test: pin daemon modularity migration contracts

[thymikee]

Summary

Pin the released behavior and architecture baselines needed to safely execute the daemon modularity work in #1478.

This adds characterization coverage for replay-test reporters and Maestro-compatible discovery, plus executable ratchets for SessionState ownership, dependency-cycle size and zone membership, external daemon/types.ts importers, logical-module direction, and private internal/ trees. Production behavior is unchanged.

The ratchet implementation keeps one source of truth for each concept: the cycle total is derived from its zone ceilings, layering violations share one model type, and SessionState field counts live with the ownership model.

Scope is limited to P0: 12 touched files. P1 and the in-progress P2 cutover are intentionally excluded.

Part of #1478.

Validation

check:affected passed formatting, lint, typecheck, layering, Fallow, metadata sync, build, and all 19 directly affected tests. The layering gate reports R6 at 7, all SessionState fields classified, R7 at 30 writer-owned fields / 42 owner claims, and R9 at 102 files.

The full unit lane saturated the host and produced unrelated five-second timeouts across eight files. All eight failure files were rerun serially; 173 tests passed, including an exact comparison run of the remaining Apple runner cancellation case on both origin/main and the PR head. CI remains authoritative for the full aggregate and platform jobs.

No device verification was required because this PR changes tests, documentation, and enforcement tooling only.

[github-actions]

PR Preview Action v1.8.1
Preview removed because the pull request was closed.
2026-07-29 16:05 UTC

[github-actions]

Size Report

Metric	Base	Current	Diff
JS raw	1.87 MB	1.87 MB	0 B
JS gzip	602.9 kB	602.9 kB	0 B
npm tarball	720.7 kB	720.7 kB	+7 B
npm unpacked	2.52 MB	2.52 MB	+43 B

Startup median (7 runs, lower is better):

Scenario	Base	Current	Diff
CLI --version	27.9 ms	28.4 ms	+0.5 ms
CLI --help	57.9 ms	57.0 ms	-0.9 ms

Top changed chunks: no changes in the largest emitted chunks.

[thymikee]

Review at 1af5d3d: not ready for the P0 migration baseline.

• P1 — reporter characterization misses the risky contract. src/replay/test/reporters/__tests__/registry.test.ts covers only non-throwing synchronous onTestResult. It still passes if live-hook isolation in registry.ts is removed or if a later extraction starts awaiting live-hook promises. refactor(daemon): extract replay, Maestro, and replay-test modules #1478 requires preserving hook timing and error handling. Add throwing and rejecting live hooks, prove later reporters/final rendering continue with diagnostics, and prove a pending live promise is not awaited.
• P2 — accidental SessionState shape ratchet. scripts/layering/daemon-modularity.test.ts hard-codes total fields as 41, but R10 only ratchets writer-owned fields and owner claims while R7 already enforces exhaustive classification. This rejects a valid new immutable/store-owned field. Remove the assertion, or explicitly make and document total state size as a deliberate migration metric.

The discovery tests and remaining edge-derived R10 guards are durable, and exact-head CI is green. No ready-for-human until these two contract gaps are tightened.

[thymikee]

Addressed both findings in 2b14abd59.

• Reporter characterization now covers synchronous throws, rejected promises, diagnostic output, continuation to later reporters/final hooks, and a pending live-hook promise that is demonstrably not awaited.
• Removed the hard-coded total SessionState field assertion. R7 still proves exhaustive one-owner/store-owned classification against the declaration, while R10 remains limited to its documented writer-owned-field and owner-claim metrics.

Validated with format, typecheck, lint, 39 layering tests, 19 focused characterization tests, and Fallow.

[thymikee]

Exact-head code review is clean. The new R10 ratchets match the declared daemon-modularity migration constraints, and the reporter/discovery tests exercise production behavior with meaningful counterfactuals rather than source spelling. Static, integration, coverage, Fallow, mutation, and non-iOS smoke checks are green.

Readiness is currently held only by the failed iOS smoke: the fixture alert wait timed out while unrelated runner/preflight/replay coverage passed. This head does not change an iOS production path, so it is not a confirmed code regression, but the required check must be rerun or explicitly classified before merge. No ready-for-human label while CI is failing.