Provir — reading the quantisation grid a lossy codec leaves in audio sold as lossless.

A lossy audio codec divides audio into blocks, transforms each block and rounds the transform coefficients onto a grid. Decoding to PCM does not remove that structure, and writing the result into a lossless container does not either. If the codec's own transform is recomputed on the decoded audio at the codec's own block alignment, the coefficients still sit on, or measurably close to, the grid the encoder used; at any other alignment, and in audio that was never lossy-coded, they do not beyond what a null of lawful audio predicts, save an open class, a signal that repeats exactly. This paper describes a detector that turns that fact into a provenance decision about a file presented as lossless: a sweep of the codec's transform over every alignment at which its grid can sit, statistics of how closely the coefficients lie on a grid at each alignment, a reduction of each profile to a peak and a ratio against the file's own off-grid floor, thresholds fixed before they are priced and measured afterwards on lawful material with a stated bound, and a rule under which a file is convicted when two independent witnesses of the same grid agree or, in a further embodiment, when one statistic of the grid clears a threshold derived from a null of lawful audio; a statistic not so priced does not decide, and the absence of a grid is never reported as proof that a file is lossless. It reads MPEG-1 Layer III at bit rates where the encoder leaves no spectral shelf, reads AAC after Derrien (2019), and reads Vorbis by the discreteness of the second difference of its log-magnitude spectrum. A reduced read visits only the alignments where a grid can sit, and a compact record carries the grid statistics without the audio. Every lawful price is given with the population it was measured on and its confidence bound, and the paper states what the method does not claim.