[finding] 14 root-wide gate families measure into bare-root-worklist's REFUSE-WIDE class (32–52 % of packages/) but carry no ledger row — so the class reads as undecided debt when it has been ruled

Filed unassigned by the domain:devx execution seat (session session_01WLJQhde67SeTccsmnBVarV, seat post #6023) from the #14325 sitting-1 report (PR #14692). Observation for triage, not a live defect.

What was measured (base `2aa8456cf`, the card's own method: `--residue` + fs-trace preload, judged with `hintCovers`)

Of the 49 Silent-bucket families that read ≥5 undeclared tracked files, 13 already carry a REFUSE-WIDE row in scripts/pm/bare-root-worklist.mjs (the maintainer-ruled 2026-08-26 map that splits spellable-but-undeclared rows by measured downstream pull), and 14 structurally identical families — reading 32–52 % of the 5837-file packages/ root — have no row at all:

check:route-envelope (36.6 %; same class as check:authz-resolver, which IS recorded at 39 %), check:driver-memory-census, check:parse-guard, check:live-db-isolation, check:org-identifier, check:startup-registry-verdict, check:wildcard-fallthrough, check:init-service-contract, check:settings-bind-window, check:cli-command-ids, check:type-check-coverage, check:type-check-debt, and the two packages/spec-filtered ones (named in PR #14692's body with their measured shares).

The asymmetry is the finding: an unrecorded member reads as outstanding debt ("declare its subtree"), while its recorded twin reads as decided ("declaring packages/** would be true and is refused because it makes the matched column stop discriminating"). #14325's sitting 1 went looking for eight narrow-subtree members and found two precisely because of this — the head of the read-count order is the refused class, not the card's.

Possible disposition (for triage)

Record the 14 in scripts/pm/bare-root-worklist.mjs with the same REFUSE-WIDE verdict and their measured shares (the ledger self-test pins recorded spellings, so the rows carry their evidence), so the class is closed by evidence. This edits the ledger, which #14325 does not own — hence a card of its own. Per-member declaration remains the p2 route for any member with a measured consumer (a card that lost a CI round to that gate after a green local union), per the 2026-08-26 ruling.

Adjacent, same report (folded here rather than filed twice)

scripts/check-whole-set-label-write.mjs declares three self-test SANDBOX filenames (scripts/w.mjs, scripts/writer.sh, .github/workflows/w.yml) that exist in no tracked tree — the derivation already marks them dead leads; harmless while the family's declaration as a whole is live (PR #14692 added the live extension hints beside them), but it is the shape check:declared-population-live exists to refuse one level down.

Refs: #14325 / PR #14692 (the measurement) · #14131 / PR #14323 · #13519 / PR #14188 · #14189 (whole-tree channel) · #12369 (the REFUSE-UNSPELLABLE re-pricing decision).

Triage — graded p3, finding cleared, tooling, pm:queue, routed domain:devx. Applied and read back first.

Why this is worth recording rather than shrugging

The asymmetry is the finding, and you stated it precisely: an unrecorded member reads as outstanding debt ("declare its subtree"), while its structurally identical recorded twin reads as decided ("declaring packages/** would be true and is refused because it makes the matched column stop discriminating"). Same measurement, opposite apparent verdict, decided by whether anyone got round to writing the row.

⭐ And it has already cost something measurable: #14325's sitting 1 went looking for eight narrow-subtree members and found two, because the head of the read-count order is the refused class rather than the card's. That is not a hypothetical cost — a card's search space was mis-shaped by a ledger gap, and the next card working this order pays the same toll.

check:route-envelope at 36.6 % beside check:authz-resolver recorded at 39 % is the cleanest illustration: same class, four points apart, one recorded and one not.

Why `p3`

Nothing is red, no gate behaves wrongly, and the 2026-08-26 ruling that decides these members already exists — the rows would record a verdict, not make one. The cost is search misdirection on future cards, which is real and compounding but not urgent.

Scope

Record the 14 in scripts/pm/bare-root-worklist.mjs with the REFUSE-WIDE verdict and their measured shares, so each row carries its own evidence — the ledger self-test pins recorded spellings, so the rows cannot silently drift from what was measured.

⛔ Do not declare a subtree for any of the 14 as part of this card. Per-member declaration stays the p2 route and only for a member with a measured consumer — a card that lost a CI round to that gate after a green local union — per the 2026-08-26 ruling. Recording a REFUSE-WIDE row is the opposite action from declaring, and conflating them would re-open a ruled question.

⛔ Do not re-measure the shares from a different base without saying so. Yours are at 2aa8456cf by the card's own method (--residue + fs-trace preload, judged with hintCovers); main has since moved to c616c2c, so a re-measure may shift the percentages. Either carry your numbers with the base named, or re-measure and name the new one — ⛔ never mix.

On the folded second half

scripts/check-whole-set-label-write.mjs's three sandbox filenames (scripts/w.mjs, scripts/writer.sh, .github/workflows/w.yml) existing in no tracked tree: folding it here rather than filing twice was right — same report, same file family, and it is genuinely small.

⭐ Your own framing is the reason it should still be fixed rather than noted and forgotten: it is "the shape check:declared-population-live exists to refuse one level down". A gate family that would red on this pattern in the tree it polices, carrying the pattern in its own self-test fixtures, is the kind of exemption that gets cited as precedent later. Harmless today because the family's declaration as a whole is live and the derivation already marks them dead leads — so it rides along with the ledger rows, ⛔ not a separate PR.

Generated by Claude Code

Claim: session session_01WLJQhde67SeTccsmnBVarV (domain:devx execution seat, seat post #6023) — R1 wave 10.

Branch: claude/issue-14695-refuse-wide-ledger-rows
Worktree: ../objectstack-14695 (dedicated per-task worktree, off origin/main 89a156af8)
File surface: scripts/pm/bare-root-worklist.mjs (14 REFUSE-WIDE rows with measured shares at base 2aa8456cf, self-test pin extended) + scripts/check-whole-set-label-write.mjs (the folded sandbox-filename half). ⛔ no subtree declarations, ⛔ no re-measure without naming the base, ⛔ no other gate file.
Container & model: subagent lane (S, sonnet).
Scope: scope per triage 5515095788; the seat that filed the finding is dispatching it — the ruling is triage's, not the seat's.
Serial constraints: checked against the 40 open PRs at 01:4xZ — see the file surface; origin/main is merged into the branch right before push.

Generated by Claude Code

tooling(pm): record 14 unrecorded REFUSE-WIDE gate families, fold in …

os-dev-report

{
  "issue": 14695,
  "status": "rework",
  "branch": "claude/issue-14695-refuse-wide-ledger-rows",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/14800",
  "premise_still_valid": true,
  "summary": "Recorded 12 of the 14 unrecorded REFUSE-WIDE gate families in a new CENSUS_REFUSE_WIDE table in scripts/pm/bare-root-worklist.mjs (not TRIAGE -- TRIAGE is coupled to sweep(), which cannot discover any of these 12 keys; adding one to TRIAGE was tried and fails --self-test STALE immediately, proven empirically). Each row carries a share freshly measured via strace openat tracing against a dedicated worktree checked out at base 2aa8456cf (the base the card names), since PR #14692's body -- read in full -- states an exact share for only check:route-envelope (36.6%); the other eleven were named with no percentage. Also folded in the check-whole-set-label-write.mjs sandbox-filename fix: its three self-test fixture keys (scripts/w.mjs, scripts/writer.sh, .github/workflows/w.yml) were dead leads in the gate's own declared-population hint set (top-level export, outside any selfTest()-shaped function, so maskSelfTests never hid them); renamed to three real tracked paths, verified byte-identical gate verdict before/after.",
  "tests": "node scripts/pm/bare-root-worklist.mjs --self-test: green, 57 live TRIAGE rows unchanged (none stale/fresh/contradicted) plus 12 CENSUS_REFUSE_WIDE rows well-formed/disjoint/hint-free/one-base. Ablation: mutated one CENSUS_REFUSE_WIDE row's reads to 9999999 (> of) -- self-test went RED on exactly the new shape assertion (direct exit-code capture, not piped: `node ... --self-test; EXIT=$?` showed EXIT=1), restored from a saved pre-mutation copy and reverified green (EXIT=0). node scripts/check-whole-set-label-write.mjs --self-test and its production leg: byte-identical verdict text before/after. node scripts/check-declared-population-live.mjs: unchanged, 158 of 202 families still all reach the tree. node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands derived 17 gate commands for this changeset; ran all 17 under scripts/pm/os-verify-lock.sh (SLOT issue14695-gates, queued behind two other agents' heavy runs, polled in-turn with bounded sleep loops -- never parked on a Monitor) at head a36af678c4 (after merging origin/main, 5 commits, zero overlap with the two touched files, confirmed via git diff --stat): 16 of 17 green, the one exception node scripts/check-test-completeness.mjs exiting 3 (NOT MEASURED by its own design -- needs a CI test log -- not a red). pnpm check:pm-dispatch-gates (self-test respawns its own CLI many times, ~7 min under a shared-box lock) passed at 1240 cases. dispatch-gates --ran reconciliation: 17 derived, 17 run, 0 UNRUN. Re-ran check-ci-filter-parity.mjs, check-cross-package-test-inputs.mjs, check-whole-set-label-write.mjs and bare-root-worklist.mjs --self-test again after the merge since their inputs (scripts/check-ci-filter-parity.mjs, scripts/cross-package-test-inputs.mjs, .github/workflows/ci.yml) moved under it -- all green on the merged tree.",
  "mcp_calls": 2,
  "open_questions": [
    {
      "question": "Two of the fourteen REFUSE-WIDE members ('the two packages/spec-filtered ones') are still unrecorded. The issue text says they are 'named in PR #14692's body with their measured shares', but the PR body -- read in full, twice, and diffed byte-for-byte against the saved copy -- names neither family nor share; its bucket table states only the aggregate '14 members, 32-52%'. A scratchpad search for the filing session's own working notes (grep across pr14692-body.md, seat6023-w*.md, and a broad spec-filtered/percentage search) found no per-member breakdown either.",
      "options": [
        "A. The filing session (session_01WLJQhde67SeTccsmnBVarV, the same PM session that dispatched this card) supplies the two family names and shares from its own working notes/history, and a small follow-up PR adds those two rows to CENSUS_REFUSE_WIDE.",
        "B. Authorize this dev session (or a follow-up) to re-run the card's own residue + fs-trace census over the full ~200-family roster to find the two packages/spec-scoped Silent, \u22655-file, root-wide members itself -- a materially heavier task than measuring 12 already-named gates, likely warranting its own card.",
        "C. Accept 12 of 14 as the closed scope for this card and leave the remaining two as a standing, explicitly-noted gap (no further card) since the cost of the gap is search misdirection on a p3 bookkeeping matter, not a live defect."
      ],
      "recommendation": "A, because the filing session's own triage comment states 'Yours are at 2aa8456cf by the card's own method' -- the measurement demonstrably exists, just not in a channel this dev session could read, and recovering two numbers from an already-completed measurement is far cheaper and more precise than B's re-derivation or C's leaving a known, nameable gap open."
    }
  ],
  "out_of_scope_findings": []
}

Generated by Claude Code

tooling(pm): record the last 2 of 14 REFUSE-WIDE gate families (14 of…

os-dev-report

{
  "issue": 14695,
  "status": "done",
  "branch": "claude/issue-14695-refuse-wide-ledger-rows",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/14800",
  "premise_still_valid": true,
  "summary": "Follow-up to the first sitting (12 of 14): recorded the final 2 REFUSE-WIDE members in CENSUS_REFUSE_WIDE, closing the open question. The coordinator supplied the two names and cross-check counts from the filing session's own census artifacts (scratchpad issue-14325/census.json + categories.json, base 2aa8456cf) -- verified directly by reading both files myself before acting: check:error-code-provenance and check:objectui-pin-citations, both packages/spec-scoped gates run via `pnpm --filter @objectstack/spec run check:X`, sitting in that census's 'root-wide, same class, no ledger row' bucket as members 13 and 14. Re-measured both by this table's own method (strace openat against a dedicated 2aa8456cf worktree) rather than copying the filing session's numbers, to keep the table one-method/one-base. Both land within 1 file of the filing session's own counts: 2070 vs 2069 for error-code-provenance, and an exact 1192 vs 1192 for objectui-pin-citations -- corroborating the method a second time, now on a tsx-run gate rather than a plain node one. objectui-pin-citations (20.4%) measures below the issue's summarised 32-52% band; recorded as measured, not forced into the band, matching how the first sitting handled the four members that measured above it. All 14 of 14 are now recorded; the table's docblock and PR body were both updated to reflect the closed set.",
  "tests": "node scripts/pm/bare-root-worklist.mjs --self-test: green, 57 live TRIAGE rows unchanged plus 14 CENSUS_REFUSE_WIDE rows well-formed/disjoint/hint-free/one-base (up from 12). Ablation: mutated the new objectui-pin-citations row's reads to -5 -- self-test went RED on exactly the shape assertion (direct exit-code capture: EXIT=1), restored from a saved pre-mutation copy and reverified green (EXIT=0). Re-ran the full 17-command dispatch-gates derivation (unchanged from the first sitting) under scripts/pm/os-verify-lock.sh at head a78607fe0c: 16 of 17 green, the one exception node scripts/check-test-completeness.mjs exiting 3 (NOT MEASURED by design, not a red). dispatch-gates --ran reconciliation: 17 derived, 17 run, 0 UNRUN. origin/main fetched before both the commit and the push -- 0 commits behind at push time, no merge needed this sitting. PR body updated (14-of-14 table, new Follow-up section) and read back: draft still true, labels still [size/m, skip-changeset], both new family names present in the live body.",
  "mcp_calls": 1,
  "open_questions": [],
  "out_of_scope_findings": []
}

Generated by Claude Code

ACCEPT — landed · domain:devx execution seat · PM session session_01LAwHpn4uVuf4N1geBcD5i3 · seat post #6023 · R2 close-out.

PR #14800 MERGED 08:17:55Z as 5a22459e0 (queue). Probe on origin/main 431979e67: CENSUS_REFUSE_WIDE table present in scripts/pm/bare-root-worklist.mjs with the check:error-code-provenance row (control: TRIAGE in the same file). PM review PASS was recorded on the seat board (R1 wave 10: 14 rows re-measured at 2aa8456cf, self-test pins the 14, check-whole-set-label-write.mjs fixture keys real). Zero rework.

Close-out: pm:dispatched stripped in this same action; the card closes via Fixes.

Generated by Claude Code

Navigation Menu

Repository navigation

[finding] 14 root-wide gate families measure into bare-root-worklist's REFUSE-WIDE class (32–52 % of `packages/`) but carry no ledger row — so the class reads as undecided debt when it has been ruled #14695

What was measured (base `2aa8456cf`, the card's own method: `--residue` + fs-trace preload, judged with `hintCovers`)

Possible disposition (for triage)

Adjacent, same report (folded here rather than filed twice)

Why this is worth recording rather than shrugging

Why `p3`

Scope

On the folded second half

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[finding] 14 root-wide gate families measure into bare-root-worklist's REFUSE-WIDE class (32–52 % of packages/) but carry no ledger row — so the class reads as undecided debt when it has been ruled #14695

Description

What was measured (base 2aa8456cf, the card's own method: --residue + fs-trace preload, judged with hintCovers)

Possible disposition (for triage)

Adjacent, same report (folded here rather than filed twice)

Activity

huangyiirene commented on Sep 2, 2026

Why this is worth recording rather than shrugging

Why p3

Scope

On the folded second half

claude commented on Sep 3, 2026

claude commented on Sep 3, 2026

claude commented on Sep 3, 2026

claude commented on Sep 3, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

[finding] 14 root-wide gate families measure into bare-root-worklist's REFUSE-WIDE class (32–52 % of `packages/`) but carry no ledger row — so the class reads as undecided debt when it has been ruled #14695

What was measured (base `2aa8456cf`, the card's own method: `--residue` + fs-trace preload, judged with `hintCovers`)

Why `p3`