fix(layering): list the whole zone when R10's type-cycle ceiling is exceeded

[thymikee]

Summary

pnpm check:layering R10 named the wrong file when a zone exceeded its type-cycle ceiling. checkTypeCycleBaseline reported members.find(<zone match>) — the alphabetically-first zone member — so the #1825 × #1779 main break read [R10] 17 daemon-server files (baseline 16) at src/daemon/daemon-command-registry.ts, a file that had been in the cycle all along. The actual +1 (src/daemon/handlers/snapshot-interactor-capture.ts) was found only by diffing largestTypeCycleMembers between two commits, ~40 min in.

LARGEST_TYPE_CYCLE_ZONE_CEILINGS records a count per zone, not a membership, and the Layering Guard job checks out at depth 1, so there is no previous state to diff against inside the gate. The violation now lists every member of the over-budget zone (sorted) and annotates the ceiling table (scripts/layering/daemon-modularity.ts:1, like the sibling R9/R10 baseline violations) instead of an arbitrary member.

Before:

[R10 daemon-modularity] src/daemon/daemon-command-registry.ts:1 — the largest type cycle now contains 17 daemon-server file(s) (baseline 16); extraction must not trade one zone's locality for another's.

After (planted: ceiling lowered 16 → 15 on the real tree):

[R10 daemon-modularity] scripts/layering/daemon-modularity.ts:1 — the largest type cycle now contains 16 daemon-server file(s) (baseline 15); extraction must not trade one zone's locality for another's. 1 over the ceiling — the member(s) that joined are among these daemon-server files: src/daemon/daemon-command-registry.ts, src/daemon/deferred-interaction-outcome.ts, src/daemon/handlers/response.ts, src/daemon/handlers/snapshot-capture.ts, src/daemon/interaction-outcome-policy.ts, …, src/daemon/session-store.ts. Cut the edge that pulled them in rather than raising the ceiling.

The count is deliberately phrased as net overflow, not a join count: a change that adds two members and removes one prints 1 over the ceiling, so the message bounds nothing except that at least one listed file is new.

No policy change: same ceilings, same rule id, same pass/fail set. Recording per-zone member lists (which would let the gate print the exact diff) would turn the count ratchet into a membership pin — a stricter policy the issue explicitly does not ask for; noted as an option, not taken.

Closes #1837. Refs #1781 A6, #1825, #1779.

Validation

• New unit test R10 zone overflow lists the whole zone so the joining member is visible plants a daemon-server +1 (src/daemon/snapshot-interactor-capture.ts, chosen to sort after the probes so the old first-member pick could not name it by luck; commands −1 keeps R9's total pinned) and asserts the single violation annotates the ceiling table and names all 17 members.
• Red against pre-fix code (git stash the production file, run scripts/layering/daemon-modularity.test.ts):

  ✖ R10 zone overflow lists the whole zone so the joining member is visible
    AssertionError [ERR_ASSERTION]: Expected values to be strictly equal:
    + actual - expected
    + 'src/daemon/probe-0.ts'
    - 'scripts/layering/daemon-modularity.ts'
  ℹ pass 9 / fail 1
  

  — the old code names the alphabetically-first probe. Restored: ℹ pass 10 / fail 0.
• Planted structural violation on the real tree (ceiling 16 → 15) prints the message quoted above; restored, node scripts/layering/check.ts is Layering guard: OK … the largest type-level cycle is 46 files (R9).
• pnpm check:affected --run: all runnable checks passed (layering, lint, typecheck, format, mutation-selection tests).

Touched files: 2 (scripts/layering/daemon-modularity.ts, scripts/layering/daemon-modularity.test.ts). Scope stayed on the layering gate. No docs/skills change: the gate's own message is the user-facing surface.

[github-actions]

Size Report

Metric	Base	Current	Diff
JS raw	2.31 MB	2.31 MB	0 B
JS gzip	759.3 kB	759.3 kB	0 B
npm tarball	882.6 kB	882.6 kB	0 B
npm unpacked	3.08 MB	3.08 MB	0 B

Startup median (7 runs, lower is better):

Scenario	Base	Current	Diff
CLI --version	30.1 ms	28.9 ms	-1.2 ms
CLI --help	72.2 ms	70.3 ms	-1.9 ms

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

[thymikee]

Diagnostic-accuracy gap: the whole-zone member list is useful, but ${zoneMembers.length - allowed} of these joined with this change is unsupported because the gate stores only a count ceiling, not prior membership. Net excess cannot prove how many entered: members may be replaced, or one edge may pull several existing files into the SCC.

Report only that the zone is N above its ceiling and direct authors to changed files/import edges; add a replacement/edge-closure regression.

Exact-head CI is otherwise green; no device evidence applies.

[thymikee]

Out of draft. CI is green on the actual head 60f88f3d (rebased onto current origin/main, so it carries #1860's corrected ratchet pin): 28/28 checks pass, Layering Guard included.

Second commit 60f88f3d applies the review nit: the overflow line was phrased as a join count, but it is net growth over the ceiling (two joins plus one departure would print 1), so it now reads N over the ceiling — the member(s) that joined are among these <zone> files: …. The pinned assertion in daemon-modularity.test.ts moved with it; the red-then-green proof re-ran on the new text (planted +1 → pass 9 / fail 1 against pre-fix code, pass 10 after), and the planted-ceiling run on the real tree prints the message quoted in the PR body.

[thymikee]

Reviewed exact head 60f88f3d: clean and good to go. The revised wording correctly reports net overflow rather than claiming an exact join count, the planted replacement case proves the zone-specific diagnostic and defeats the old alphabetical attribution, and listing the whole zone satisfies #1837 without changing policy. Scope/body are accurate; all 28 exact-head checks are green. No device evidence applies to this layering-only change.

[github-actions]

PR Preview Action v1.8.1
Preview removed because the pull request was closed.
2026-08-19 09:08 UTC