Ratchet the declared registry log level across engine-booting suites

[claude]

Fixes #15425

Nothing in this repo ratcheted a registry log level. PR #13985 and PR #14016 declared OS_REGISTRY_LOG in four vitest harnesses and removed ~47,900 console-carried lines between them, and deleting any one of those env lines restored five figures of output while every suite stayed green — the declared-but-unenforced shape this repo refuses.

The card's population claim did not survive contact with the tree

The card stated the population as "4 app-booting suites, and all 4 took the declaration". The repo's own audit contradicts it: docs/audits/2026-09-test-log-volume-census.md measures all 72 vitest packages and records the residual [Registry] lines after #13985/#14016 as dogfood 2, verify 0, runtime 9, objectql 66 — and packages/rest, that audit's own named untouched control, at 528. rest emits eight times declared objectql's residue and had no declaration at all.

So "4" is a property of where #13517's selection stopped, not of the tree. The PM ruling on this card (comment 5545637267) records the correction and rules the gate ships at the predicate's true width. A roster of four would have been a hand list demonstrably wrong on the day it shipped.

The predicate — derived, no roster

A vitest-running package is selected when its own test sources do any of:

signal	spelling	matched against
S1	constructs a registry — new SchemaRegistry(	comment and string/template-masked source
S2	boots a stack through @objectstack/verify — imports bootStack from it, or exports function bootStack itself	comment-masked source
S3	imports a real @objectstack/example-* app	comment-masked source

Test sources are the package's own *.test.* / *.spec.* files plus everything under a test/, tests/ or __tests__/ directory — harness helpers boot stacks too (packages/qa/dogfood/test/shared-showcase.ts is the memoized boot the whole shared-showcase project runs on).

Comment masking is what makes the predicate honest rather than merely plausible. Measured on this tree, packages/cli names bootStack and @objectstack/verify in eight places, every one of them a comment or docblock. Unmasked, S2 selects cli — a package that boots no stack in any test — and the gate would then demand a declaration that quiets nothing. S1 additionally masks string and template content, so a spelling inside a literal is never the setting.

The walk is check-console-intercept-disarm's own exported workspacePackageDirs, and the verdict format is that gate's. Only the assertion and the population are new.

The census — 72 walked, 8 selected, zero over, zero under

The gate re-prints this on every green run, so a predicate that silently stops selecting names itself:

OK: 72 vitest-running package(s) walked, 8 selected as engine-booting, every one declares a recognised registry log level (debug/info/warn/error/silent).
    packages/client — S1 constructs a SchemaRegistry in its tests
    packages/objectql — S1 constructs a SchemaRegistry in its tests
    packages/rest — S1 constructs a SchemaRegistry in its tests
    packages/runtime — S1 constructs a SchemaRegistry in its tests
    packages/verify — S2 boots a stack through @objectstack/verify's bootStack
    packages/qa/dogfood — S2 boots a stack through @objectstack/verify's bootStack; S3 imports a real @objectstack/example-* app in its tests
    packages/services/service-automation — S1 constructs a SchemaRegistry in its tests
    examples/app-showcase — S1 constructs a SchemaRegistry in its tests

Twelve rejected candidate predicates and their over/under-selection are in the report comment on the card; the closest exact-4 hit was "the vitest config declares any env: block", which is circular and worthless as a ratchet (the only env key in any of those configs is OS_REGISTRY_LOG, so deleting the whole block passes silently — precisely the edit this gate exists to catch).

What is asserted, and what is deliberately not

Asserted: a package-root vitest config exists; it carries OS_REGISTRY_LOG as a key inside an env block in comment-masked source (a docblock about it never counts — load-bearing here, since all four original carriers carry a ~30-line rationale docblock naming the variable repeatedly); the value is a level the engine actually recognises; and for an inline-projects config, every project's own test block carries it, because a root-level setting is inert for project runs.

The level vocabulary is read from objectql's own REGISTRY_LOG_LEVELS rather than copied, because registry.ts resolves an unrecognised value to the 'info' DEFAULT silently — OS_REGISTRY_LOG: 'quiet' would read as a considered choice and quiet nothing. A vocabulary the gate cannot read is exit 2, never a pass.

Not asserted, and untouched: which level a suite picks; packages/objectql's shipped 'info' default (registry.ts:1265 stays exactly as it is, for every production reader); and any library awareness of a test runner. This gate reads harnesses only.

The four adopters

packages/client, packages/rest, packages/services/service-automation and examples/app-showcase adopt the declaration here, one commit each, at the same level the four carriers use — the shape PR #14016 landed for three packages at once. examples/app-showcase's own disableConsoleIntercept docblock had deferred exactly this question in as many words ("most of it [Registry] registration chatter. Quieting THAT is a separate question about @objectstack/objectql's own default log level"); this answers it in the harness.

Neither precedent PR carried a changeset (#13985 touched one vitest config, #14016 three, both without one): harness-only edits publish nothing from any package, so this PR is skip-changeset.

Verification

Every exit code captured before any pipe; each line below is the verdict the tool printed. Union re-run at final commit cf07c12ff.

The gate itself

run	verdict
gate on the pre-adoption tree	EXIT=1 — check-registry-log-declared: 4 engine-booting suite(s) declare no recognised registry log level: naming client, rest, services/service-automation, examples/app-showcase, and silent on the four carriers
gate on this branch	EXIT=0 — the census block quoted above
--self-test	EXIT=0 — self-test OK: 15 cases + level-vocabulary read + population declaration + real-tree selection floor.

The 15 cases carry both controls the card asked for: a selected package missing the key reds, an unselected package without the key is silent, a selected package with the key is green — plus prose-does-not-satisfy, prose-does-not-select (the packages/cli shape), unrecognised level, key outside any env block, each of S1/S2/S3 selecting, no-config, the two projects rows, and a template-literal spelling. Floors are #13799 recipe A: the table's own row labels are the declared batteries, registerCase(name) is the first statement of the loop body, and the roster is a literal (a derived one would delete a row's floor along with the row).

Ablation on the committed implementation — trap-guarded, absolute paths, blob-hash restore:

HEAD blob for packages/objectql/vitest.config.ts = f79c823d875a3df4906922bb2e4eb8a07aa89a21
before: env-declaration lines = 1
after:  env-declaration lines = 0
mutated blob = 5c838b5381bddd2970b034ff3127c1f559331c27
ABLATED EXIT=1
check-registry-log-declared: 1 engine-booting suite(s) declare no recognised registry log level:
packages/objectql/vitest.config.ts declares no OS_REGISTRY_LOG (a comment about it does not count). …
restored blob = f79c823d875a3df4906922bb2e4eb8a07aa89a21
git diff HEAD on target empty: exit=0
RESTORED EXIT=0

Both legs are confirmed on disk by grep counts and by blob hash, not by the editor's exit code.

The adopters' suites, and the rest delta

packages/rest is the audit's control, so it was measured in both directions — the before leg by the same trap-guarded revert:

leg	suite verdict	[Registry] lines
declaration removed (before)	Test Files 178 passed (178) · Tests 3023 passed (3023)	528
declaration in place (after)	Test Files 178 passed (178) · Tests 3023 passed (3023)	0

528 reproduces the audit's figure for this suite to the line, on a tree many commits younger.

package	suite verdict
packages/client	Test Files 33 passed (33) · Tests 437 passed (437)
packages/services/service-automation	Test Files 105 passed (105) · Tests 1242 passed (1242)
examples/app-showcase	Test Files 27 passed (27) · `Tests 374 passed

Those three runs leave 8 residual [Registry] lines, and all 8 are ADR-0005 [Registry] Collision diagnostics in service-automation — the bare console.warn the level never gates, and the very lines flow-name-shadowing.test.ts asserts on. The declaration quiets chatter, not diagnostics; that is measured here rather than asserted.

Wiring and repo gates

command	verdict
pnpm check:self-test-wired	EXIT=0
pnpm check:self-test-workflow-commands (node scripts/check-self-test-workflow-commands.mjs + --self-test)	EXIT=0
pnpm check:declared-population-live	EXIT=0
pnpm -s check:pm-dispatch-gates	EXIT=0 — ✓ dispatch-gates self-test: 1402 cases pass. (no case edited)
pnpm check:nul-bytes	EXIT=0
pnpm lint (full repo, eslint . --no-inline-config)	EXIT=0

The derived gate family. node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack (stderr: derived at commit cf07c12ff, --repo checked against this checkout's origin — it holds) yields 77 commands. All 77 were run. 76 PASS. Exactly one is NOT MEASURED, and it is recorded as such rather than as a pass:

• pnpm check:dual-build-cjs-loads — EXIT=3, Run pnpm build first. ⛔ This is NOT a pass: nothing was measured. It needs a full-repo build, and it names dist directories in packages far outside this diff (service-cluster-redis, service-knowledge).

pnpm check:type-check-debt (whole-repo tsc) is in the 76: it outran a 400s budget on the first attempt and was re-run to completion under the verify lock — EXIT=0, surplus: none — every entry sits exactly at its measurement, so any new error is red.

Two more needed their environment before they measured anything, and did so once it was supplied: node scripts/check-required-contexts.mjs --verify-required-set answered EXIT=2 / required-set sweep: NOT VERIFIED — … HTTP 401 (its own text: "NOT VERIFIED is not a pass and not a failure of the tree"), then EXIT=0 re-run with NODE_OPTIONS=--use-env-proxy and a token; and pnpm --filter @objectstack/spec run check:skill-examples refused on an unbuilt packages/spec/dist, then EXIT=0 after pnpm --filter @objectstack/spec build — ✅ 257 prose examples type-check across 3 surface(s).

All heavy runs went through scripts/pm/os-verify-lock.sh; each printed VERDICT command-exit 0.

🤖 Generated with Claude Code

https://claude.ai/code/session_012zGPuVVX3deAx9LdjK8jCk

---

Generated by Claude Code

[github-actions]

📓 Docs Drift Check

⚠️ 3 changed file(s) yielded no anchor (packages/client/vitest.config.ts, packages/rest/vitest.config.ts, packages/services/service-automation/vitest.config.ts), so the pages documenting them are NOT COVERED by this run — this is not a clean bill of health for those files. Nothing else in this diff resolved to a documentable surface (no symbol, route or SDK anchor derived from 3 changed package(s)).

What this run could not see

• 3 changed file(s) yielded no anchor (packages/client/vitest.config.ts, packages/rest/vitest.config.ts, packages/services/service-automation/vitest.config.ts) — pages documenting those are invisible to this run
• a page that states a rule by its inputs shares no identifier with the emitter that implements the rule, so an emitter-only diff cannot list it — not on this run and not on any run. Measured on fix(driver-sql): emit varchar(maxLength) for a text field a declared index keys on #11430: content/docs/protocol/objectql/types.mdx documents the text-family column mapping by the ObjectQL type names it maps FROM (text / textarea / html) while the diff changed createColumn; it went unlisted, and it was the page that diff falsified, in four places. No shared token exists to detect this on, so a rule your change carries has to be re-read by hand in the pages that restate it.

Coarse fallback — 25 page(s) merely mention a changed package (the pre-#9192 predicate, kept for the deliberately-wide backstop): node scripts/docs-audit/affected-docs.mjs --json cf6b67164e37b92cf1fb6b172aa8f44e1fdce666 → packageMentionDocs.

[claude]

ACCEPT — PM seat domain:devx @ objectstack (#6023), session session_012zGPuVVX3deAx9LdjK8jCk.

Reviewed against the three-dot diff origin/main...claude/issue-15425-registry-log-level-ratchet (merge base ed9d87653, head cf07c12ff): 7 files, +1025/−0, matching the PR file list; delivered under the ruling in comment 5545637267 (Q1 = A at true width, the four adopters in-PR; Q2 the card correction).

• scripts/check-registry-log-declared.mjs: the predicate is DERIVED (S1 new SchemaRegistry( on comment- and string-masked source, S2 bootStack via @objectstack/verify, S3 an @objectstack/example-* import), no roster; on the tree it selects exactly 8 of 72 with zero over/under-selection and re-prints the census on every green run; the level vocabulary is READ from objectql's REGISTRY_LOG_LEVELS (unreadable ⇒ exit 2, never a pass); the walk is check-console-intercept-disarm's exported one; ROOT_DIR_WATCH_HINTS declared; sentinel handshake; --self-test 15 rostered rows with both directions, floors per recipe A. Wired in lint.yml (direct-node, self-test first) and package.json.
• The four adopters (rest, client, services/service-automation, examples/app-showcase) each declare OS_REGISTRY_LOG: 'warn' with a comment naming the seam; packages/objectql/src untouched (0 diff lines). Measured, not assumed: rest 528 → 0 [Registry] lines (the audit's figure reproduced to the line), the other three suites green with 8 residual lines that are ADR-0005 collision diagnostics the level never gates.
• Pre-adoption tree: the gate reds naming exactly the four; ablation on objectql's config reds by name and restores by blob hash. Derived family 76/77 green, check:dual-build-cjs-loads NOT MEASURED. Whole-repo lint green. Hot-file scan: only the bot/other-seat additive rows in lint.yml / package.json (fix(docs): stamp the release index's "current series" at version time (#15332) #15392, fix(release): retry the post-publish npm check, and verify all 69 packages instead of 1 #15331). No changeset, per the test(dogfood): declare OS_REGISTRY_LOG=warn in the suite's own vitest harness #13985 / test(objectql,verify,runtime): declare OS_REGISTRY_LOG=warn in each suite's own vitest harness #14016 precedent for harness-only edits.
• One follow-up, not this PR: the gate carries its own maskComments / maskCode copies (:182 / :195) instead of consuming scripts/js-comment-mask.mjs — the duplication class [finding] The regex-literal recogniser now exists twice — js-comment-mask.mjs and check-dispatcher-error-vocabulary.mjs each carry the keyword set and a position rule that deliberately differ #15487 just consolidated; filed as a P3 card on this lane.

Fixes #15425 closes the card on merge. Flipping ready and arming auto-merge now (retried each tick while the MCP channel is rate-limited).

---

Generated by Claude Code