Two scripts, four complete films
Both numbered plays now have complete films and fresh repeats through a frozen route, with delivery, sampled visual review and listening kept distinct.
There are now four complete films where earlier reports had plans, failed preparation windows and unfinished recordings. Two scripts have been performed in the permanent Blender theatre, then both have been made again through the recorded frozen method.
That was the practical test I had asked for. One completed film could still conceal repairs particular to its script. A second script would require different action. A fresh repeat with the software held unchanged would show whether the factory could run the corrected method again, rather than merely retain one successful assembly.
The four files exist. Each lasts 180 seconds and contains 4,320 pictures at 24 frames per second, with stereo sound. The recorded visual reviews used frame sampling; the sound records explicitly say no listening was performed.
Two different plays
SC-000002, Leave It Open, uses the office cart, opposite-wing entrances, conversation and two desk contacts. Its first completed version began with a hand-written prompt book. That book gave the new compiler a concrete performance to translate into actor controls, scenery cues, projection, music and camera choices.
SC-000001, Where Next?, adds the whole office, a trial board with two hinged covers, a stagehand operating them, and a supported bell. Nell and Leon prepare a science-news recording. The sequence must show the evidence changing as covers open and close, preserve their dialogue cards, and reach Leon’s bell contact and the final fade.
Both initial films included repairs on the way. The curtain-cache fault, movement combinations, camera operator, fog startup and interrupted work all remain in the record. Calling those first builds unattended would erase the very work that made the repeat worth attempting.
Claude Code then preserved both initial performances in history and began fresh runs, including planning, so the new books could make different staging and editing choices while preserving the plays’ required action and physical limits.
Repeat the method, not the old movie
The fresh Leave It Open ran from about 05:16 to 06:56 UTC on October 18. Its planner wrote a book in one round of approximately 136 seconds. The scene passed its seven physical obligations on the first full stage attempt, then reached seven-camera capture and the final cut.
Fresh Where Next? followed from about 06:56 to 09:22 UTC. Planning took one round of approximately 232 seconds. Its six physical obligations passed on the first full stage attempt, including both puppet fastenings, the bell touch, cart positions and curtain opening. It then recorded and assembled the complete play.
The worklog records no human action between each fresh command and its release. A seventeen-file list identifies the preserved software used for these repeats by digital fingerprints, which change if the file changes. Codex matched those fingerprints against the retained code, checked that each movie matched its delivery receipt, inspected its format and frame count, and read the complete video and audio streams to detect decoding errors.
The delivery receipt includes literal file properties:
"duration_s": 180.0,
"video": "960x540 h264 24/1",
"video_frames": 4320,
"audio": "aac 48000 Hz 2 ch"
Exact release excerpts, full film identities and verification scope. Retained source file.
The file fingerprints differ across all four films and match their respective release receipts. The repeats are new productions with retained footage and scene records, rather than four names for one export.
Completion still has particular limits
The repeat took about an hour and forty minutes for Leave It Open and two hours and twenty-six minutes for Where Next?. Cloth and fog dominated the machine work. In the latter run, the fog stage also included copying 17,284 cloth-cache files for the filmed scene’s changed name. Suggested speedups remain future maintenance; they were not quietly added between the successful repeats.
The visual review is also stated precisely. The worklog calls the films watched, then describes frame sampling across the script blocks: office arrival, speaker-owned screens, gestures, touches, trial-cover changes, bell and fade. That supplies sampled visual observations. It does not establish continuous normal-speed viewing with sound, and the listening records claim none.
The four movies nevertheless change the project’s result. The two different scripts and their fresh repeats have complete delivery files, retained physical checks and documented visual inspection. The earlier failures remain failures under their own rules. The later change allowing different planning agents is a separate amendment, whose tests must be assessed at their own scope.
For the first time, the source record can point to the complete plays and their repeated productions, rather than asking another set of process checks to stand in for them.
