[finding] `check-doc-route-spelling.mjs` declares `skills` as a scan root in a shape BOTH the deriver and the tool that audits for this exact gap are blind to — a lowercase object property

[os-trump]

Filed by the domain:services PM seat (#6021) from #17050's triage. Blocked-by: #17048 — see the disposition at the end; the fact below is measured and final, only the fix design waits.

The measurement, verified by this seat against origin/main

scripts/check-doc-route-spelling.mjs declares its scan population as a lowercase object property:

const REAL_CONFIG = {
  roots: ['content/docs', 'skills'],
  …

scripts/pm/bare-root-worklist.mjs — the tool whose entire job is to find undeclared bare-root populations — recognises constants by name, uppercase only:

const POPULATION_CONSTANT = /^(?:[A-Z0-9_]*_ROOTS?|[A-Z0-9_]*_DIRS?|ROOTS|DIRS|POPULATION|[A-Z0-9_]*_SCOPE)$/;

⇒ A lowercase roots: property cannot match, so the gate's skills root is invisible to it. And dispatch-gates derives no hint from that shape either, so the family sits Silent for a change under skills/** — even though the gate's own header declares skills/** in scope.

⭐ The sharp part: this is the same bare-root species the tool already tracks two instances of (check-position-name-fold-loaders, check-whole-set-label-write). It is missed not because it is a different defect but because it is spelled differently — and the thing doing the missing is the auditor built for precisely this.

Why this is not #16828's class, stated so it is not re-litigated

#17050's triage placed this row firmly in (b) — an undeclared wide/subtree population — not (a), #16828's self-concealing-registry shape. It has no table of already-declared conformant members; every file it walks is judged the same way, old or new. So this is not a third instance of #16828; it is a first instance of a blind spot in the auditor, which is a different and arguably cheaper thing to fix.

Two candidate fixes — ⛔ not a ruling, and both want the post-#17048 tree

1. Widen the recognition. Teach POPULATION_CONSTANT (or the surrounding extraction) the object-property shape, so roots: [...] inside a config literal is seen. Fixes the class, not the instance; ⚠️ it will also surface however many other gates spell their population that way, which is a population nobody has counted — see below.
2. Rename the declaration in check-doc-route-spelling.mjs to a recognised shape (const SCAN_ROOTS = […]). Fixes the one instance, cheap, and leaves the auditor as blind as it was.

⭐ (1) is the one that removes the blind spot; (2) is the one that makes today green. ⚠️ Whoever takes it should say which problem they are solving before choosing.

⚠️ The unmeasured half, which decides how big (1) is

Nobody has counted how many gates declare a population as a lowercase property rather than an uppercase const. #17050's sweep could not have told us — it keyed on tree-walk signatures, not declaration shapes. ⛔ A zero from any sweep here needs a positive control that re-finds check-doc-route-spelling.mjs's own roots:, or it is NOT MEASURED, not a clean tree.

Disposition — filed now, against the delivering seat's recommendation, and why

The delivering seat recommended leaving this for whoever next touches bare-root-worklist.mjs after #17048 lands, reasoning that filing now risks a card whose fix design goes stale.

⛔ Overridden, and the reason is a standing correction in this lane's own ledger: a "waiting for someone else" list rots silently. The seat's staleness concern is real but applies to the fix design, not to the fact — and the fact is measured, verified twice, and does not move when #17048 lands. So the fact is recorded here now; the design is explicitly deferred.

Unblock condition, written to be executed rather than interpreted: #17048 merges or closes. It changes the exact predicates bare-root-worklist.mjs imports from dispatch-gates.mjs, so any fix designed before then is designed against a tree about to move. At that moment this card returns to pm:queue and both candidate fixes above are re-derivable against the merged tree.

Provenance

#17050's per-script triage (report issuecomment-5596...), row 8 of 12. That triage's headline result was 0 of 12 rows matching #16828's class — this row is the only genuinely new finding in the twelve, and it is a different species from the one being hunted.

Related: #16828 (the class the sweep was hunting) · #17050 (the sweep) · #17032 · #16770 — all about the distance between what a check is named or shaped like and what it actually reads.

[os-trump]

Claim: session session_012zTkyNHJ7TkuN2oXtP5x37 · branch claude/issue-17057-doc-route-spelling-population-shape · seat domain:services PM (#6021)

Clause-②: no — the whole diff lives in repo-local scripts/**, which no package publishes (root package.json is private: true; no package's files array names scripts/, verified on the tree). ⚠️ STOP-and-report if the delivery finds itself adding an exported symbol to a published package, a new key on a published payload, or any packages/spec/src/** path — that flips Clause-② and this card is not carrying the review tier.

Unblock condition met, verified on the tree — not from a report

#17048 merged as a886fe2d9 (git log origin/main --grep='(#17048)'). pm:blocked → pm:dispatched. The fix design may now be derived against the merged tree, as the card's own unblock sentence specifies.

Premise re-verified by this seat against today's origin/main (91f65c4ea), post-#17048

Both halves of the card still read exactly as filed:

• scripts/check-doc-route-spelling.mjs:131 → roots: ['content/docs', 'skills'], inside const REAL_CONFIG = {
• scripts/pm/bare-root-worklist.mjs:119 → const POPULATION_CONSTANT = /^(?:[A-Z0-9_]*_ROOTS?|[A-Z0-9_]*_DIRS?|ROOTS|DIRS|POPULATION|[A-Z0-9_]*_SCOPE)$/;

⇒ the lowercase property cannot match the uppercase-only name regex. The card's fact did not move when #17048 landed, as it predicted.

The ruling the card deferred: take fix (1), widen the recognition

The card refused to rule and asked whoever takes it to say which problem they are solving first. Ruling now, so it is not re-litigated:

⭐ Fix (2) — renaming roots: to const SCAN_ROOTS — is refused as the whole answer. It makes today green while leaving the auditor's output silently under-counted by an unknown amount. That is precisely "a zero from a probe that cannot answer yes is NOT MEASURED", one level up: bare-root-worklist.mjs would keep reporting a number that is wrong in a direction nobody can see. This lane spent today deleting exactly that shape (#16919, #17039 → #17055, #16828 → #17048); ⛔ do not re-introduce it here.

Fix (1) is the deliverable. Fix (2) is acceptable in addition, if renaming also makes the instance clearer — but never instead.

⚠️ Measure the tail BEFORE committing to scope — this is the part that can go wrong

Widening recognition will surface every other gate that spells its population as a lowercase property, and each surfaced row then needs a disposition (declared / REFUSE-WIDE / fixed) or the gate goes red. Nobody has counted that population. So:

1. Measure first. Widen the recognition, then run the auditor and report the count of newly-surfaced rows before deciding what the PR carries.
2. ⛔ A zero, or a small number, is not a result without a positive control. The control is fixed and named by the card: the sweep must re-find check-doc-route-spelling.mjs's own roots: under the widened recognition. If it does not, the instrument is broken and every number it produced is NOT MEASURED — report that, do not proceed.
3. A crude first cut from this seat, offered as a lead and ⛔ explicitly NOT a census: a line-anchored grep for ^\s*(roots|dirs|scanRoots|searchRoots|population|scope)\s*:\s*\[ over scripts/** returns 22 hits, but nearly all are inside self-test fixture literals (check-type-check-coverage.mjs, check-self-test-wired.mjs, check-self-test-workflow-commands.mjs); the only production-config instance it finds is check-doc-route-spelling.mjs itself (lines 131 and 787). ⚠️ That grep is anchored and name-listed, so it under-counts by construction — treat it as a hint that the tail may be small, never as the measurement. Your instrument must be the widened auditor, not my grep.

Scope gate, written to be executed: if the honest count of newly-surfaced production rows is ≤ 3, carry the widening and their dispositions in this one PR. If it is more, ⛔ stop and report the count with the row list — I will re-scope rather than have you disposition a population you had to guess at.

Fences

• ⛔ Governed surface: skills/** is governed. This gate scans skills/**; the fix must not edit anything under it. If a surfaced row's only fix is inside skills/**, docs/adr/**, .claude/**, AGENTS.md or CLAUDE.md, that is the maintainer's floor — report it, do not write it.
• ⛔ Never skip, disable or quarantine a test to get green.
• ⛔ Never git stash — the stash stack is shared across all worktrees. Use a patch or a wip commit.
• Worktree-first: objectstack-issue-17057, dedicated. ⛔ Never edit the shared primary checkout.
• Single-writer: only .objectui-sha is gate-protected (verified on the tree — scripts/check-single-claim-paths.mjs's SINGLE_CLAIM_PATHS has exactly one entry). ⚠️ scripts/pm/bare-root-worklist.mjs is not a claimed path and no fence applies to it; check for an in-flight PR touching it as ordinary courtesy, not as a gate.
• Commit, push and open the draft PR as soon as there is anything worth showing — before verification finishes. Verification results get appended to the body as they arrive. ⛔ Do not end a turn waiting on a background job or a watcher; nothing wakes a seat that is waiting.

---

Generated by Claude Code

[os-trump]

os-dev-report

{
  "issue": 17057,
  "status": "done",
  "branch": "claude/issue-17057-doc-route-spelling-population-shape",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/17113",
  "premise_still_valid": true,
  "summary": "Both premises re-verified on origin/main 91f65c4ea before any edit: bare-root-worklist.mjs:119 POPULATION_CONSTANT is uppercase-only, and check-doc-route-spelling.mjs:131 declares roots: ['content/docs', 'skills'] as a lowercase object property inside REAL_CONFIG. Took fix (1) as ruled: a new POPULATION_PROPERTY mirrors POPULATION_CONSTANT term for term (camelCase hump where the underscore is; bare singular root/dir refused as ROOT/DIR already are), and populationSpans now returns const-shaped and property-shaped spans as one list so no caller keeps a roster. The const scanner is byte-for-byte unchanged and the property value gets its own depth-tracking delimiter, so the before/after has one variable. Fix (2) was NOT taken. Dispositions for the three surfaced rows: check-doc-route-spelling.mjs gains ROOT_DIR_WATCH_HINTS = ['skills/**'] (the idiom two sibling doc gates already carry at this root, 46 of 47 tracked files, 97.9% precise) so its row is now REACHABLE and the family stops sitting Silent for skills-only cards; the two check-console-injection.mjs rows are join() path components in a gate that never walks the root and are recorded in UNJUDGED with measured notes -- UNJUDGED's docblock binds a recogniser-widening pass to measure and not to judge, which is what #15468's widening did before it. Appending forced the bucket's shared base from 66e68adc6 to 91f65c4ea: 7 of the 11 inherited entries name byte-identical gate sources (blob id), one cited line number was corrected (:436 to :446), no population boolean or recorded reason was otherwise rewritten.",
  "tests": "MEASUREMENT (the number the card asked for first): origin/main 91f65c4ea unmutated = 78 triples across 44 gate source files; after the widening = 81 across 46. Exactly 3 newly-surfaced rows, 0 gone, full row-set diff taken: check-console-injection.mjs distDir packages, check-console-injection.mjs specDir packages, check-doc-route-spelling.mjs roots skills. All three are production config -- maskSelfTests blanks the fixture copy of the same literal at :787+. 3 <= 3 so this PR carries the dispositions. POSITIVE CONTROL HOLDS: the widened sweep re-finds check-doc-route-spelling.mjs's own roots:, so the instrument is not answering zero because it cannot answer yes. ABLATION on the exact instrument, predicted before running, a shell trap on EXIT INT TERM with absolute paths: neuter POPULATION_PROPERTY alone -> 78 triples across 44 files, i.e. the pristine origin/main baseline to the row, with the doc-route-spelling row and both console-injection rows absent; and it is measured with the new skills/** declaration still on disk, so the declaration alone moves nothing and the recogniser is what surfaces the rows. On-disk landing proven both ways before reading (original alternation gone = 0, injected marker present = 1). Restoration proven by empty git diff HEAD AND blob equality (86fb61af2f829cb9fc088ec721c8c84b44089ece == HEAD blob), never by an exit code. HONEST NOTE: the FIRST ablation attempt was a silent no-op (a perl -0pi substitution matching nothing, exit 0); the script's own landing check caught it and no number from that attempt was used -- the run above is the re-anchored one. GATES: node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack derived 33 families on a tree merged up to origin/main f87fdf359 (no STALE TREE caveat; change set is exactly the 2 edited files). All 33 run in the foreground, every exit code captured by redirect-then-capture, 0 non-zero. Reconciled by the tool: 'dispatch-gates --ran: 33 derived famil(ies) accounted for -- 33 run, 0 NOT-MEASURED.' => 33 derived / 33 run / 0 NOT-MEASURED / 0 UNRUN. Also ran the 4 artifact-roster gates the derivation flags as evidence in NEITHER direction because their roster sits under a directory one of my paths is in (check-published-list-mirrors.mjs and its --self-test, check:pm-label-desc-cap, check:console-injection) -- all exit 0. LINT: no narrowing claimed -- ran the whole farm. pnpm lint (eslint . --no-inline-config) exit 0 in 81s at 5e5df1e13; scoped --format json over the 2 changed files reports 2 files, 0 errors, 0 warnings. BYTES: check:nul-bytes green plus a direct control-character grep over both changed files, no match. NOT OWED, stated: no test file in the tree names either edited script (git grep over *.test.ts / *.spec.ts is empty) -- their test suites ARE their --self-test entry points, both inside the 33 and both green. The derivation's 48 artifact-roster, 11 declared-wide and always-runs families sit outside the derived total by construction; CI runs them. The 13 pending-changeset families do not apply (skip-changeset).",
  "mcp_calls": "0 -- every GitHub read and write went through the container's REST channel (repo-scoped probe green) or plain git; the card and its Claim comment were read from the public-repo zero-quota page payload, edges 11 == totalCount 11, hasNextPage false",
  "open_questions": [],
  "out_of_scope_findings": [
    "noted, not filed: both of check-console-injection.mjs's REAL populations (packages/console/dist and packages/spec) are themselves invisible to dispatch-gates, for the different reason that they are assembled from bare join() components rather than spelled -- recorded in the UNJUDGED note in-tree where the maintainer judges the rows, not filed as a card",
    "noted, not filed: UNJUDGED's one-shared-base pin means every future recogniser widening must re-read all inherited notes before it can append; that cost is the pin's design and is now documented in its docblock, but it will grow with the bucket"
  ]
}

---

Generated by Claude Code

[os-trump]

os-dev-report

{
  "issue": 17057,
  "status": "done",
  "branch": "claude/issue-17057-doc-route-spelling-population-shape",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/17113",
  "premise_still_valid": true,
  "summary": "Both premises re-verified on origin/main 91f65c4ea before any edit: bare-root-worklist.mjs:119 POPULATION_CONSTANT is uppercase-only, and check-doc-route-spelling.mjs:131 declares roots: ['content/docs', 'skills'] as a lowercase object property inside REAL_CONFIG. Took fix (1) as ruled: a new POPULATION_PROPERTY mirrors POPULATION_CONSTANT term for term (camelCase hump where the underscore is; bare singular root/dir refused as ROOT/DIR already are), and populationSpans now returns const-shaped and property-shaped spans as one list so no caller keeps a roster. The const scanner is byte-for-byte unchanged and the property value gets its own depth-tracking delimiter, so the before/after has one variable. Fix (2) was NOT taken. Dispositions for the three surfaced rows: check-doc-route-spelling.mjs gains ROOT_DIR_WATCH_HINTS = ['skills/**'] (the idiom two sibling doc gates already carry at this root, 46 of 47 tracked files, 97.9% precise) so its row is now REACHABLE and the family stops sitting Silent for skills-only cards; the two check-console-injection.mjs rows are join() path components in a gate that never walks the root and are recorded in UNJUDGED with measured notes -- UNJUDGED's docblock binds a recogniser-widening pass to measure and not to judge, which is what #15468's widening did before it. Appending forced the bucket's shared base from 66e68adc6 to 91f65c4ea: 7 of the 11 inherited entries name byte-identical gate sources (blob id), one cited line number was corrected (:436 to :446), no population boolean or recorded reason was otherwise rewritten. REVIEW ROUND 2: the coordinator found that the comment beside the property scan claimed the leading character class excluded BOTH a ternary's `a ? roots : dirs` and a `foo.roots:` member expression. Reproduced independently: only the member-expression half holds, because the `?` is followed by a space and \\s is in the class -- the TIGHT `a?roots:dirs` is the spelling that gets excluded, and a `roots:` label statement matches too. Took option 1 (correct the comment) plus the pins; the regex is NOT tightened, because excluding the ternary needs a look-behind and excluding the label form needs real parsing, so a tightening would be a second change owing its own full row-set diff and would still leave the label form matching. One correction to the review's severity premise, measured not argued: sweep admits a literal ONLY when a span contains it (if (!span) continue), so a span is an INCLUSION filter and a spurious one can only ADD a row, never hide one.",
  "tests": "MEASUREMENT (the number the card asked for first): origin/main 91f65c4ea unmutated = 78 triples across 44 gate source files; after the widening = 81 across 46. Exactly 3 newly-surfaced rows, 0 gone, full row-set diff taken: check-console-injection.mjs distDir packages, check-console-injection.mjs specDir packages, check-doc-route-spelling.mjs roots skills. All three are production config -- maskSelfTests blanks the fixture copy of the same literal at :787+. 3 <= 3 so this PR carries the dispositions. POSITIVE CONTROL HOLDS: the widened sweep re-finds check-doc-route-spelling.mjs's own roots:, so the instrument is not answering zero because it cannot answer yes. ABLATION on the exact instrument, predicted before running, a shell trap on EXIT INT TERM with absolute paths: neuter POPULATION_PROPERTY alone -> 78 triples across 44 files, i.e. the pristine origin/main baseline to the row, with the doc-route-spelling row and both console-injection rows absent; and it is measured with the new skills/** declaration still on disk, so the declaration alone moves nothing and the recogniser is what surfaces the rows. On-disk landing proven both ways before reading (original alternation gone = 0, injected marker present = 1). Restoration proven by empty git diff HEAD AND blob equality (86fb61af2f829cb9fc088ec721c8c84b44089ece == HEAD blob), never by an exit code. HONEST NOTE: the FIRST ablation attempt was a silent no-op (a perl -0pi substitution matching nothing, exit 0); the script's own landing check caught it and no number from that attempt was used -- the run above is the re-anchored one. GATES: node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack derived 33 families on a tree merged up to origin/main f87fdf359 (no STALE TREE caveat; change set is exactly the 2 edited files). All 33 run in the foreground, every exit code captured by redirect-then-capture, 0 non-zero. Reconciled by the tool: 'dispatch-gates --ran: 33 derived famil(ies) accounted for -- 33 run, 0 NOT-MEASURED.' => 33 derived / 33 run / 0 NOT-MEASURED / 0 UNRUN. Also ran the 4 artifact-roster gates the derivation flags as evidence in NEITHER direction because their roster sits under a directory one of my paths is in (check-published-list-mirrors.mjs and its --self-test, check:pm-label-desc-cap, check:console-injection) -- all exit 0. LINT: no narrowing claimed -- ran the whole farm. pnpm lint (eslint . --no-inline-config) exit 0 in 81s at 5e5df1e13; scoped --format json over the 2 changed files reports 2 files, 0 errors, 0 warnings. BYTES: check:nul-bytes green plus a direct control-character grep over both changed files, no match. NOT OWED, stated: no test file in the tree names either edited script (git grep over *.test.ts / *.spec.ts is empty) -- their test suites ARE their --self-test entry points, both inside the 33 and both green. The derivation's 48 artifact-roster, 11 declared-wide and always-runs families sit outside the derived total by construction; CI runs them. The 13 pending-changeset families do not apply (skip-changeset). ROUND 2 EVIDENCE: reproduced the review's table exactly by calling populationSpans directly (member expression 0 spans; ternary 1; tight ternary 0; label 1; positive control 1). Direction measured by splicing synthetic families over throwaway fixtures under a trap (removed; git status clean after): bare root inside a spurious ternary span = 1 row ['roots apps']; the same root in NO span = 0 rows; a root inside BOTH a real const span and a spurious ternary span = 1 row attributed to ['SCAN_ROOTS apps'], so a spurious span cannot even steal a row's key. Six new self-test cases pin all of it. Row-set diff re-measured on the new head: still 3 surfaced / 0 gone. Re-derived: same 33 families, same 2-path change set. ⚠️ HONEST NOTE: the first re-run of the gate batch returned six exit-3 PREREQUISITE NOT MET results because the worktree had been torn down and recreated without pnpm install -- those are NOT MEASURED, never passes; pnpm install then a full re-run of all 33 is where the round-2 numbers come from. Round 2 totals: 33 derived / 33 run / 0 NOT-MEASURED / 0 UNRUN, scoped eslint 2 files 0 errors 0 warnings, control-character grep clean, --self-test green, report exit 0.",
  "mcp_calls": "0 -- every GitHub read and write went through the container's REST channel (repo-scoped probe green) or plain git; the card and its Claim comment were read from the public-repo zero-quota page payload, edges 11 == totalCount 11, hasNextPage false",
  "open_questions": [],
  "out_of_scope_findings": [
    "noted, not filed: both of check-console-injection.mjs's REAL populations (packages/console/dist and packages/spec) are themselves invisible to dispatch-gates, for the different reason that they are assembled from bare join() components rather than spelled -- recorded in the UNJUDGED note in-tree where the maintainer judges the rows, not filed as a card",
    "noted, not filed: UNJUDGED's one-shared-base pin means every future recogniser widening must re-read all inherited notes before it can append; that cost is the pin's design and is now documented in its docblock, but it will grow with the bucket",
    "noted, not filed: the review's stated mechanism for spurious spans (that they SWALLOW rows) is the opposite of what sweep does; the inclusion-filter direction is now measured, pinned in the self-test and written into the comment, so the next reader inherits the measurement rather than either claim"
  ]
}

---

Generated by Claude Code