Never ship a fix that made it worse.
Acuity scores your broadcast audio and video, cleans what's broken, then re-scores — and keeps the fix only when it measurably beat the original. Every enhancement arrives with a receipt.
Score → clean → re-score. Ship the win, discard the regression.
Every "enhanced" clip you've ever shipped was a guess. Acuity turns the guess into a receipt: score → clean → re-score, and if the fix didn't measurably beat the original, it's thrown away before anyone hears it. You only ever ship the win.
A closed loop, not a filter.
Acuity scores the input (NISQA MOS + noise, coloration, discontinuity, loudness), runs a DeepFilterNet denoise + loudness pass, then re-scores the output — and adopts it only when risk actually dropped. A worse asset is never stored, never returned as the winner. The same call reads video health too — black frames, freezes, A/V presence — so you're judging the whole broadcast, not just the mic.
One score, spent across all of WAVE.
"Good" audio skips the enhance pass and drops to an economy transcription tier; "bad" routes to premium plus review. Acuity is the quality gate the whole pipeline meters cost against.
Federated — auth, entitlement, and per-minute metering run through api.wave.online, so scoring the broadcast and acting on the score are the same governed motion.