Hearably web-preview limiter: standalone reproduction kit
https://hearably.app/research/browser-audio-limiter-benchmark/

Measurement published: 2026-08-28. Downloadable kit added: 2026-09-20.

1. Save run-benchmark.mjs, limiter-worklet.mjs and results.json in one folder.
2. Inspect the source. Run: node run-benchmark.mjs
3. Read the new limiter-results-.../results.json and results.csv files.

Node.js is required (the original report used v22.16.0). No third-party
packages, account, repository access or network calls are required by the
runner. Use an empty working folder; each run creates its own result folder
and does not overwrite the downloaded reference report.

The worklet snapshot is byte-identical to the measured Studio/web-preview
source. It is frozen here so later product changes cannot silently change
the August measurement. The runner checks this SHA-256 before importing it:
efa1af972f93d864c363de66dc16b3ce876bcd459c60d97183991fcd7e9da5ce

Four deterministic stereo signals are rendered at 48 kHz in blocks of up
to 128 samples: a 0.5 impulse, a 2.0 impulse, a 2.0/0.5 linked stereo
impulse and an overdriven 1 kHz tone. The last block may be shorter.
The original report's pass flag checks only the 0.95 sample-peak ceiling
plus 0.000001 tolerance. The reference JSON and CSV have not been rewritten.

The September runner adds separate reproductionChecks for finite output,
239-sample delay, unity below the ceiling and a 4:1 stereo ratio. It also
compares fixed numeric metrics to the August report within 1e-12. Timestamp,
Node/platform/architecture and machine-dependent timing are not compared.
Success exits 0. A failed check or checksum mismatch exits 1. Current output
uses schemaVersion 2 to distinguish these extra checks from the original.

The Node harness mocks AudioWorklet registration and calls process(). It
does not run in a browser and does not capture extension or website audio.
It does not verify DRM compatibility, true peaks, LUFS normalization,
perceived quality, hearing safety, battery usage or browser CPU performance.
The two-second noise throughput diagnostic is local and machine-dependent.
No competitor comparison is performed.