docs(layering): kill criteria on every rule module

[thymikee]

Summary

Every layering rule/policy module under scripts/layering/ now carries a four-line header —
Catches:, Evidence:, Cost:, Kill criterion: — stating what defect class the rule catches
and why no other gate sees it, a real PR/commit that motivated or exercised the rule, its LOC
cost (rule + test), and the concrete condition under which the rule gets deleted. This covers R2,
R4-R7, R9-R14, R16, R18, R19, R65-R73, and the rule-id uniqueness gate (rule-ids.ts). Modules that
share one file (daemon-modularity.ts for R9/R10; check.ts for R4/R5/R6) get one header per rule.
No behavior change, no code lines touched — comment-only.

R3 platforms-seam has no active declaration to annotate (zone-policy.test.ts asserts its
absence); R17 devices and R98/R99 are not declared anywhere under scripts/layering/ (R17 lives
outside this directory, R98/R99 are collision-test fixtures only) — see Tradeoffs.

Validation

• pnpm check:layering — green, 198/198 node:test assertions pass.
• pnpm check:quick (oxlint + tsc -b) — clean.
• npx vitest run --project unit-core scripts/layering — reports "No test files found": the
  layering suite runs under node --test via check:layering, not vitest's unit-core project,
  which doesn't include scripts/layering/** in its include globs (confirmed pre-existing).
• Planted-red: added a quoted 'R2 imposter-rule' string inside zone-policy.ts's new header
  comment (the exact mistake the task's CAUTION warns against), then ran pnpm check:layering.
  Observed failure:
  ✖ the layering rules currently in tree carry no unallowed collision —
  AssertionError: Expected values to be strictly deep-equal: + ['two rules answer to one id: R2 names commands-floor and imposter-rule. Allocate the next free number.'] - []. Reverted; gate
  is green again.
• Full affected gate: pending (serialized gate stage will append the result).

Tradeoffs / follow-ups

• R3 and R17 have no module under scripts/layering/ to annotate today (R3 is retired with only
  a negative test; R17 "devices" lives outside this directory per rule-ids.ts's own history
  comment). R98/R99 only appear as example ids inside rule-ids.test.ts's collision fixtures, not
  as declared production rules — rule-ids.ts itself got a header instead, describing the
  uniqueness gate it implements.

[github-actions]

Size Report

Metric	Base	Current	Diff
JS raw	2.58 MB	2.58 MB	0 B
JS gzip	868.4 kB	868.4 kB	0 B
npm bundled raw	2.58 MB	2.58 MB	0 B
npm bundled gzip	868.4 kB	868.4 kB	0 B
npm tarball	1.02 MB	1.02 MB	+6 B
npm unpacked	3.46 MB	3.46 MB	0 B
npm clean-installed	3.46 MB	3.46 MB	0 B

npm unpacked components

Component	Base	Current	Diff
JS / dist source	2.74 MB	2.74 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
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	46.3 kB	46.3 kB	0 B

Startup median (7 runs, lower is better):

Scenario	Base	Current	Diff
CLI --version	32.9 ms	31.0 ms	-1.9 ms
CLI --help	85.3 ms	82.4 ms	-2.9 ms

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

Top changed packed files

No changed packed files.

[thymikee]

BLOCKED — several new kill criteria claim replacement enforcement that does not exist:

1. [P1] R18/R70 cannot be retired based on package manifests. Node built-ins are not package.json dependencies, globals need no import, and those rules also protect additional mechanics/construction and retired-root invariants. A zero dependency constraint would leave the violations legal. Name complete enforced replacement boundaries or retain the AST gates.

2. [P1] R72/R73 are not replaced by exports maps. Exports maps restrict external package specifiers, not internal relative calls, required engine→presentation topology, or provider-local residue construction/discard/reassignment. State structural replacements for each invariant or retain these gates.

The same overclaim appears in R11/R13/R16: project references, publishing, or typed ports do not by themselves make relative tunnels, eager imports, or Node/global mechanics unresolvable. Audit every kill criterion to require either an actually enforced replacement or an explicit decision that the invariant no longer matters. Exact-head CI is green; comments only, but the guidance is not safe to merge as written.

[thymikee]

Updated exact-head status: still blocked at 92bd3dc on the previously reported unsafe kill-criterion guidance, and now also blocked by a current-main conflict in scripts/layering/check.ts after #2240. Rebase, preserve main's R4/R5 commentary, correct the challenged guidance, and rerun affected exact-head checks. Existing green checks predate the rebase.

[thymikee]

Rebased onto main at d04e213: main's R4/R5/R6 commentary in check.ts kept verbatim (including #2241's "sole owner of the ratchet" note), R14/R71 headers carried into retired-paths-policy.ts, headers for the two deleted modules dropped.

Every kill criterion re-audited against the A4 spike (project references are a build cache, not a boundary). None names an enforced full replacement, so all 25 now read "none enforced today; retire only by maintainer decision that no longer matters": R2, R4, R5, R6, R7, R9, R10, R11, R12, R13 (x3), R14/R71, R16 (x2), R18, R19, R65-R70, R72, R73, rule-ids. Partial mechanisms are named as partial only: R11 cites tsc TS2307 (NodeNext exports) and TS6059 (composite rootDir) for two of four branches; R18/R70 state manifests do not govern built-ins or globals; R72/R73 state exports maps do not restrict relative calls or residue handling.

Reran check:layering (198/198, guard OK), format:check, check:quick. Head: 447e7cf.

[thymikee]

BLOCKED at 447e7cf1917d250801d2d74ff6c02ec647cfc6dd.

No remaining code/documentation finding. The unsafe replacement kill criteria are corrected, the check.ts conflict is resolved, and the headers now document structural invariants rather than excuse workarounds.

The only blocker is required iOS Smoke CI: TEXT_INPUT_COMMIT_NOT_OBSERVED while filling the fixture email field. This is unrelated to the comments-only layering diff and strongly consistent with a runner/device flake, but this exact head still needs a successful rerun or explicit baseline control before readiness. All other exact-head checks are green and package size is unchanged. Do not label ready while required CI is red.