Study 16 · Listening
The chart reads barmark: the adoption validated end to end, the migration grid-perfect, and a wrong pair finally defeats the ratio clause
Listening, the adoption
Study Overview
The Adoption, and the Registration
Stem Shovel retired its in-house chord-chart parser and adopted the open barmark standard end to end: barmark 0.5.0 from npm parses every chart behind a production adapter, the renderer draws from the same parse, and sections are known by name rather than numeral (the app's pins 371831ae, b5ed9ed2 and a0926c65). All five stored Badverbs charts were migrated under a refuse-not-guess rule and confirmed by the band against its recordings' section markers; the four public ones are the corpus, fetched verbatim, with the fifth not on a public page and disclosed as out of scope. Registered at b589daf with one amendment, recorded before results were accepted, for handling a refusal in the exploratory cross-pairs. Zero spend, 95 mechanically graded cells, no model calls, and the alignment ports unchanged from Study 15's pin.
The Parsers Agree Bar for Bar
G1 passed with nothing to report, which is the point. The three committed app-dialect texts ran through the retired chordChartBars parser and through barmark 0.5.0 behind the production adapter, identical bodies with each parser reading its own registered dialect header: 76, 152 and 86 bars, zero differences under the registered per-bar comparison, zero warnings from either parser. Nothing the app could read was lost in the adoption.
The Migration Is Grid-Perfect
G2 passed at a precision the gate did not ask for. The gate allowed each band-declared section start within a tenth of a second on the chart's own grid; every public migrated chart lands every declared start at 0.0000 seconds with zero parser warnings. letter-lands parses 76 bars with 10 sections, peaceful-dreams 153 with 10 (the verified 152 plus the grammar's ring-out attack bar), eat-all-the-clocks 71 with 9 in the band's corrected play-order form, and exploratory mop-heart 140 with 8, in 3/4 at 224 bpm with a mid-chart key change. letter-lands and peaceful-dreams also match the verified content lists exactly under the registered normalization.
A Real Wrong Pair Finally Crosses the Ratio Gate
G3 passed, and it carries the study's headline. The right pairs pass both same-piece clauses comfortably: ratios 1.90, 3.18 and 2.63 with median local costs of 0.177 to 0.210, plus exploratory mop-heart's first-ever alignment at 1.96 and 0.248. All six registered wrong pairs fail, ratios 1.13 to 1.57, caught by local costs of 0.333 to 0.475. Then the exploratory cross-pairs delivered what three studies of eroding margin predicted: the eat-all-the-clocks recording against the mop-heart chart scores ratio 1.74, the first real wrong pair above the 1.6 gate, and is rejected only by the second clause, its median local cost of 0.323 sitting over the 0.30 line. Production's shipped samePiece is still ratio-only, so the bench's standing recommendation to AND the local-cost clause in is now urgent rather than prophylactic: a real catalog pair crosses the ratio gate. One other exploratory pair never produced a number at all, and that is also a result: production's own rule that a chart running more than twice the recording's length must be refused turned the mop-heart recording against the peaceful-dreams chart away, the right answer by construction.
No Regression, the Chart Was the Bottleneck, and the Verdict
G4 passed against its named baseline: through the entirely new parse path, letter-lands' median section error is 0.05 seconds against Study 15's 0.08, and peaceful-dreams holds at 0.11. The predictions went four for four, and the second is the satisfying one: eat-all-the-clocks, exploratory in every prior study because its chart was wrong, drops from a 1.99 second median error to 0.09 with the band's corrected chart and is now gate-grade on every measure, confirming Study 12's verdict that the chart, not the aligner, was the bottleneck, this time by the fix rather than the failure. The verdict delivered: adoption validated, ship with confidence. Two asks stand: the LAST-SOUND ring-out refinement from Study 15, still visible in peaceful-dreams' tail sections, and the medianLocal clause in production's samePiece, promoted from recommendation to urgent ask.