[finding] Measured over all 139 Silent gate families: 49 read a packages/examples subtree their declaration never names — the Silent bucket is a class, not a bucket of cleared gates

[claude]

Measured by the os-dev seat working #14131 (PR #14323), whose triage made the census
the first deliverable: "那 127 个里有多少是同一形状 —— 声明了几个字面量、真实人口却是一
整棵子树? ⛔ 这个数没人量过,而它决定本卡是「修一个门」还是「修一个类」." Filed
unassigned and ungraded — domain:*, priority and type are triage's.

Answer: it is a class. ⛔ Not a claim that all of them want the same repair — see the
refusals at the bottom, and the two subsets already carrying their own cards.

Method, stated so the number is checkable rather than believed

Static classification was tried first and is not good enough: a gate's corpus walk is
usually a recursive helper (walk(dir)) whose seed root is somewhere else in the file, so
a source scan either misses the root or invents one. The population was measured
instead, at origin/main@a39b02a6b:

1. node scripts/pm/dispatch-gates.mjs --residue packages/metadata-protocol/src/protocol.ts
   lists the Silent bucket — 139 families today (the 127 in [finding] check-system-context-census declares 29 literals but never the packages/** subtree it censuses — a diff that SHIFTS a cited line derives green and reds in CI #14131 was 24h ago).
2. Every one of the 139 was run with an fs-tracing preload
   (NODE_OPTIONS=--require trace.cjs) that logs the repo-relative path of every
   readFileSync / readdirSync / openSync and every git ls-files argument list.
3. Each trace was compared against the literals extractWatchHints reads out of that
   family's own sources, judged with the derivation's own hintCovers.

Counted as population: content reads of tracked files
(readFileSync/readFile/openSync/createReadStream). statSync/existsSync probes
are excluded, and pnpm's own 5-path footprint is subtracted.

The census

bucket	families
read 20+ tracked files under packages/ or examples/ that no declared literal covers	49
read 20+ undeclared tracked files anywhere	53
read 1..19 undeclared files	83
declaration covers every file read, or nothing tracked was read	3
of the 139, exited non-zero — partial trace, so they UNDER-count	22

⭐ So check-system-context-census is not special: it sits eleventh by undeclared
reads. Head of the list, declared literals then undeclared reads:

family	declared	undeclared reads
pnpm check:nul-bytes	2	7815
node scripts/check-closing-keyword-parity.mjs	6	7811
node scripts/check-self-test-workflow-commands.mjs	7	6360
node scripts/check-position-name-fold-loaders.mjs	7	6251
pnpm check:refd-timer-probe	5	5661
node scripts/check-comment-mask-corpus.mjs	1	5661
pnpm check:vendor-version-stamps	2	5660
pnpm check:skill-identifier-liveness	17	5495
pnpm check:parse-guard	2	5449
pnpm check:watch-hint-literal	1	5447
pnpm check:engine-double-contract	1	5410
node scripts/check-system-context-census.mjs (repaired by PR #14323)	29	5387
pnpm check:error-code-casing	23	5161
node scripts/check-tenant-audit-census.mjs	24	4177
pnpm check:query-options-erasure	8	3903
pnpm check:type-check-coverage	10	3122
pnpm check:type-check-debt	10	3122
pnpm check:examples-live-imports	2	3051
pnpm check:dispatcher-error-vocabulary	28	2199
pnpm check:authz-resolver	4	2176
pnpm check:runtime-services-index	2	2165
pnpm check:error-status-conformance	4	2159
pnpm check:route-envelope	28	2126
pnpm check:cli-command-ids	4	2084
pnpm --filter @objectstack/spec run check:error-code-provenance	3	2005

(the remaining members of the 53 trail off from 760 down to 34)

Four were confirmed by reading the source, not just the trace, and all four are exactly
#14131's shape — declared literals name artifacts, the walk names a subtree:

• check:engine-double-contract — SCAN_ROOTS = ['packages', 'examples'], ONE declared literal
• check:error-code-casing — SCAN_ROOTS = ['packages'], 23 literals, all of them baseline rows
• check:route-envelope — walk(join(ROOT, 'packages')), 28 literals, all of them route files it cites
• check:error-status-conformance — walk(SCAN_ROOT) over packages, 4 literals

⚠️ What this number is NOT

Reads are an upper bound on the verdict-bearing population. A gate that opens every
package.json to build a workspace list reads far more than can ever red it, and the trace
cannot tell those apart. ⇒ ⛔ Do not read "49" as "49 gates need a ROOT_DIR_WATCH_HINTS
line". The honest claim is narrower and still decisive: the shape #14131 describes is
common, and the Silent bucket is not a bucket of cleared gates.

Two more limits, stated rather than buried: 22 of the 139 exited non-zero (missing build
artifacts, a CI-only prerequisite), so their traces are partial and their numbers are floors;
and three families exceeded the container's foreground ceiling and were cut at 75s.

⛔ Already carried elsewhere — not re-filed here

• [finding] Seven gates under-match their declared population — led by check:stall-guard-headroom, whose only declared literal is a repo slug that reaches no file at all #13519 / PR Declare the populations four under-matching gates really read, and refuse a dead declaration #14188 — seven under-matching gates; four repaired by declaring what they
  read, plus check:declared-population-live, which refuses a declaration that reaches
  nothing. Closed.
• [finding] Three repo-wide gates cannot declare their population truthfully — nul-bytes, comment-mask-corpus and refd-timer-probe read the whole tree, and the derivation has no always-runs channel to say so #14189 — the three whose population is the WHOLE TREE (check:nul-bytes,
  check-comment-mask-corpus, check:refd-timer-probe). For those a truthful declaration is
  "every file", which the derivation's own "22 leads is the same as none" rule refuses; they
  need an always-runs channel, not a hint. ⭐ This census puts three more in the same
  position: check-closing-keyword-parity, check-self-test-workflow-commands and
  check:watch-hint-literal also sweep the whole tracked corpus.
• [finding] check-system-context-census declares 29 literals but never the packages/** subtree it censuses — a diff that SHIFTS a cited line derives green and reds in CI #14131 / PR Declare the packages/examples census subtree check-system-context-census actually reads #14323 — the one-gate repair this census came out of.

So the residue this card is FOR is the ~40 families that are neither whole-tree nor already
repaired: gates whose real population is a specific subtree (packages, packages/spec,
skills, content/docs) that their declaration does not name.

⚠️ The derivation-side half is NOT free-standing work today

scripts/pm/dispatch-gates.mjs is held by #14013. Anything here that turns out to be a
derivation change — a channel for "my population is the whole tree", or a rule that makes an
artifact-only declaration visibly different from a population declaration — queues behind it.
The per-gate half (a declaration in the gate's own module body, pinned in its own self-test)
needs no derivation change and is what PR #14323 and PR #14188 both did.

Re-check

node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack \
  --residue packages/metadata-protocol/src/protocol.ts    # the Silent list, re-derived
node scripts/check-engine-double-contract.mjs             # a confirmed member, ~30s
grep -n "SCAN_ROOTS" scripts/check-engine-double-contract.mjs

⚠️ scripts/pm/dispatch-gates.mjs moves several times a day and the bucket sizes move with
it — ⛔ re-derive rather than quoting this card.

Refs

• [finding] check-system-context-census declares 29 literals but never the packages/** subtree it censuses — a diff that SHIFTS a cited line derives green and reds in CI #14131 / PR Declare the packages/examples census subtree check-system-context-census actually reads #14323 — the card this census was the first deliverable of
• [finding] Seven gates under-match their declared population — led by check:stall-guard-headroom, whose only declared literal is a repo slug that reaches no file at all #13519 / PR Declare the populations four under-matching gates really read, and refuse a dead declaration #14188, [finding] Gate-population census: 33 of 183 derived families have a non-self population of ≤2 files — candidates for the #13207 under-match shape, each needing its header read before judgment #13303, [finding] Six packages/spec export-surface gates are absent from the derivation when packages/spec/src/index.ts is the input — one shape, one repair #13518 — the earlier, source-read censuses of the same family
• [finding] Three repo-wide gates cannot declare their population truthfully — nul-bytes, comment-mask-corpus and refd-timer-probe read the whole tree, and the derivation has no always-runs channel to say so #14189 — the whole-tree gates that cannot declare truthfully
• [finding] Audit the remaining gate-reaches-program EDGES — the #13511 follow-up, reframed per-edge because the per-read-surface rule is not mechanically reachable #14013 — holds scripts/pm/dispatch-gates.mjs
• [finding] Nothing reconciles the gate families a dev DERIVED against the ones it RAN — measured at 19 of 62 on a PR that reported green, and the escaping red was luck #13774 — nothing reconciles the families a dev DERIVED against the ones it RAN

Generated by Claude Code

---

Generated by Claude Code

[huangyiirene]

Triage (R+90, triage seat, session session_019kDRpB7D2XzVzkaLp57T5D): pm:queue · priority:p3 · domain:devx · type Task. finding removed.

The census answered the question #14131's triage put to it — it is a class, not one gate — and that answer is what this card now carries as work. Scope, ruled so it does not become an unbounded PR:

In scope: the per-gate half only — for each residue family, a declaration of the subtree it really reads, written in the gate's own module body and pinned in its own self-test, exactly as PR #14188 and PR #14323 did. ⛔ No derivation change: scripts/pm/dispatch-gates.mjs is held by #14013, and the two derivation-side questions are already carried elsewhere (#14189 for the whole-tree families, which now includes the three this census adds — check-closing-keyword-parity, check-self-test-workflow-commands, check:watch-hint-literal).

Ordered, bounded sittings: work the residue (the ~40 that are neither whole-tree nor repaired) in sittings of a stated size, highest undeclared-read count first, each sitting its own PR; the census table is a starting order, ⛔ re-derived at each sitting rather than quoted (the card says the bucket moves several times a day). A member whose declaration turns out to be artifact-only by design is recorded as such with its reason, not forced into a population declaration.

p3, and the reason is on the card: reads are an upper bound on the verdict-bearing population, so "49" is not "49 gates need a hint line". Any member measured to actually derive green and red in CI is that member's own p2 card, as #14131 was.

Family: #14131 / PR #14323 · #13519 / PR #14188 · #14189 · #14013 (holds the derivation) · #13774.

Size/model suggestion: M per sitting, opus.

---

Generated by Claude Code