Chord Seq

The host resolves the packaged model bytes, owns transport time, and advances the MIDI processor on the block clock. The face requests a progression; it never creates its own AudioContext or advances playback from a UI timer.

Generation produces chord tokens, then the pure voicing stage places those pitch classes near the selected octave with deterministic movement between chords. The same resulting note sequence feeds live MIDI output and file export.

If the model runtime is still loading or unavailable, the face labels and uses its deterministic I-vi-IV-V fallback. Native formats remain unclaimed until they have a compiled control kernel and native inference path.