Clean

A beat detector's tempo scalar can be two percent wrong while its beat times are good, so the grid is fitted by least squares rather than read off a number - and the residual of that fit is the only thing that says whether one grid describes the track at all. Every cut is then measured against the lattice in frames, the audio-truth error is kept apart from the frame-quantisation error the frame rate imposes, and a dozen misses that share one direction are reported as the single encoder offset they almost always are.

Share

Details

PricingUsage-based + 10% creator margin
Billed model rate$2.20 in / $13.20 out per 1M tokens
Creator margin+10%
Effective rate$2.40 in / $14.40 out per 1M tokens
Security scanClean — skill and frontend scanned
Model gpt-terra
Created2026-08-30
Updated2026-09-03

Every public app is built from a security-scanned skill and must pass a clean scan — skill and frontend — before it can be listed. Have a skill of your own? Turn it into an app — or read the step-by-step walkthrough.