Study 11c · Listening
The declared-meter gate validated, and the bench's own rule retired on the record
Listening, the ladder thread closed for good
Study Overview
The Patch, the Rejection, and the Registered Defeat
Study 11b named two fixes and Stem Shovel answered with a patch and a disagreement: the 3:2 and times-3 rung gates moved to the song's DECLARED time signature, off the detector's flag entirely; a times-3 rung joined the ladder for the fast waltz; and the bench's proposed alternative, evaluating the compound flag at the declared tempo's lag, was rejected with a constructed counterexample and a request that the bench test both rules on a mirror fixture built to defeat its own. The bench registered the prediction of its own rule's defeat before running anything. Verbatim port pinned at the patch commit; arms 1 and 3 are pure functions over Study 11b's stored reads; arm 2 renders three new mirror items, straight running eighths declared wrongly at two-thirds. Zero spend.
The Gate That Strictly Dominates
C1 passed perfectly: over the full 11b corpus, the declared-meter gate accepted 0 of 332 corrected wrong rows, lost no true-count match, and GAINED three over the shipped v0.200.0 rule. The two compound items the acoustic margin had missed now match at two-thirds through their declared 6/8, and the fast waltz matches at times-3 through its declared 3/4: 224, counted three to the bar. The one item that must stay an error stays loud, now by data instead of by luck. C3 passed as cleanly: the new rung opens nowhere but declared-triple meters, zero leaks, zero off-tolerance accepts, with the count ambiguity quantified and living where it belongs, in the rung picker.
The Mirror Did Its Job Twice Over
C2 is the study's heart. The bench's declared-lag rule accepted 3 of 3 wrong mirror values, exactly as the registered prediction traced: on straight running eighths, a two-thirds-wrong declared tempo puts its lag's third on the eighth grid, and every clause of the bench's amended rule licenses the rung, beat floor included. The bench's own proposed rule is retired on the record, by the measurement its proposer registered. The letter of the gate failed on the other half (the declared-meter side accepted 1 of 3), and the anatomy deserves plain statement: on the 138 bpm mirror the DETECTOR itself misread the straight eighths at 92 with a false compound flag, and the wrong declared 92 then matched the wrong detection directly at rung 1, which no gate governs. A ladder cannot reject declared-equals-detected; this is Study 11's lesson-six collision class again, not a gate leak, and the gates themselves never opened.
The Bonus Fixture
The mirror incidentally built the first SYNTHETIC reproduction of the detection-side confound: truth 138, detected 92 with a false compound flag, the true tempo sitting on the alternates at strength 0.96. The confound that previously existed only on one real recording now has a committed regression fixture, and the patched comparison surface already contains it correctly: the declared-meter gate refuses to call 138-against-92 a count, so the warning stays loud.
Verdict, Scorecard, and the Thread's Ledger
The verdict delivered: ship the patch to production as is. The declared-meter gate strictly dominates both its predecessors on everything a gate controls, and the one C2 accept is detection-side and pre-existing, with mirror-138 as its new regression fixture. The scorecard: prediction 1 right in full, 2 half (the bench's defeat traced exactly, the collision-class accept unanticipated), 3 right in full. The ladder thread closes at five studies, Study 10, its amendment, 11, 11b and 11c, for about eight cents of total model spend, all of it Study 9's pilot, and with each side having corrected the other at least once on the record.