Performance patch over v1.11 (package 1.11.1, voice format still v7; no API or protocol changes).
- Sine operators use a deterministic polynomial
fastSin(|error| ≤ 2e-10) instead ofMath.sin, and all-sine voices take a faster render path. - Node 26
npm run benchmark(single machine, not browser/device): standard 8 voices 0.265 → 0.215 ms/block (−19%), standard 32 voices 1.132 → 0.900 (−20%), high 32 voices 1.977 → 1.515 (−23%). - Sine output differs from v1.11 by about 1e-10 (not bit-identical).
- CI (Node 22/24/26, Chromium/Firefox/WebKit) passed; scoped static security review found no evidence-backed issue.
- Not published to npm. Physical-device, MIDI-hardware and human-listening acceptance remain unverified.
Tarball SHA-256: 4f5ccfcc499c5100645bba97caf9a80d6d2a6e5229c404fc11006ee1a785d58e