문제
레지스트리 호환성 분류기(scripts/lib/registry-compatibility.ts)는 값의 JS
타입이 다르면 breakingFields를 보기 전에 field-type-changed(breaking)로
판정한다. null과 문자열은 타입이 달라서, governance가 string|null로
선언하고 breaking 필드에서 뺀 필드(navigationLabel, navigationOrder, API
timeoutMs, 저장 ttl, ENV defaultValue)를 null <-> 값으로 바꾸기만 해도
breaking 증거를 요구한다. contract의 allowedValues에 값을 더하는 것도
어떤 변경이든 breaking이다. 그래서 증거 5필드에 쓸 내용이 없는 변경에
"마이그레이션·롤백"을 지어 쓰게 된다. 템플릿 자신도 VITE_BUILD_ID
defaultValue null -> "local-build"에 "the value itself did not change"라고
적은 breaking 증거를 남겼다.
고른 방법 (리뷰 최소 개선 B)
- 행 필드: 한쪽이 null이고 두 값이 모두 그 필드의 선언 타입
(contract.fieldTypes, 예 string|null)에 맞으면 타입 변경이 아니라
field-changed로 보고 breakingFields로 등급을 정한다. 선언이 없거나
선언 밖 값이면 지금처럼 field-type-changed(breaking)다.
- contract allowedValues: 필드 집합이 같고 필드마다 이전 허용값을 모두
포함하면 allowed-values-extended(additive)다. 값 하나라도 빠지거나,
새 필드를 제한하거나, 제한을 없애면 지금처럼 breaking이다. 나머지
contract 변경도 그대로 breaking이다.
- 선언 타입 판정은 check-registries.ts에 있던 runtimeType·
matchesDeclaredType을 lib로 옮겨 검사기와 분류기가 같은 함수를 쓴다.
- semantic-diff.json fixture에 nullable-field-cleared(behavior-change),
registry-contract-extension(additive)을 더하고 artifact id 목록에도
넣었다.
버린 대안
- 현 상태 + "이런 변경은 재승인하라"는 문서: 재승인이 증거를 대신하는
길을 더 넓힌다(0929/07-04).
- nullable 필드 판단 전체를 breakingFields로 옮기기(리뷰 C): 다른
레지스트리 영향을 다시 봐야 하는 넓은 변경이다.
- 제한 해제(allowedValues에서 필드 키 삭제)를 additive로 보기: 리뷰
기준("이전 값을 모두 포함")에 맞지 않고 판단이 하나 더 필요해 보수적으로
breaking에 남겼다.
위험·한계
- fail-closed 범위가 줄어든다. 선언 타입 안의 null 전환이 의미("기능
꺼짐")를 갖는 필드는 breakingFields에 넣어야 breaking으로 남는다.
paramsSchema·searchSchema는 이미 breakingFields라 그대로 breaking이다.
- 현재 기준선 diff에서 VITE_BUILD_ID defaultValue가 breaking에서
behavior-change로 바뀐다. registry-change-evidence.json의 해당 증거 행
(field-type-changed)은 이제 어떤 변경과도 맞지 않는다. 오래된 증거 정리는
0929/07-01 범위라 이번에는 지우지 않았다.
- 다운스트림의 같은 종류 증거 행도 쓰이지 않게 된다. 게이트 판정이
바뀌는 변경이라 다음 릴리스 CHANGELOG에 [계약] 행이 필요하다(이번
라운드는 CHANGELOG 미출시 절이 없어 적지 않았다).
- fieldTypes 키가 점 경로(a.b)인 필드는 행의 최상위 필드 이름과 맞지 않아
계속 breaking으로 남는다.
검증
- RED(분류기 수정 전): registry-compatibility.test.ts 새 사례 5건 실패
(null 전환 3건이 breaking/field-type-changed, nullable breaking 필드의
kind, allowlist 확장 breaking). check-registry-compatibility-fixtures:
nullable-field-cleared actual breaking, registry-contract-extension
actual breaking.
- GREEN: registry-compatibility·registry-governance·release-artifacts·
second-api-feature-registries 49/49 통과. compatibility fixtures 11 PASS.
check-registries "11 registries PASS"(남은 breaking은
REFERENCE_RESOURCE_DETAIL.path 하나, 증거 있음).
ci-artifact-contract "semantic validator" 사례 통과.
- 회귀: tsc node/test rc=0, 바꾼 파일 eslint rc=0.
- 미실행: check:architecture(pnpm). tests/system 나머지(provider guardian
exit=126 환경 문제로 이 변경과 무관하게 실패).
Refs: docs/reviews/2026-09-29-maturity-review/07-contracts.md
The classifier called type, enum and nullable changes "none" and treated new
keys under strict readers as additive; FE-GATE-014 only checked hand-written
fixtures. It now fails closed and compares the committed baseline with the
schema the build emits. The doc lists per-boundary reader rules.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The unit test read README.md, which removal fixtures do not copy, so
test:sample-removal failed. verify:documentation now requires README to name
every build and deploy variable the generator and admission read.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A misspelt kill-switch key was ignored without a trace, so the feature kept
running. Boot now records feature.override.unknown once per such key, and
release admission lists them as warnings in its record without refusing the
artifact (a shared document may still name another build's features).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The module inventory named dependency modules external:<file>, so no artifact
said which package versions reach the public bundle. It now records
npm:<name>@<version> ids and a packages list. supply-chain.md states what the
SBOM, bundle list and provenance prove, and one SBOM scope mismatch (scheduler).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
FE-GATE-019 ran before the release existed and passed by comparing the declared
headers with a copy of themselves. The fixture result is now UNVERIFIED, the
gate runs in the production stage after deployment, and the live probe
requests the entry script the served index.html names (D-14).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Provider evidence expires two hours after issue, so no repo tool could
re-verify a release before deploying it later or rolling back to it.
verify:promotion --archived keeps every signature, digest and member check and
judges freshness at the recorded promotion time; the FE-RB-005 drill takes the
extracted release with --release-root.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
dependency-cruiser 18 cannot parse TypeScript 7, so it cruised no module, passed,
and its report and the docs claimed it checked the layers. The import graph is
the only checker; .dependency-cruiser.json stays as its rule file (D-11).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Icon vendor, design-system deep import and realtime composition were regular
expressions beside graph rules, and the regex missed relative deep imports
inside presentation. They are graph rules with fixtures now, with the
components/ui facades and page-templates named. package-script-graph.ts stays;
the reason is recorded.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Drills held constant true verdicts and read public/ placeholders, so they
passed with no build or with JS assets deleted. They now check the local dist,
mark boot, live traffic and provider steps UNVERIFIED (exit 3), and FE-GATE-016
is retired because it repeated FE-GATE-025's FE-RB-005 drill.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Three boundaries the layer contract declares had no executable rule behind
them, so the code drifted across all three while every gate stayed green.
`src/contracts` reached back up into `src/application` for the shared `Result`
carrier and the compatibility predicate. Neither package owned the shared
vocabulary and the dependency pointed both ways. Both now live in contracts —
the lower package — and application re-exports them, so no caller moves.
A concrete adapter was not supposed to depend on another concrete adapter, but
only adapter-to-presentation was enforced, and `diagnostics` imported a guard
out of `telemetry`. The guard belongs to neither, so it moved to the adapter
kernel. Stating the rule needed the checker to resolve `$1` in a `to` pattern
against the importing module's own directory; the alternative is one rule per
adapter group, which silently stops covering a group the moment one is added.
Product assembly leaks out of bootstrap: generic presentation reads the
installed-feature registries. That is a real refactor, so the rule freezes the
exact set of modules doing it today rather than pretending it is fixed — a new
edge fails. The two remaining open edges are named in the config, not silent.
Each rule was verified by introducing the violation it forbids and confirming
the gate rejects it.
The documentation drifted the same way. README and the manual accessibility
checklist both said six routes while ten were registered, which left the
platform overview and three reference-resource screens outside the declared
manual review scope without anyone deciding they should be. The scope is now
derived from the route registry by `verify:documentation`, so the sentence
cannot outlive the registry again. The review ledger also named a canonical
path that does not exist in this tree; it is upstream provenance, and it now
says so instead of looking like a broken repository reference.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The provider sandbox never ran. bubblewrap 0.9.0 stops parsing an `--args`
file at the first non-option and never hands the remainder back, so the
command written into that file was silently dropped: bwrap printed its usage
text, exited 1, and the provider produced no evidence at all. The options
still travel in the args file — that is what keeps host paths and credentials
out of `/proc/<pid>/cmdline` — but the command now rides on real argv, and
`encodeProviderBwrapInput` refuses a `--` so the drop cannot come back.
The scope wrapper then could not exit. It read the supervisor's liveness pipe
through `fs`, which runs a blocking `read(2)` on a threadpool thread; the
supervisor holds that pipe open for the scope's whole life, so the read never
returned and closing the descriptor did not interrupt it. Once bubblewrap
finished the wrapper deadlocked in `process.exit`, the scope outlived the
provider, and a completed run was reported as a timeout kill. The channel is
now read through the event loop, so teardown is observable and terminal.
Creation modes were left to the ambient umask. `mkdir(mode)` and `open(mode)`
are requests the kernel subtracts the umask from, so a runner exporting a
restrictive umask produced directories it could not enter and handed `tar` a
file it could not re-open. Private modes are pinned instead of inherited.
Promotion cleanup deleted before it checked. Removals run through a pinned
descriptor, so a leaf substituted after validation had this promotion's exact
five destroyed first and the substitution reported afterwards, leaving a
half-emptied directory a retry could not tell from a completed one. The name
is re-bound to the inode before anything is removed, so the failure is total.
Separately, release coherence proved the artifacts agreed with each other but
never that they belonged where they were going: a build whose runtime document
said `APP_ENV: local`, `AUTH_MODE: demo` and a loopback API is coherent with
itself and passed every gate. `public/` is copied verbatim into `dist/`, so
that local document shipped with every build regardless of what the build was
for. Runtime configuration now comes from a declared profile, and FE-GATE-027
refuses to admit an artifact to an environment it does not match — including
refusing an undeclared destination, so nothing is admitted by omission.
`REQUEST_TIMEOUT_MS` and `VITE_ROUTER_BASE_PATH` were validated and then
dropped: the V3 executor ran every operation on its contract's own deadline,
and Vite emitted root-absolute assets for a sub-path deployment. The timeout is
now a ceiling that may tighten a contract but never loosen one, and one base
path feeds the router, the Service Worker scope and the asset base together.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>