The Same 206 Seconds Measured Twice

The folder is the same folder and the project is the same project. Two masters sit in 03_Finals/v1.0 and both declare a duration of 206.0 seconds. The first is sal_y_miel_final_nolan.0V1.0.mp4 at 128,100,856 bytes. The second is sal_y_miel_nolan_complete.0V1.0.mp4 at 60,714,422 bytes, which is 52.6 per cent of the size of the first. The reports written beside them carry the threshold that produced the cuts, 0.3, and the moment they were generated, 2026-09-11 at 15:50:07 and at 15:50:14. Seven seconds apart. Two documents, 13,819 bytes and 11,657 bytes. The detecting pass behind them was identical. The films it read were not.

The first report measures 34 segments across 206.0 seconds. The second measures 28. Same declared duration, same 0.3 threshold, and 6 fewer segments in the second file. Every segment in both carries 5 tags drawn from a fixed list and 1 sampled value, and that value reads 3 in every segment of both reports, so no cut in either file is a single-frame guess. Both reports count 0 segments flagged dead. Both run from 0.0 to 206.0 with no gap between one segment's end and the next segment's start. What differs is not coverage. It is where the cuts land.

The second file grades 30.36 points warmer on average. The first report's warmth averages minus 9.80, running from minus 37.40 up to plus 38.50. The second averages plus 20.56, running from minus 8.20 up to plus 65.90. Luma averages 43.456 against 40.193, a gap of 3.263. Contrast averages 44.879 against 39.082, a gap of 5.797. The floor of the luma range is identical in both reports at 22.5. The ceiling is not: 64.9 against 62.9. Two passes over the same clip, read by the same instrument, and one comes back cool and lowkey while the other comes back warm.

Both reports carry a vocabulary of exactly 15 words. Thirteen of those words appear in both: action, amber, dark, dim, drifting, fast, golden, lowkey, moving, neutral, night, slow and warm. Two words appear only in the first report, blue and cool. Two appear only in the second, flat and soft. The union is 17 words. The first report spends its top 3 evenly, on dim, lowkey and cool at 20 uses each. The second report's most used word is warm, at 22, with fast and action next at 18 each. The same instrument, the same word list, and a different word on top.

Two masters. One folder. 206.0 seconds each.

Raw motion averages 27.835 in the first report and 31.964 in the second, so the second file moves more at the pixel level. Normalised motion averages 9.466 in the first and 6.950 in the second, so the second file scores lower on the same measure. The 2 readings point in opposite directions. Segment lengths explain part of it. The first report's segments average 6.0585 seconds with a median of 5.000, and its longest is 10.040. The second averages 7.3579 with a median of 8.085. Longer cut, less motion inside each one. The segment durations sum to 205.990 in the first report and 206.020 in the second. Neither sum is the 206.0 the header declares.

The probability that 2 masters would be cut, graded and filed under one declared duration without anyone ever writing down what separated them is not a small one, and most rooms price it at zero right up to the moment a client asks which version was approved. The record here is 2 JSON files, 13,819 bytes and 11,657 bytes, readable a year later without opening either movie. On the 210 energy systems deployed across Africa and Asia the rule was the same: the unit that ships carries the specification of the unit that was tested, because a rebuild from record and a rebuild from memory are 2 different machines. Two documents, 2 films, and the difference sits on the page instead of in a room. Keep more of yourself, for longer.