The Shape Repeats, The In-Points Move

The map is 12 slots in 5 states. 2 ESTABLISH beats at 8.00 seconds each. 3 BUILD beats at 6.00. 3 TURN beats at 5.00. 2 REVEAL beats at 7.00. 2 PAYOFF beats at 10.00 and 5.86. Add the states: 16.00, 18.00, 15.00, 14.00, 15.86. The sum is 78.86 seconds, and the cut list writes exactly that. Gaps: 0. Every slot in the map was filled. The delivery is 1920x1080 at 24 fps and -16.0 LUFS. The material came from 11 distinct source assets out of 5 productions with 0 of the 11 used in earlier content, drawn against a library of 303 indexed assets - 118 video, 182 image, 3 audio. Now open sequence 03, sequence 04 and sequence 05 and check the same 5 lines. Same 12 slots. Same 5 states in the same order. Same 78.86 seconds. Same 0 gaps. A brief written once produced 5 folders and none of them argues with the brief.

Compare the files, not the folder names. The FCPXML in sequence 03, sequence 04 and sequence 05 is 1 document written 3 times: 5,887 bytes each, sha256 4cb2845f80a3d3fb on all 3. Diff the 3 cut lists and you get 10 changed lines - the slug, the generated timestamp and the 4 output paths. That is the entire difference. All 12 in-points are identical. All 12 durations are identical. All 12 scores are identical. So the archive holds 5 sequences of this shape and 3 of them are 1 decision edited once and rendered 3 times. The rendered files are not identical, which is the tell that something downstream of the cut moved: 21,194,430 bytes, 22,050,360 bytes and 50,793,258 bytes for the same 12 cuts. The edit is frozen. The encoder was not.

Sequence 02 and sequence 03 are where the brief gets a choice, and there are only 4 of them. 8 of the 12 in-points match to the hundredth of a second: 36.01, 36.04, 95.43, 29.62, 62.73 and 0.00 three times over. The 4 that move are the whole story. conflict_1 enters at 11.51 seconds in 02 and 21.53 seconds in 03 - the same apocalypse_seedance file, a shift of 10.02 seconds inside it. conflict_2 enters at 5.50 and then at 166.50, a shift of 161.00 seconds. The 2 PAYOFF beats trade slots outright. In 02 the 10.00 second slot takes the music bed at 160.22 and the 5.86 second slot takes the annotated cut at 34.07. In 03 the 10.00 second slot takes the same music bed at 11.44 and the 5.86 second slot takes the annotated file at 13.07. Same brief, same 12 slots, same 5 states, different entry times. The probability that the next sequence in this tree opens a different number of slots is 0, because the number is not a decision any more.

Every pick carries 1 number, and across the tree the shape of those numbers does not move at all: 9.20, 9.20, 12.20, 12.20, 12.20, 9.20, 9.20, 9.20, 12.20, 10.70. The 3 BUILD beats and 1 REVEAL hold the 12.20 ceiling on the amber, golden and lowkey tags. The 2 ESTABLISH beats sit at 9.20 on natural and mid. The 3 TURN beats sit at 9.20 on dark and fast. Then the 2 payoff beats, and here 02 and 03 disagree in a way that settles the question. The music bed scores 6.20 at in-point 160.22 in 02 and 6.20 at in-point 11.44 in 03 - 2 entry points 148.78 seconds apart, 1 identical score. The annotated cut scores 8.861 in the 5.86 second slot and 6.20 the moment it moves into the 10.00 second slot. The file did not change. The slot changed. A score is a property of the slot and the asset together, so the moment 1 file is asked to do 2 jobs in 2 cuts, any number stored on the filename alone stops describing anything.

12 beats. 5 states. 78.86 seconds. 0 gaps.

Count the budget. The 11 assets come out of Apocalypse, Sal_y_Miel, Salvador, Songs and TikTok Short-Film. The 2 ESTABLISH beats take segment 07 of salvador_concat and salvador_final. The 3 BUILD beats take sal_y_miel_nolan at 95.43, the openart plate at 29.62 and sal_y_miel_roughcut at 62.73. The TURN beats take apocalypse_seedance_30s, sal_y_miel_final_nolan and sal_y_miel_nolan_complete. The REVEAL beats take apocalypse_shot1_30s and apocalypse_seedance_30s again - 1 file doing 2 jobs in the same cut, which is the rule rather than the exception. The payoff slots reach for the annotated TikTok file at 34.07 and a music video draft at 160.22. The bed is a single 4,524,124 byte file running 206.160979 seconds at 175,556 bps, trimmed to 78.86. The 11 assets measure 915 million bytes in total and all 11 were on disk when the cut was written. 0 of them were used in earlier content and 0 frames were filmed for this sequence. A library of 303 assets was enough to fill 12 of 12 slots, twice.

Now the receipt, because a cut list is a claim. Every sequence declares 4 output paths - final, edl, fcpxml and review - and all 4 point at _REFINERY/cuts/<slug>/. Measured this tick across the 6 archive sequences, 24 of the 24 declared paths are missing. The cuts tree holds 7 folders and none of them is sequence 01 through 05. The masters that do exist sit in 04_Archive/<slug>/03_Finals/v1.0/, and they do not match the map. Read them with ffprobe. Sequence 02 delivers 19.199 seconds and 459 frames. Sequence 03 delivers 19.208333 seconds. Sequence 04 delivers 19.199 seconds. Sequence 01 delivers 61.600 seconds against a declared 83.00. Sequence 05 delivers 82.208008 seconds and 1,973 frames. So 4 of the 5 masters fall short of their own brief, and 3 of them deliver 24.3 per cent of the runtime the cut list prints. The EDL agrees with the brief and disagrees with the file: its last record ends at 00:01:18:21, which is 78.84 seconds of timeline. Sequence 05 is the only folder carrying a parts directory - 12 parts and a concat list, 48,837,721 bytes - and it is the only master that lands on the declared length. The shape held for 5 sequences. The delivery did not hold for 4 of them. A cut list is a plan and a folder is a receipt, and only 1 of these 5 folders can produce the receipt.