Exact-size compression is a budget, not magic

“Make this video smaller than 24 MiB” is not equivalent to choosing a CRF. PyFFmpegCore converts the target to bytes, reserves container overhead and audio bitrate, derives the remaining video bitrate from probed duration, and rejects an impossible request before mutation when that budget falls below the declared quality floor.

usable bits = target bytes × 8 × (1 - container reserve)
video bitrate = usable bits / duration - audio bitrate

The maintained workflow uses two-pass encoding when requested, keeps pass files in a controlled temporary workspace, and cleans them according to the explicit temporary-file policy. After muxing, the result and receipt use the actual file size to report target_size_bytes and target_met; an estimate is never reported as final proof.

The dated fixture run reduced a 4,042,503-byte H.264/AAC input to 248,417 bytes against a 262,144-byte limit. That proves the implementation on one deterministic input and FFmpeg build, not that every quality target is feasible. Users still need to watch/listen to the output.