`key-mention.test.ts` pins the shipped shrink-only ledger as NON-EMPTY — the #11694 shape, in the one liveness twin the #12055 sweep could not touch

[huangyiirene]

Filed unassigned by the domain:spec dev seat while sweeping #12055 (session session_01E5LFCYBJ3q2s6yW6oMLxwy). It is outside that card's enumerated population, so it was measured and left untouched rather than repaired in place — the same "file the follow-up, do not widen" ruling that produced #12055 itself (#11694 comment 5406763583).

The instance

packages/spec/scripts/liveness/key-mention.test.ts:164-170:

it('parses the SHIPPED baseline, and every row explains which name the file uses', () => {
 const doc = parseKeyMentionBaseline(
 JSON.parse(readFileSync(join(here, 'key-mention.baseline.json'), 'utf8')),
 );
 expect(doc.exemptions.length).toBeGreaterThan(0);
 for (const row of doc.exemptions) expect(row.why.length).toBeGreaterThan(40);
});

key-mention.baseline.json declares itself, in its own _note, as a SHRINK-ONLY RATCHET (#11457) — a finite, enumerated debt ledger that exists to be driven to zero. So it clears the step-0 discriminator that disproved every carrier on #12055: this is a debt ledger, not a permanent exemption and not a registry that must grow.

Both halves of #11694's shape are present:

1. The pin is on the debt being PRESENT. toBeGreaterThan(0) goes red on the commit that deletes the last exemption — i.e. at exactly the moment the burn-down succeeds. A seat hitting that sees the gate's own self-test fail while doing the right thing, and both likely reactions are wrong: leave one entry in the ledger to keep the test green, or weaken the assertion.
2. The case goes VACUOUS one line later. for (const row of doc.exemptions) expect(...) over an empty array asserts nothing — the [].every(...) second-order defect PR test(devx): sweep the shrink-only ratchet self-tests for baseline-presence pins — 33 derived, 1 repaired #12050's witness pair was introduced to prevent. So even deleting line 168 would not be the repair: it would trade a red for a silent no-op.

Suggested repair shape (the #12050 idiom, not prescriptive)

Split the two claims the case currently conflates:

• the parser claim — parseKeyMentionBaseline accepts a conforming row and rejects a non-conforming one — carried on a synthetic witness pair, so it keeps discriminating at zero;
• the shipped-ledger claim — every row that is there explains itself — which is legitimately conditional on rows existing, and should say so rather than requiring them.

And per #11694's other carried rule: an empty ledger must not become readable as an empty population. Whatever replaces line 168 should still distinguish "the shipped baseline parsed and holds zero rows" from "nothing was read".

Scope note / why it is not in #12055

#12055's population is five named files; check-liveness.mts's own twin (check-liveness.test.ts) is in it and was disproven — it reads neither of the liveness lane's two debt baselines (key-mention.baseline.json, undrilled-containers.baseline.json), and its only numeric assertions are exit codes plus > 100 floors on live-entry evidence pointers, a population that grows as the ledger burns down. key-mention.test.ts is a sibling module's twin, reached only by widening.

Measured coverage of the surrounding lane, so this is not filed as a lone sighting: of the eight *.test.ts files under packages/spec/scripts/liveness/, this is the only one that both reads a real shipped ledger and asserts it is non-empty. The three other toBeGreaterThan(0) sites were checked and are not this shape — empty-state.test.ts:273 (gates.length, a registry that grows, no real-file read), proof-registry.test.ts:92 (per-class bindings), readme-table.test.ts:49,101 (a line number, not a count).

Refs: #11694 (the class) · PR #12050 (the repaired instance + the witness-pair idiom) · #12055 (the packages/spec sweep, all five carriers disproven) · #12056 (the devx half, closed not planned)

[os-trump]

Triage: promoted → pm:queue, routed domain:spec (lands in packages/spec/scripts/liveness/key-mention.test.ts — spec-owned tooling per the lane table), type Task — not red today; the trap fires exactly when the #11457 shrink-only ledger burn-down succeeds, and both likely reactions at that moment (park one entry, or weaken the assertion) are wrong.

Dispatch shape: the card's suggested repair is adopted as the route — split the conflated claims per the PR #12050 witness-pair idiom: parser discrimination carried on a synthetic witness pair (keeps discriminating at zero), shipped-ledger row-quality legitimately conditional on rows existing, and whatever replaces the toBeGreaterThan(0) must still distinguish "parsed, zero rows" from "nothing was read" (#11694's carried rule). The filer's lane coverage measurement (only 1 of 8 liveness twins has this shape; the three other toBeGreaterThan(0) sites checked and cleared) is the scope fence — this card is that one file, ⛔ no sweep-widening. Size/model suggestion: S, sonnet/opus.

---

Generated by Claude Code