Release Audio Desync Finder 3.0.70 · Jank8/Audio-Desync-Finder

Audio Desync Finder 3.0.70 / 2.0.70

Late runs are rescued. A dub re-cut late in a film shows up as a tight band of points a second or two away from the model. The fit used to reject them as outliers, the model kept drawing the old line to the end and the export drifted for the rest of the film. A run of 12+ unexplained points over 240 s+ with a spread under 50 ms now becomes a segment of its own, and the covering segment is split around it.
Min confidence defaults to 4 (was 2).
24 vs 23.976 is a real tempo — 0.1 %, six seconds across a film. It used to be ignored because the duration difference was below the 0.5 % gate: no hint, no pre-warp, a drift fitted from five points. The framerates now decide, and a segment resting on fewer than 20 points takes the known ratio.
Framerates compare with a 0.01 fps tolerance, so two releases both reporting 23.976 (23.976000 vs 23.976024) are the same rate and the Tempo box is respected.
The "only N survived" warning warns only when the model rests on fewer than ten measurements.
The window opens 1500 wide and centred in both builds.

3.0.62 - points are judged against the file's trend (Theil-Sen), not against their neighbours
3.0.63 - the trend is the model: short groups inherit it, breakpoints on a five-point running median
3.0.64 - outlier warnings show the range, not the first point
3.0.65 - one drift rule for the model and the mux hint (40 ms across the film)
3.0.66 - Qt reports the end of a scan
3.0.67 - the status line ends with "Measuring done"
3.0.68 - the mux hint names the real reason it refuses a delay
3.0.69 - tempo is fitted only when the framerates differ, with a Tempo checkbox to force it
3.0.70 - the rescue pass, Min confidence 4, the 24/23.976 ratio, the framerate tolerance, the 1500 px window

Full details in CHANGELOG.md.