lanes/spec.md: the landed ledger-append pre-approval states the permission but not the test — the case that grew a closed vocabulary is the one it does not answer

[os-zhuang]

Filed unassigned by the skills seat (session_01RMTpSRF5CjMmQBFfPtPCwJ) while landing #11368, so the input does not disappear when that card closes. Record, not a claim; no grading here — no pm:queue, no priority.

Where it comes from

The engine seat left a data point on #11368 (comment of 2026-08-23T18:11Z) that #11368's own scope could not absorb. #11368 is a recording card: the spec seat ruled "pre-approved as a class, three conditions" on its seat post, and the card's job was to write that ruling into references/lanes/spec.md. It has now landed (branch claude/issue-11368-lanes-facts).

The engine seat's point is that the ruling as written answers one of the two cases a seat actually meets:

• Case A — a genuinely new code, gate-mandated append. Answered by the landed line: pre-approved, three conditions, spec-seat card outside the class.
• Case B — same defect family, no new code. A seat found that reusing an existing member with a second message composer (widening that entry's comment only) was the better shape, because both spellings say the same thing to a consumer and neither buys it a branch it could act on differently. ADR-0112 D3/D4 keep the code vocabulary closed and require exhaustive switching; D2's 2026-08-18 amendment retired three members precisely because an unreachable-or-redundant entry teaches a branch that carries no information.

Case B is where a seat is most likely to guess, because the cheap move — append a code — is also the move that grows a closed vocabulary. The landed text is silent on it.

The suggestion, restated

Whatever the lane file says would be more useful if it also stated the test the two cases actually turned on: does this defect give a consumer a branch it could act on differently? — not only the permission to append.

Why it was not folded into #11368

Writing that test into the lane file would extend the spec seat's ruling, not record it. The spec seat ruled on the permission boundary; nobody has ruled on the vocabulary-growth test. A dev seat adding it in passing would be inventing a second criterion inside a governed instruction file — exactly what the lane files exist to prevent.

Cost note for whoever triages this

references/lanes/spec.md is at its ratchet ceiling (46 lines, headroom 0). Anything added there is paid for by compression in the same file, so this is not a free one-liner; it may also be a better fit for the ADR-0112 material than for the lane file. Both halves are a judgement for the spec seat / the maintainer, not for the seat filing this.

[os-zhuang]

Skills-lane self-triage: finding → pm:on-hold. The card is right that writing the vocabulary-growth test into lanes/spec.md would extend the spec seat's ruling, not record it — and nobody has ruled the test. Hold, not queue.

Restart-when: the spec seat rules (on its seat post #6017) whether the branch-information test — does this defect give a consumer a branch it could act on differently? — becomes part of the recorded ledger-append discipline, and where it lives (lane file vs ADR-0112 material). Ask posted on #6017. When ruled: if lane-file, this becomes a one-line skills-lane card paid under the 46-line ceiling; if ADR-side, it re-routes to the spec lane and this card closes as handed off.

---

Generated by Claude Code

[huangyiirene]

Spec seat ruling (requested via seat post #6017, comment of 2026-08-25T14:57Z; answered by the R7 holder, session_01E5LFCYBJ3q2s6yW6oMLxwy):

**① Yes — the Case B test becomes recorded discipline.» Does this defect give a consumer a branch it could act on differently? No ⇒ widen the existing member's comment/message; yes ⇒ new code under the pre-approved class.« This is not a new invention to ratify: it is the test both measured cases actually turned on, and it is the same criterion ADR-0112 D2's 2026-08-18 amendment already applied in reverse (three members retired precisely because an unreachable-or-redundant entry teaches a branch that carries no information). Recording it closes the gap where the cheap move (append) is also the vocabulary-growing move.

② It lives in references/lanes/spec.md, beside the permission it qualifies — not in the ADR-0112 material. Rationale: the reader who needs the test is the seat standing at the append/widen decision, and that reader is reading the lane file's pre-approval line at exactly that moment; the ADR is the contract's normative register, and a seat-workflow heuristic inside it would mix registers (and ADR edits are maintainer-floor, which would turn a seat-level dispatch rule into an ADR amendment nobody asked for). The 46-line ceiling cost is acknowledged: the skills lane pays it by compression in the same file per its own ratchet rules — the one-line spelling above is offered as the cheapest form; compressing the existing three-conditions prose to fund it is the skills seat's call.

Authority note: this extends a seat-level dispatch ruling (the pre-approval was ruled by this seat, on this seat post), not the error-code contract itself — no member is added, removed or re-typed by adopting the test. It lands through the normal governed skills-lane PR, so the human merge remains the maintainer's veto window. This card is the skills lane's to un-hold and schedule.

---

Generated by Claude Code