chore(gates): layering baselines ratchet against merge-base

[thymikee]

Summary

R6 (type-only spine inversions per zone pair), R9 (largest type-level cycle, whole and per zone)
and R10's R7 ownership pressure stop comparing against hand-edited constants. Each now measures
the merge-base with origin/main and compares it with the working tree, through the same
functions on both sides.

For a change author: growth still fails with the same message shape (now "baseline N at the
merge-base"), a shrink needs no edit and is banked on merge, and no PR can leave a number above
the tree to spend later. TYPE_INVERSION_BASELINE, LARGEST_TYPE_CYCLE_ZONE_CEILINGS,
TYPE_CYCLE_BASELINE and DAEMON_MODULARITY_BASELINE.sessionState are deleted;
externalDaemonTypesImporters stays, because it names files rather than counting them. R9's
per-zone overflow now names the file that joined instead of listing the whole zone, because the
reference carries membership.

The base tree is read through the committed-tree reader from #2257 (one git ls-tree, one
git cat-file --batch, no second checkout) and one memoizing parser serves both trees, so
pnpm check:layering is unchanged at ~12 s. Repo Guards now checks out full history for it.

Validation

At b7a15dd: pnpm check:layering (209 rule tests + gate), depgraph:test,
check:gate-manifest, check:fallow, check:quick, format:check, and the reader's existing
consumer scripts/__tests__/eager-closure-budgets.test.ts (418 tests).

Planted red, each via node --experimental-strip-types scripts/layering/check.ts:

• R6, one new commands -> client type import: inversions grew to 4 (baseline 3 at the merge-base)
• R9, one file pulled into the provider-webdriver cycle: grew to 7 files (baseline 6 at the merge-base)
• R10, one new writer of session.snapshot: R7 ownerFileClaims grew to 23 (baseline 22 at the merge-base)
• Shrink: deleting one mcp -> client inversion leaves the gate green (exit 0, "R6, 4 remaining")
  with no edit; the old equality pin failed that case.

Full affected gate: green at c3dca6e (pnpm check:affected --run; 32 local gates incl. format/lint/typecheck, layering, di-seams, fallow, mcp-metadata, build, package, integration-node/progress, macos-coverage, vitest-related, replay-compat, daemon-wire-compat, affected-selector, gate-manifest(-model), depgraph, tmpdir-leaks(-model), coverage-model, wire-compat-model, production-exports, bundle-owner-files, fixture-cache/fallback, command-docs, agent-guidance, xctest-selection, maestro-conformance, mutation-model; rest GitHub-authoritative)

[github-actions]

Size Report

Metric	Base	Current	Diff
JS raw	2.61 MB	2.61 MB	0 B
JS gzip	879.1 kB	879.1 kB	0 B
npm bundled raw	2.61 MB	2.61 MB	0 B
npm bundled gzip	879.1 kB	879.1 kB	0 B
npm tarball	1.04 MB	1.04 MB	-3 B
npm unpacked	3.52 MB	3.52 MB	0 B
npm clean-installed	3.52 MB	3.52 MB	0 B

npm unpacked components

Component	Base	Current	Diff
JS / dist source	2.77 MB	2.77 MB	0 B
Apple runner source/project	544.4 kB	544.4 kB	0 B
Apple snapshot presentation source	33.8 kB	33.8 kB	0 B
Apple Simulator snapshot bridge source	30.5 kB	30.5 kB	0 B
macOS helper source	54.8 kB	54.8 kB	0 B
Android helper artifacts	43.3 kB	43.3 kB	0 B
Other package files	46.7 kB	46.7 kB	0 B

Startup median (7 runs, lower is better):

Scenario	Base	Current	Diff
CLI --version	26.6 ms	27.0 ms	+0.4 ms
CLI --help	82.1 ms	77.5 ms	-4.6 ms

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

Top changed packed files

No changed packed files.

[thymikee]

Rebased onto current origin/main (past #2293 R12 retirement and #2294 navigation-projection retirement). Conflicts were in scripts/layering/check.ts and docs/dependency-graph-findings.md, both on the same hunk: main's rewritten R6 commentary (reflecting the retired projection) vs. this PR's replacement of that commentary with a short placeholder ahead of deleting TYPE_INVERSION_BASELINE. Resolved by keeping this PR's placeholder/rewording while carrying forward main's updated projection-retired content, since the whole block is deleted by this PR's second commit anyway. No rule id or deleted constant was resurrected.

New head: d9db522555.

Focused checks all green post-rebase: pnpm check:layering, pnpm check:quick, pnpm depgraph:test, pnpm check:fallow, pnpm format:check.

Re-verified the planted reds still fire with the same shape: R6 (commands→client grew 3→4), R9 (largest type cycle grew 6→7 via provider-webdriver), R10 (ownerFileClaims grew 22→23), each reported against the merge-base as before; reverted after confirming.

The full affected gate will be re-run on this head by the serialized gate stage.

[thymikee]

Reviewed d9db522: no actionable findings. R6/R9/R10 use the existing committed-tree reader and shared measurements; growth fails and shrink needs no hand-edited ceiling. Full-history CI checkout supplies the merge-base, and post-rebase planted-red evidence covers each ratchet. Tooling-only, zero unpacked growth. Code-review clean; remaining CI must finish before an unconditional merge-ready claim.

[github-actions]

PR Preview Action v1.8.1
Preview removed because the pull request was closed.
2026-09-05 18:49 UTC