Commit History
-
Update benchmark documentation and point estimates
Verified signature4f89269
-
Add complete JAAH Maj/min chord evaluation and confidence intervals
Evaluate all 113 JAAH recordings: SheetSage2-AR 64.50%, ChordFormer 59.45%, and Prober 62.94% Maj/min recall. Document reference-span scoring, coverage, per-track scores, fixed-model provenance, and paired BCa intervals. Preserve previous benchmarks and the Chords1217 five-fold explanation.2c45f22
