Files
product-vamp/docs/composer/musical_capability_matrix.md
T
khodowany 0e68f5364b Cleanup
2026-10-04 18:48:02 -05:00

2.6 KiB
Raw Blame History

Demonstrated capabilities and limits

Behavior Active implementation Evidence
New tune without a compulsory head rhythm Source-independent action construction Empty-head and held-head synthetic tests
Arpeggiated expansion / scalar route / neighbor motion Complete contextual pitch route Operation reachability and route tests
Development, answer, return, optional head reference Heard-theme/source feature used only by the selected action Each action requested and realized in eligible authored contexts
Open continuation / local settling / completion / handoff Different timing and destination contracts Candidate diversity and terminal-hold tests
Rest, pickup, syncopation, cross-bar group, tuplet Exact rhythm groups and complete authored cells Group, notation and playback tests
Early destination, approach, enclosure, connection Joint proposals and existing typed validation Positive/negative contract tests and final expression reports
Stable/resolving colors, prepared suspension, blues General chord/function policy; complete realized evidence Retained validator tests and construction coverage
All six strings; no open strings; frets 1–12 Shared pitch domain and complete-melody placement Physical tests and published TAB validation
No target dependency / no other generated-role dependency Input audit and independent role cache Dependency checks and full-song document hashes

Not every enabled treatment must be selected in every song. Rejections and selected uses are separate from capability. A failed bounded search is not a proof that a musical solution is impossible.

The source melody is optional material, not a compulsory blueprint. A return currently preserves the rhythm and direction of a short heard signature with harmonic adaptation; this is weaker than requiring its exact intervals. This can still be perceptually insufficient. Phrase-level context and the user's listening assessment remain important.

The route solver selects a complete path but internally uses a bounded pitch frontier. It does not perform a global whole-chorus optimum or arbitrary two-sided rewriting of committed phrases. One-phrase lookahead chooses a different complete alternative instead. Faster passages use a coarse tempo policy; legal TAB is not proof of human comfort.

Unsupported written chords still produce explicit mapper errors. No missing initial harmony is invented. Source-defined vocal protection remains authoritative even when a comparison reference continues beyond it. No new dynamics, swing, accents, velocities or timbral effects were introduced.