Fradan is perceptual-codec forensic analysis: four pages of charts showing where the artifacts sit, across time and across frequency. No scores for anything.

In theory you can detect lossy samples or even AI audio artifacts.

The whole analysis compiles to wasm and draws the same four pages onto canvases here. No bindgen, no npm, no bundler. The interface is four pixel buffers and a string, so a raw C ABI covers it. Decoding stays in Rust rather than going through WebAudio, because that resamples to the AudioContext rate and would destroy the lowpass cutoff that page one exists to find.

Nothing is uploaded. All is local, and in theory can be used offline. The file is decoded and analysed inside this tab, on the main thread. So a long file will freeze it for a few seconds.