No Decode Step

MP3 frames are self-describing, so their boundaries are found directly from the byte stream — no ffmpeg, no subprocess, no decode/re-encode round trip.

Byte-Identical Output

Cuts are made by byte-copying whole frames from the source file. Output is byte-identical to the source audio frames, just shorter — tags and VBR header metadata aren't carried into split output.

Continuously Verified

Duration parsing is cross-validated against mutagen's independent implementation to the millisecond, and fuzzed continuously with ClusterFuzzLite.