The #14878 absence pin covers `packages/` only — ruling D's full tree radius needs the dispatch-gates residue measurement redone first

[claude]

Split out of #14878 by the domain:devx seat (session session_012zGPuVVX3deAx9LdjK8jCk) at 2026-09-03T22:0xZ, as the measured residual of that card's ruling D.

What ruling D asked for, and what landed

The ruling (#14878 comment 5528359266, maintainer 「同意」) says the absence pin widens «from one file to the tree». PR #15085 landed it at packages/ only. This card is the rest.

Why it stopped at packages/ — measured, not chosen

Declaring an examples/-wide .ts glob in scripts/cross-package-test-inputs.mjs makes that glob an inherited watch hint on every importer of that table, check:cross-package-test-inputs included. dispatch-gates.mjs's self-test pins that no hint of that gate reaches a test file outside packages/** — precisely because it is listed as a change-KIND rather than a path derivation, since the hint route cannot reach the population it judges.

Measured on that tree: all 41 tracked test files outside packages/ are under examples/. So one examples-wide glob does not shrink the residue class that case guards — it empties it, and the case's own remedy ("re-point at another member of its class") is impossible because no other member exists.

⇒ The implementing dev correctly refused to edit that self-test case: doing so would weaken a gate to fit a declaration, in a scripts/pm/ file its card did not own. It narrowed its own pin instead and wrote the cost into the pin's docblock.

Why this is not urgent

On origin/main at filing time, every surviving actionObjectKey reference is under packages/** — zero examples/ hits, #14669's showcase site having already been fixed. The delivered pin therefore has complete coverage of live sites; what is missing is prospective coverage of a new stale mention written under examples/.

What this card actually needs — ⛔ not a two-line change

1. Redo the residue measurement behind dispatch-gates.mjs's "no hint outside packages/**" self-test case. That is a scripts/pm/ decision and may belong to another lane — route accordingly rather than assuming it is this one's.
2. Only then widen, and widen in two places, never one: SCANNED_ROOTS in both pin files (packages/objectql/src/action-owner-key-single-source.test.ts, packages/runtime/src/action-owner-key-single-source.test.ts) and the package globs in scripts/cross-package-test-inputs.mjs. A new top-level root needs a matching glob or the pin scans a radius nothing re-runs it for — the failure mode the declaration exists to prevent.
3. ⛔ Editing that self-test case is not the repair. The pin's own docblock says so at :126–:148; read it before starting.

⚠️ The same reasoning applies to any other widening — docs/, content/, skills/, apps/ — not just examples/.

Acceptance criterion

SCANNED_ROOTS covers the ruling's full radius in both pins, check:pm-dispatch-gates is green without its self-test case having been edited, and an ablation shows the pin redding on a planted stale mention under the newly covered root — not merely passing.

Provenance: #14878 ruling D (5528359266); implementation PR #15085; ACCEPT and scope-shortfall note on #14878. domain:* and grading left for triage.

---

Generated by Claude Code

[huangyiirene]

Triage — domain:devx, as the designated single owner of a cross-domain card. tooling · priority:p3 · pm:queue kept. Read 2026-09-04T07:26Z.

You asked triage to "route accordingly rather than assuming it is this one's" on step 1. Taking that seriously produced a cross-domain answer rather than a single-lane one, so this is a designation, not a plain lane label.

The file surface spans three lanes

file	lane by the table
scripts/pm/dispatch-gates.mjs (step 1, the residue measurement)	domain:devx
scripts/cross-package-test-inputs.mjs (the globs)	domain:devx
packages/objectql/src/action-owner-key-single-source.test.ts	domain:engine
packages/runtime/src/action-owner-key-single-source.test.ts	domain:cli

⇒ Under the lane table's cross-domain exception — 「真拆不动的跨域单 PR 由分诊座位指定一个车道 PM 认领」 — this seat designates domain:devx, on two grounds:

1. The gating decision is devx's subject. scripts/ gates split by SUBJECT: governance of the agent-instruction / governed surface goes to domain:skills, code-and-doc quality stays devx. This pin's subject is actionObjectKey single-sourcing — code quality. ⛔ It is not the scripts/pm/check-governed-merges.mjs case, which the table names explicitly for skills.
2. Step 1 gates steps 2–3. The residue measurement decides whether widening is possible at all, so the lane that owns it should hold the card rather than receive it.

⚠️ The two SCANNED_ROOTS edits in objectql and runtime are mechanical constant changes riding this card, ⛔ not a reason to split it — the ruling's radius is one decision and splitting it would let the two pins diverge, which is the failure the single-source pin exists to prevent. The claiming seat should declare the full file surface in its claim comment, per the cross-domain path.

⛔ Preserved, not re-litigated

• ⛔ Editing the self-test case is not the repair — the pin's docblock says so at :126–:148, and the previous dev was right to refuse. Read it before starting.
• Widen in two places, never one: both pins' SCANNED_ROOTS and the globs in scripts/cross-package-test-inputs.mjs. A root without a matching glob gives a pin a radius nothing re-runs it for.
• The acceptance criterion stands as written, including that the ablation must show the pin redding on a planted stale mention under the newly covered root — ⛔ passing is not evidence.

p3 confirmed against your own reasoning: every surviving actionObjectKey reference is under packages/**, so live coverage is complete and only prospective coverage is missing.

---

Generated by Claude Code

[claude]

Claim: PM loop round R11 — triage designated domain:devx as the single owner of this cross-domain card (5537138325). Declared full file surface, per the cross-domain path: scripts/pm/dispatch-gates.mjs (step 1 — the residue MEASUREMENT only, ⛔ its self-test case is not edited), scripts/cross-package-test-inputs.mjs (the globs), packages/objectql/src/action-owner-key-single-source.test.ts and packages/runtime/src/action-owner-key-single-source.test.ts (both pins' SCANNED_ROOTS, widened together or not at all). Step 1 gates steps 2–3: if the measurement shows the ruled radius cannot be declared without changing that gate's contract, the deliverable is the measurement + a fork for the maintainer, pr: null — not an edited gate.
Session: session_012zGPuVVX3deAx9LdjK8jCk (domain:devx execution seat)
Branch: claude/issue-15097-absence-pin-tree-radius
Worktree: objectstack-issue-15097
Domain: domain:devx (designated; touches domain:engine / domain:cli pin files mechanically)

Assignee read as empty BEFORE the write; assign + pm:queue → pm:dispatched as one read-modify-write with comparative read-back; race re-read follows.

---

Generated by Claude Code

[claude]

Claim: os-dev execution seat, dispatched by PM claim 5537366546 (PM loop round R11, domain:devx designated single owner per triage 5537138325).

Restating the declared four-file surface, unchanged from the dispatch:

• scripts/pm/dispatch-gates.mjs — READ ONLY, step 1 residue measurement only. Its self-test case is not edited.
• scripts/cross-package-test-inputs.mjs — the globs.
• packages/objectql/src/action-owner-key-single-source.test.ts — SCANNED_ROOTS.
• packages/runtime/src/action-owner-key-single-source.test.ts — SCANNED_ROOTS.

Nothing else. Step 1 gates steps 2-3: if the measurement shows the ruled radius cannot be declared without changing that gate's contract, the deliverable is the measurement plus a fork for the maintainer with pr: null — not an edited gate.

Session: session_012zGPuVVX3deAx9LdjK8jCk
Branch: claude/issue-15097-absence-pin-tree-radius (pushed empty at claim time)
Base: 460134af8

Comments re-read to the last page before writing: no competing Claim: present (2 comments total — triage 5537138325, PM claim 5537366546).

---

Generated by Claude Code

[claude]

os-dev-report

{
  "issue": 15097,
  "status": "needs_decision",
  "branch": "claude/issue-15097-absence-pin-tree-radius",
  "pr": null,
  "premise_still_valid": true,
  "summary": "STEP 1 (the measurement, on origin/main 460134af8; scripts/pm/dispatch-gates.mjs read-only, diff empty). The case is the #11199 block's XPKG battery: specimens OUTSIDE_PACKAGES='examples/app-crm/test/smoke.test.ts', TSX_TEST='packages/client-react/src/realtime-hooks.test.tsx', APPS_TEST='apps/docs/src/x.test.ts', plus the class-level case xpkgResidue.length greater than 0. NUMBERS: gate carries 106 inherited hints; tree has 8287 tracked files and 3197 tracked test files; RESIDUE = 13, not 41 — 10 under examples/ and 3 .tsx under packages/client-react/ (the battery prints '(13)' itself). Tests outside packages/ = 41, ALL under examples/, 0 under scripts/ (the dispatch's '39 examples + 2 scripts' is wrong: isTestFilePath admits .ts/.tsx only and scripts/ holds no .test.ts). 31 of those 41 are ALREADY reached, via the inherited 'examples/app-showcase/**' hint that @objectstack/example-showcase declares. WHICH DECLARATIONS THE CASE REFUSES: 'examples/**/*.ts' and 'examples/**' red neg1; 'apps/**/*.ts' reds neg3; 'examples/*/src/**/*.ts', 'examples/**/src/**/*.ts', 'docs/**/*.md', 'content/**/*.mdx', 'skills/**/*.md', '.changeset/**/*.md' red NOTHING; a bare 'examples' is refused by the extractor and contributes no hint at all. ADMISSIBLE SHAPE: only partially, and not for the ruled radius. (a) docs/, content/, skills/ are admissible at zero cost — already inherited hints of this gate, 0 test files under them, already in ci.yml crosspkg — but INERT, because SCANNED_EXTENSION is the single string '.ts' and those roots hold 0 .ts files (docs 0 ts / 205 md, content 0 / 408, skills 0 / 46); adding them to SCANNED_ROOTS widens the declared radius while scanning zero extra files, which is coverage in appearance only. Making them live needs SCANNED_EXTENSION to become a set — a shape change beyond this card's SCANNED_ROOTS scope. (b) examples/ has NO admissible whole-root shape; the only admitted one is 'examples/*/src/**/*.ts' (153 of 210 examples .ts), leaving the 57 outside */src/ — the test/, e2e/ and config files — uncovered, i.e. exactly the sub-tree where the historic miss lived (one of the five surviving references was a test under the showcase example). (c) apps/ is refused by neg3, whose class is documented as already having no tracked member, so it cannot be re-pointed even in principle. (d) A hint-scoped or masked declaration is CLOSED: extractWatchHints masks only comments and self-test bodies, there is no ignore directive in the file, and check:watch-hint-literal exists precisely to forbid computing a declaration so it stops contributing a hint (57 cases, 8 computed spellings rejected). => No shape delivers the ruled radius without changing that gate's contract, so pr is null and the fork is below. THREE CORRECTIONS to the card's stated basis, all measured: (1) 'that glob does not shrink the residue class, it EMPTIES it' is FALSE — xpkgResidue goes 13 to 3 and the class-level case stays GREEN; what empties is the narrower class the SPECIMEN stands for (test files outside packages/**, 41 to 0). Exactly ONE case reds, and it is the specimen, not the class guard. (2) Both pins' docblocks say a new top-level root needs a matching ci.yml crosspkg entry gated by check-ci-filter-parity.mjs. That does NOT bind for examples/: 'examples/**' is already in ci.yml's core filter and the parity gate's SCHEDULING_FILTERS is ['core','crosspkg'] judged as an OR. Measured with the glob planted, node scripts/check-ci-filter-parity.mjs exits 0 — 'OK: all 144 declared cross-package glob(s) (100 unique) are covered by core or crosspkg'. No .github/workflows/ci.yml edit is required. (3) The 2 files now naming actionObjectKey outside packages/ are .changeset/cli-action-dedup-global-key-constant.md and .changeset/converge-standalone-action-owner-key.md — both LEGITIMATE past-tense records of the removal (#14667 / #14422), and already excused by construction by NOT_A_STALE_MENTION's existing file.startsWith('.changeset/') rule, which the pin's docblock says is kept for exactly the day the extension set widens. Nothing to disposition; the card stays prospective, p3 confirmed.",
  "tests": "All exit codes captured by redirect BEFORE any pipe; verdicts quoted from each gate's own printed line. Heavy runs went through bash scripts/pm/os-verify-lock.sh -c (slot issue-15097-dev), detached per dispatch-gates.mjs's own header ban on running --self-test in the foreground on an agent container, and POLLED to completion inside this turn — no run was left in the background. POSITIVE CONTROL, clean tree (pnpm check:pm-dispatch-gates, lock held 360s): '+ but no hint of this gate reaches a test file outside packages/**' PASSES, alongside '+ the tree still holds test files no hint of this gate reaches (13), so the entry is not redundant', '+ nor a .tsx test file inside it', '+ nor one under apps/**, the class with no tracked member to lose'. That run's ONLY failure is 'x CONTROL: --changed --commands is legal and answers, so the moved usage line describes a real combination' — ALREADY FILED as #15278 (reds on any branch with no diff against origin/main; this branch has none). It is not the case under study and not caused by this work. ABLATION, 'examples/**/*.ts' planted into @objectstack/objectql's globs (lock held 402s): EXACT REFUSAL LINE = 'x but no hint of this gate reaches a test file outside packages/**', summary 'x dispatch-gates self-test: 1 of 1353 case(s) failed.' In the SAME run '+ the tree still holds test files no hint of this gate reaches (3), so the entry is not redundant' — the class-level case SURVIVES the widening — and the #15278 CONTROL case PASSES here, because the plant gives the branch a non-empty changed set. The two runs therefore fail DIFFERENT single cases, which is what isolates the widening's cost to exactly one specimen. MUTATION CONFIRMED ON DISK each time, not by an editor exit code: HEAD blob 32daef3c5dfa1d2acb3fa95eaff1b17d95b0356a recorded, python anchor asserted to hit exactly 1 occurrence, injected text counted with grep -cF (=1) and git diff HEAD --name-only non-empty; an unconditional trap on EXIT/INT/TERM restored via git checkout HEAD -- ABSOLUTE_PATH (never a bare git checkout --, which restores from the index), and every restore printed 'git diff HEAD is []'. A first attempt was correctly read as a VOID no-op because grep without -F treated the glob's '*' as regex; it was rerun, and this is reported rather than quietly retried. Final tree: git diff HEAD --stat EMPTY — dispatch-gates.mjs and every other file unmodified. PIN SUITES (lock held 381s, after pnpm --filter '@objectstack/objectql^...' --filter '@objectstack/runtime^...' build): objectql action-owner-key-single-source.test.ts 'Test Files 1 passed (1) / Tests 5 passed (5)'; runtime same file '1 passed (1) / 16 passed (16)'. FAST GATES all exit 0: check:cross-package-test-inputs ('All 117 self-test cases passed.' + 'OK: 26 package(s) read outside themselves, all declared, and turbo.json hashes every declared glob.'), check:declared-population-live ('201 of 252 famil(ies) ... reaches this tree's 8287 tracked file(s)'), check:watch-hint-literal ('57 cases pass -- 8 computed spellings rejected'), check:nul-bytes ('scanned 8280 text file(s) ... no raw ASCII control bytes'). NOT MEASURED, reported as such: 'pnpm check:ci-filter-parity' is NOT a script name — it exits 254 with 'Did you mean \"pnpm check:pnpm-filter-targets\"?', i.e. nothing ran. The gate's real spelling, per lint.yml:3198, is node scripts/check-ci-filter-parity.mjs, which was run directly and exits 0. DERIVED FAMILY: node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack exits 2 with an EMPTY command list, which is the correct reading for a zero-edit changeset — there is no diff to gate. Control-byte scan (grep -naP over the class) clean on this report and on the claim comment; both posted bodies read back after posting.",
  "mcp_calls": "0 — every GitHub read and write went through repo-scoped REST with $GH_TOKEN; no MCP GitHub tool was called. Global /search/issues is blocked for this session ('sessions are bound to their configured repositories'), so the dedup search used the repo-scoped issues list endpoint (536 open issues) plus local grep.",
  "open_questions": [
    {
      "question": "The ruled radius cannot be declared without changing check:cross-package-test-inputs' hint contract. Which fork? (Step 1 confirms the card's blocking premise; it also shows only the SPECIMEN case blocks, not the class guard.)",
      "options": [
        "A. Accept packages/ as the radius and land only the prose corrections. Cheapest, no gate edit, no ci.yml edit. Live coverage is already complete and the 2 outside mentions are already excused by NOT_A_STALE_MENTION's .changeset/ rule. Cost: prospective coverage of a NEW stale mention under examples/ stays absent — the hole the pin's docblock already prices. Residual work = correct the 'it EMPTIES it' claim (both table entries and both pin docblocks) and the ci.yml crosspkg claim, both now measurably wrong.",
        "B. Re-point the specimen at the class that did NOT empty. The case's own stated remedy is 're-point at another member of its class', and as the code computes it (xpkgResidue) the class keeps 3 live tracked members (the packages/client-react .tsx tests) under any examples widening. Retiring OUTSIDE_PACKAGES in favour of TSX_TEST is ONE line in dispatch-gates.mjs — the edit this card forbids the dev to make, so it is the maintainer's to rule. Not weakening-to-fit in the usual sense (the guarded class genuinely still has members), but it does drop 'outside packages/**' as a distinct guarded reason, and the docblock says one specimen per residue reason.",
        "C. Widen to examples/*/src only ('examples/*/src/**/*.ts' plus a matching examples/*/src pathspec in both pins' SCANNED_ROOTS). Admissible TODAY with no gate edit and no ci.yml edit; covers 153 of 210 examples .ts. Cost: the 57 files outside */src/ — test/, e2e/, configs — stay out, which is exactly the sub-tree the historic miss lived in, so it buys the half the evidence says matters least while reading as an examples widening.",
        "D. Change the gate's contract — mark a declaration non-hint-contributing, or give the pin a population table the gate reads directly instead of inheriting. A scripts/pm/ design decision; as stated it collides head-on with check:watch-hint-literal, whose whole purpose is to forbid a declaration spelled so it stops contributing a hint."
      ],
      "recommendation": "A, plus the prose corrections, with B as the follow-up if prospective examples coverage is judged worth a dispatch-gates.mjs edit. C buys the wrong half of examples/ and leaves a pin that reads tree-wide with a hole precisely where mentions have historically been written; D is a contract change disproportionate to a p3 PROSPECTIVE gap. A leaves the tree honest: the pin's docblock already prices the hole, and what is actually wrong today is only the recorded REASONING for it, which this measurement corrects. Note the corrections are this card's own step-1 output, so they belong on #15097 rather than a new card — recording them here so they cannot vanish under any fork."
    },
    {
      "question": "Widening to docs/, content/ and skills/ is admissible at zero cost but INERT while SCANNED_EXTENSION is the single string '.ts' (those roots hold 0 .ts files). Should SCANNED_EXTENSION become a set?",
      "options": [
        "Leave it. A root added to SCANNED_ROOTS that scans zero files is coverage in appearance only, and is worse than not widening.",
        "Turn SCANNED_EXTENSION into a set covering .md/.mdx and widen the three prose roots with it. The pin is already provisioned for this: NOT_A_STALE_MENTION's CHANGELOG.md and .changeset/ rules are documented as kept for exactly this day, and they are what would keep the release record from re-entering as false reds. Scope: a shape change beyond this card's SCANNED_ROOTS wording, and it decides which prose counts as a live mention."
      ],
      "recommendation": "Leave it under fork A, and fold it into whichever card rules the radius — it is a separate decision from the examples/ question and should not ride in silently as a SCANNED_ROOTS edit."
    }
  ],
  "out_of_scope_findings": [
    "NOT filed as new — already open as #15278: the dispatch-gates self-test's '--changed --commands' CONTROL case reds on any branch with no diff against origin/main, which is every fresh worktree's baseline run. Hit here on the clean-tree control run and confirmed pre-existing by the dedup search, not filed again.",
    "NOT filed as a separate card — recorded in open_questions above, because they ARE this card's step-1 output and #15097 stays open under every fork: the 'it EMPTIES it' claim (scripts/cross-package-test-inputs.mjs objectql + runtime entries, and both pins' docblocks) is measurably false (residue 13 to 3, class case stays green), and both pins' ci.yml crosspkg claim does not bind for examples/ (already in the core filter; parity gate exits 0 with the glob planted)."
  ]
}

---

Generated by Claude Code