Music.
Ship the composition, not the recording.
The deepest audio lever. Spatial audio carried recorded sound objects; a
wai.audio.score carries no recording at all — just the notes. A few
hundred bytes of note events become music your browser synthesizes with the same
deterministic Rust engine (compiled to WASM) that made the conformance corpus, and checks the
BLAKE3 of the rendered PCM against the corpus's: mixdown-equivalence, the same criterion
spatial audio uses. The reference synth is trig-free — a phase accumulator and a linear
envelope in the fixed-point floor — so every machine renders bit-identical samples. A sink
with a soundfont or a neural instrument plays the same score richly; that is presentation.
Press play.
Composition as instructions
A note is a few bytes — pitch, timing, envelope. The whole piece is the note list; the recording is megabytes. The sink synthesizes from the score, so a minute of music ships as kilobytes. Lever 3, applied to sound.
Mixdown-equivalence
The reference render is bit-exact: the wai.det.fixed64 floor, a phase-accumulator
oscillator, a linear ADSR — trig-free, so + − × ÷ alone reproduce it on
any machine. The hash in the HUD equals the corpus's expected_hashes.json.
Pitch is a per-sample phase increment authored once — the floor never needs an exponential.
One floor, every medium
The same fixed-point floor and mixdown discipline as spatial audio — recorded objects there, synthesized notes here. Sight, hearing, touch, motion, state, and now the composition itself: one standard, one verification, on any machine.