CAP-33 does not own the engine's OTA arm, so a correct trailer fails the relevance lint (09048b84)

[dc0sk]

Follow-up from #1371 / PR #1401. This is the one remaining failure in the relevance corpus that is not a mislabel.

The case

09048b84 — "fix(daemon): let an OTA-enabled daemon receive uncoded non-ladder traffic" — carries Refactors: CAP-33 CAP-38. Its body describes the change as "without feeding the rate controller or keying an ACK", and it names on_rx_frame(RxOutcome::Failed) and ota_rate_decision_events. CAP-33 "Over-the-air receiver-led rate control" is semantically the right capability.

It fails the lint because CAP-33's code: is:

crates/openpulse-core/src/ota_rate.rs
crates/openpulse-core/src/profile.rs

— and the commit touched the OTA arm in openpulse-modem/src/engine.rs and openpulse-daemon/src/server.rs. CAP-33 owns neither. The rate controller's policy is in core; the OTA arm that drives it lives in the engine and the daemon, and the map does not say so.

Why it was not fixed in #1401

It is a map gap, not a rule error, and the 2026-09-15 maintainer decision already names curation as the remedy for exactly this. It briefly caused a worse mistake: I proposed relaxing the lint's predicate (fail only when every named capability is irrelevant) to absorb this one case. Review rejected that — it would have been a silent second relaxation on top of the gap, with a measured bypass (CAP-66 owns a touched file in 49 % of production commits). The predicate stayed strict and this case stayed failing, deliberately.

Enforcement is not retroactive, so 09048b84 blocks nothing; a future commit to the OTA arm naming CAP-33 would be refused, which is the live cost.

The decision this needs

Widening CAP-33 to engine.rs is not free:

• engine.rs is 1 335 mutants and already has 6 owners; server.rs is 218 with 6.
• code: feeds trace.py scope → req-mutation.sh, so a capability's size is its mutation cost.
• It also makes engine.rs a 7-owner hub, and hub ownership is what makes the relevance rule weak where commits concentrate (already 9 of 22 pass only weakly).

So the options are roughly:

1. Widen CAP-33 to the OTA arm files. Fixes the case; grows a hub and a mutation scope.
2. Narrow the commit's trailer convention — accept that a change spanning policy and its driver names the capability that owns the files it touched (CAP-38 here), and let CAP-33 describe policy only.
3. Function-level code: (-F-style scoping) so a capability can own the OTA arm functions in engine.rs without owning all 1 335 mutants. Biggest change; also the one that would help scripts/req-mutation.sh is the only tier-2 verification instrument and runs in no workflow #1279's cost problem.

Option 3 is the one that generalises, and it is the same lever the #1279 cadence work wants. Worth deciding together rather than separately.

Related: #1371, #1279, #1401.

🤖 Generated with Claude Code

https://claude.ai/code/session_0188ATCj6DZ9aRVQ2vSirua6

[dc0sk]

Closed by #1416 (8b5896fa) — no map change, per the maintainer decision of 2026-09-19.

Refactors: is structural (#1402, from check-trailer.sh's own header), so it names the capability whose code was changed. engine.rs is CAP-38's; CAP-33 keeps ota_rate.rs + profile.rs — the policy, not its driver. 09048b84's Refactors: CAP-33 should have been Refactors: CAP-38, and is recorded in the ledger alongside #1410's nine.

The cost figure in this issue was wrong: widening CAP-33 would have added zero enforced mutants, not 1335 — CAP-33 satisfies only REQ-FUN-06 and CAP-38 only four others, all baseline, which --all-enforced never touches. Correcting it did not change the answer; the real cost is a seventh owner on engine.rs, weakening the relevance rule where commits concentrate.

Also recorded: 09048b84 carries three Refactors: lines because a squash concatenates its commits' messages, so a PR body left as GitHub's default squash text inherits every commit's trailer.