The Caption Ships With The Cut

The finished object is 1 file: sal_y_miel_social_teaser_02_final.v1.0.mp4, 2,138,653 bytes, h264 with aac audio, 1080x1920 at 30 fps. Probed this tick, it runs 6.100 seconds against a 6.0 second brief - 0.1 second of trailing frame, printed rather than rounded away. The brief is 6 beats in 6.0 seconds: a 1.2 second hook, three 0.6 second beats, a 1.0 second payoff and a 2.0 second release. The longest beat in the film is 2.0 seconds, the average is 1.0 second, and 4 of the 6 beats are shorter than that average. Those 6 beats were assembled from 6 rendered parts totalling 1,993,279 bytes and concatenated into the 1 delivered master, and the audio was trimmed to 0:6.000 off a 206.160978 second bed at -14.0 LUFS with a 0.7 second fade in and a 1.4 second fade out starting at 4.60 seconds. Every one of those numbers is in the folder. Not 1 of them is in the caption. That is the whole problem: the file is finished and the delivery is not, and the difference between those 2 states is where a cut sits for a week.

The package writes the same 4 lines to 3 places and the only thing that moves is the count. Instagram Reels gets 233 of 2,200 characters with 5 hashtags. TikTok gets 165 of 2,200 with 4. YouTube Shorts gets 220 of 5,000, a title that reads Everything here already existed, and 3 hashtags. The copy itself never changes: everything here already existed, cut from what was already shot, nothing new was filmed, keep more of yourself for longer. So the headroom is 1,967 characters on Reels, 2,035 on TikTok and 4,780 on Shorts, and 1 of the 3 formats is using 4.4% of what it is allowed. That looks like waste and it is not. A caption written to a ceiling gets padded to the ceiling, and padding is where a brand starts writing sentences nobody asked for. Writing to 10.6% of the limit on 1 platform and 7.5% on another is the same discipline as cutting a 6.0 second film out of 1,000 seconds of master: the length is a decision, and the limit is only the wall behind it. 3 platforms, 3 walls, 1 piece of copy, and 0 rewrites because 1 of them needed a different tone.

The publish file names 3 routes to the platforms and the count of working ones is 1. Composio MCP is not wired, with the reason written beside it: OAuth is required and the ak_ keys are rejected. The Postiz posting agent is not installed. Manual is ready now. So the file does not describe an automated pipeline and it does not pretend the missing pieces are coming - it prints 2 dead rails, 1 live rail and the exact next action for each: a browser login for 1, an install plus an account key for the other, and for the third, open the asset and paste the caption. That last line is the one that matters, because it is the route every published cut has actually taken. The software that does not exist yet has not shipped a frame. The route with a person in it has shipped all of them. A rail list is a promise you pay for later, and 2 of the 3 promises here are marked unpaid in the same file that makes them. That is the honest form: not a roadmap, an inventory.

6 distinct sources were cut into this film and 0 frames were filmed for it. The sources come from 3 projects, Apocalypse, Sal_y_Miel and Salvador, and 1 of the 6 had already been used in earlier content, which the cutlist records as a use count of 1 rather than hiding behind a fresh filename. The hook comes off a 30 second clip at 21.53 seconds in. The proof beat comes off a finished master at 47.80 seconds. The release beat comes off another finished master at 156.00 seconds. Five beats are video and 1 is a still, held for 1.0 second with the camera not moving. The sources do not agree with the canvas either: 2 of the 6 were shot 1280x720 and 4 were shot 1920x1080, and all 6 land inside a 1080x1920 vertical frame. The fill behind a mismatched aspect ratio is that same frame, scaled down, blurred and blown back up behind itself, so nothing is letterboxed and nothing is cropped away. That is the trade the whole engine rests on: instead of filming 6 seconds, you measure 303 library sources, pick 6, and pay for 1 recut.

6.0 seconds of picture. 3 captions. 4 lines of copy. 0 automatic posts.

The refinery holds 7 cut folders. 4 of them carry a publish package - a PUBLISH.md, a post.json and a final master in 03_Finals/v1.0 - and 3 do not. The 3 without are sal_y_miel_social_hook_01, tiktok_short_film_social_hook_01 and tiktok_short_film_social_hook_02. The last of those has a REVIEW.md and no package. tiktok_short_film_social_hook_01 has a cutlist.json, no REVIEW.md, no publish folder and no final file, so nothing in it can be posted by anybody, including the person who made it. That is the number worth printing: 4 of 7 cuts can leave the building and 3 cannot, and the 3 that cannot are not missing the film, they are missing the caption, the JSON and 1 finished file. A cut folder without a caption is a screenshot of a decision. It documents work that will never be seen, which is a worse outcome than work that was never started, because the cost has already been paid and the asset is already sitting on the shelf expiring.

The last line of the file reads: review before posting, nothing here goes out on its own. So run the receipt test on the package itself. Entrade built 200 wood-gasification machines across the UK and Europe, and 200 is a count someone can walk and check. The book published on 8 April 2026 runs 393 pages and can be verified in 1 minute by anybody holding it. A EUR 75 million industrial group was restructured and 210 energy systems were deployed, and both are figures a counterparty can recount from their own records. Now apply the same test to a caption. Either the asset exists, the 3 ceilings are written down and the character counts are printed next to them, or the package is a description of a delivery nobody has made. The probability that a cut folder with no publish file gets posted by whoever finds it 6 months later is not 1, and 3 of the 7 folders here are in exactly that state. Keep more of yourself, for longer. The package is where that gets decided, and it is decided before the first post, not after the last one.