161d147767
A score, haptic signal, audio scene, splat4d clip or film declares its length in its contract, and rendering allocates in proportion: a score's mixdown allocated `duration` samples outright, a haptic channel `duration` × 2 bytes, a scene's stereo mix 20 bytes a sample, and the splat4d and film receipts a hash a frame. Nothing bounded the declaration, so a few bytes could ask a renderer for more memory than any machine has. SECURITY.md recorded it as a gap when the fuzzing PR found the reader-side reservations. Each extension's contract section (score, haptics, spatial-audio, volumetric, linear-film and its reel) now has a reader limit shared with SPEC §2, through a new always-on module `wai::limits`: - score and haptics: `duration × channels` at most 2^27 samples; - audio scene: `duration` at most 2^26, so the stereo floor is at most 2^27 samples; - splat4d `n_frames`, film `frames`, a reel's shots together: at most 2^20 frames. A reader may refuse past them (`duration_too_long`) when it reads the contract, before anything is rendered; a conforming encoder writes none. The film schedule's tick, `start_tick + f × ticks_per_frame`, wrapped in release builds and panicked in debug ones. A reader must now refuse a schedule whose last tick passes 2^64 - 1 (a film's, and each shot's of a reel); `Film::frame_tick` saturates and `Reel::frame_tick` uses checked arithmetic. tests/extension_duration_limits.rs rewrites a committed vector of each format to declare exactly its limit (it reads) and one past it (refused by name), and checks the film schedule ending at 2^64 - 1 (reads) and past it (refused). Falsified: with the readers as they were, all three tests fail. The world, audio, haptic, splat4d, score and film corpora verify unchanged; the 275 lib tests with the container features pass. CI runs the test beside container_reservations on every leg. SECURITY.md drops the gap and keeps the one this exposes: replaying a world to a far tick costs time in proportion (memory is bounded, time is not). Co-Authored-By: Claude Opus 5.5 <[email protected]>
6.3 KiB
WAI Extension: Score (Music as Instructions)
Status: Draft. Phase 6 — the deepest audio lever. Ship the composition, not the recording: a score of note events the sink synthesizes. Conformance is mixdown-equivalence, the music counterpart of spatial audio. Reference impl: the
scorefeature ofwai-rs; corpus:score-conformance/; live sink: wai.transaction.science/score. Keywords MUST, MUST NOT, SHOULD, MAY are RFC 2119/8174.
1. Scope and model
A wai.audio.score object carries no recorded audio — only note
events. Where wai.audio.scene mixes recorded sound objects, a score
is reconstructed from instructions: a deterministic reference synth
turns the notes into a canonical PCM mixdown, and a sink with a richer
instrument (soundfont, neural) renders the same score for presentation.
It is lever 3 (instructions-at-the-sink) for music — a minute of music
is kilobytes of notes, not megabytes of samples.
In scope: the container, the note model, the deterministic reference synthesis, the mixdown-equivalence conformance criterion, the receipt profile. Out of scope: a transport; the sink's instrument (soundfonts, neural synths — presentation, not conformance); MIDI-file import (an authoring concern).
Relationship to the core spec: mirrors spatial audio. The reconstruct
runs in the wai.det.fixed64 floor; media takes the value "audio";
wai.audio.score is authoritative.
2. Container (WSCR)
"WSCR" | u16 n_sections | section table (u8 kind, u32 off, u32 len) | sections
| kind | section | required |
|---|---|---|
0x01 |
contract (canonical JSON) | REQUIRED |
0x02 |
note events | REQUIRED |
2.1 Contract
{ "channels": 1, "duration": 34000, "numeric": "wai.det.fixed64",
"sample_rate": 22050 }
duration is the rendered mixdown length in samples.
Reader limit. A reader MUST NOT refuse a score for its length while duration × channels is at most 134 217 728 (2²⁷, the audio limit of SPEC §2), and MAY refuse one
past it (duration_too_long), when it reads the contract, before it renders anything.
A conforming encoder writes no score past it.
2.2 Notes
n_notes u32, then per note (65 bytes):
start u64 | dur u64 | phase_inc i64 Fx | gain i64 Fx | waveform u8 | attack u64 | decay u64 | sustain i64 Fx | release u64.
phase_inc is the per-sample phase advance freq / sample_rate,
authored off-line from the pitch — so the floor never evaluates an
exponential to turn a note into a frequency (the same authoring
discipline as the avatar quaternion). waveform selects a reference
timbre: 0 saw, 1 square, 2 triangle, 3 25%-pulse.
2.3 Authoring form (informative)
wai score build writes a WSCR from a JSON document that holds every
field as an integer, so the committed document is the definition and no
platform's floating-point pow can change a pitch:
{ "sample_rate": 22050, "channels": 1, "duration": 34000,
"numeric": "wai.det.fixed64",
"notes": [ { "start": 0, "dur": 3500, "phase_inc": 50960238,
"gain": 1717986918, "waveform": 0, "attack": 220,
"decay": 600, "sustain": 2791728742, "release": 1500 } ] }
phase_inc, gain and sustain are the raw Q32.32 bits. numeric may
be left out. The tool refuses a number with a fraction or an exponent, a
member it does not name, a member given twice (readers differ on which of
the two values they keep), more than one channel, a phase_inc of a whole
cycle or more, a gain or sustain above 1.0, a waveform above 3, and a
note whose release would end after duration, so the mixdown always
holds every note whole. score-conformance/notes/ holds each vector in
this form, and wai score build turns it back into the committed WSCR
byte for byte.
3. Reference synthesis (trig-free)
For each note, a phase accumulator advances by phase_inc each sample
(wrapping at 1.0); a piecewise-linear oscillator maps the phase to
[-1, 1] (no trig); a linear ADSR envelope (attack ramp → decay to
sustain → held until dur → release to zero over release) scales
it; the result times gain accumulates into the mixdown buffer. Each
output sample is quantized to i16 (round(v · 32767), half away from
zero, clamped) in integer space. The whole path is + − × ÷ in the
Fx floor — bit-identical on every machine.
4. Conformance — mixdown-equivalence
Given the same
wai.audio.score, a conforming sink MUST reconstruct the identical canonicali16PCM via the reference synth, hence the identical BLAKE3 — on every machine, no tolerance parameter.
mixdown_hash = BLAKE3("wai:score-mixdown\x01" || i16le PCM)
score_hash = BLAKE3("wai:score\x01" || contract || notes) // identity
A conformance vector also states pcm_sha256, the SHA-256 of the same
i16le PCM (the audio buffer of SPEC §7 Digests), so a verifier that
holds only SHA-256 checks the samples. wai score render prints both
and, with --pcm, writes the samples.
A sink MAY play a soundfont/neural instrument instead; that is presentation and does not affect conformance, exactly as the audio-scene HRTF render or the worlds renderer does not.
5. Receipt (JWP profile)
score_hash is the identity; the group's objects are the mix PCM blocks
(per a fixed sample span), Merkle-bound; one Ed25519 signature (verified as
SPEC §7 Signatures requires) over score_hash + root + exact work (= duration × notes) + carried
joules_micro; parent_receipt_hash chains — the worlds/audio receipt
shape reused. The figure's acquisition class rides beside it as an optional signed label
(energy-measurement §2.8): absent, the receipt keeps
its legacy bytes and the figure is unlabelled; present, it is inside the
signature. The reference meter labels every figure it measures
OnChipCounter, with its declared uncertainty.
Appendix
Mixdown-equivalence is spatial audio's criterion with the recorded
objects replaced by note events: both render a deterministic i16 mix
in the Fx floor and hash it. The score is the most extreme
instructions-at-the-sink case in audio — there is no recording anywhere
in the payload, only the composition. Sight, hearing, touch, motion,
state, captured-content provenance — and now the composition itself —
are one floor, one verification.