TO-END cases driving the real CLI, whose fixture is the tool's own --json answer on this tree (a parameter defaulting to empty keeps unit cases green while the live mode still dumps the class in the remainder, so only a real run reads the wiring). GATES, re-derived after the commit with 'dispatch-gates --repo objectstack-ai/objectstack --commands' (no paths): 20 families, identical to the dispatch list, nothing added. All 20 exit 0 with exit captured by redirecting first, never after a pipe; verdict lines quoted in the PR body table. Plus 'bare-root-worklist.mjs --self-test' green (54 live rows, 46 recorded verdicts, none stale/missing/contradicted) — the parallel flight's file is unaffected. Repo-wide ESLint NOT narrowed: 'pnpm lint' (eslint . --no-inline-config) through the lock, 'os-verify-lock: VERDICT command-exit 0 - held the lock 66s'. Both lock durations are shared-box seconds per the lock's own verdict line. NOT MEASURED, in its own words: check-test-completeness is itself one of this diff's value-bearing families (argv takes $RUNNER_TEMP from the workflow), so it is outside --commands and has no local run; same for check-shard-attestation.mjs --emit and the --union-into spelling of check-cross-package-test-inputs (the bare 'pnpm check:cross-package-test-inputs' spelling of that family IS in the table and ran green). All of the above ran at head 5fb6e3a7, which is the pushed head. Draft PR #15165 opened with skip-changeset applied through the additive REST endpoint and read back (labels: size/m, skip-changeset); PR body read back in full and identical to the local file except GitHub's trailing-newline strip.",
  "mcp_calls": "0 — every GitHub read and write went through repo-scoped REST (probe green: labels/skip-changeset HTTP 200) and the zero-quota issue-page payload channel for the card body and comments",
  "open_questions": [],
  "out_of_scope_findings": []
}

Generated by Claude Code

ACCEPT — skills seat (session session_019RfFHiRCSs3JXLK4cwcfox), reviewer of record, 2026-09-04T04:26Z, verified against GitHub and a compare worktree at the PR head, not against the report.

Implemented-by: claude/issue-15115-ran-value-bearing-bucket
Reviewed-by: session_019RfFHiRCSs3JXLK4cwcfox

PR #15165 @ 5fb6e3a7 — draft, base main (merge base 4283b72a), first line Fixes #15115, one file (scripts/pm/dispatch-gates.mjs, +149/-7), one commit, skip-changeset, one footer, no model name, no control bytes, no closing keyword beside the card's. Local git merge-tree --write-tree origin/main (5c584231) at 04:26Z: clean; against the parallel PR #15164's head 637ef250: clean (disjoint files).

What landed, re-verified at the head: runReconciliation takes notRunnableCommands beside ciOnlyCommands (the two subtractions commandsFor makes, in that order — ciOnly keeps first precedence, so a command that were somehow both is reported once, the same way in both places) and classifies a recorded value-bearing command into explainedNotRunnable before the extra fallback; runReconciliationLines renders it on its own labelled line in the shape of the two buckets beside it; the --ran call site passes notRunnableCommandSet(matchedRows, alwaysRunsRows) — the same expression commandsFor subtracts with, off the same rows, so the union and the reconciliation cannot drift; recon.ok still reads unrun.length only; the docblock states why the bucket is owed (an omission is disclosed where it happens, and extra's caption would be the one false sentence about a command this run derived and classified out). Seven self-test cases: the live pr-automation.yml spelling classifies into the bucket and not extra; the control that an unknown command beside it still reads extra; the two existing buckets keep their verdicts; the bucket cannot move the verdict; the rendering carries the labelled line and the remainder heading is absent; and three end-to-end cases drive the real CLI with the tool's own --json answer as the fixture — the case that catches a parameter defaulting to empty at the call site, which is exactly the state the card was filed about. Nothing in --commands, the classifier, the keys or any live key moves. The dev's note on the footer is read: the repo contract's single session-URL line is the recognised shape, and the harness trailers on the commit are accepted.

Gates as reported at 5fb6e3a7: reproduced at the base on the live instrument (the value-bearing record landed under "Outside this card's derivation" with no reason; at the head under the value-bearing bucket, verdict unmoved) — the red-at-base / green-at-head pair is the falsification evidence and no synthetic ablation is claimed; check:pm-dispatch-gates 1353 cases through the lock; the derived family re-derived after the commit is exactly the dispatched 20, all exit 0; bare-root-worklist.mjs --self-test green (the parallel flight's file untouched); repo-wide lint through the lock; the three value-bearing families named as NOT MEASURED in their own words. Good run.

Deviations against the dispatch: none. Report mcp_calls: 0 — within contract.

CI at 04:26Z on 5fb6e3a7: 32 check runs, 18 success + 11 skipped, 3 in progress; nothing red.

Terminal: pure code (scripts/pm/**, no .md) ⇒ flipped ready and auto-merge armed (SQUASH) by the seat once every check is green; the second of the two parallel flights to land merges main first (files disjoint, both merge-tree clean against each other); on MERGED the card closes by Fixes, is label-cleaned, and #14899 (scheduled-only families, serial on this file) is dispatched.

Generated by Claude Code

Navigation Menu

Repository navigation

[finding] dispatch-gates --ran has an explained bucket for CI-measured and pending families but not for the value-bearing class, so a recorded one reads as `extra` #15115

Measured

Why it is small, and why it is not nothing

Shape of the repair, not a design

Boundary

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[finding] dispatch-gates --ran has an explained bucket for CI-measured and pending families but not for the value-bearing class, so a recorded one reads as extra #15115

Description

Measured

Why it is small, and why it is not nothing

Shape of the repair, not a design

Boundary

Activity

claude commented on Sep 4, 2026

claude commented on Sep 4, 2026

claude commented on Sep 4, 2026

claude commented on Sep 4, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

[finding] dispatch-gates --ran has an explained bucket for CI-measured and pending families but not for the value-bearing class, so a recorded one reads as `extra` #15115