Two seconds worked. Three minutes did not

The two-second equipment movie worked. It recorded every required camera, combined picture and sound, decoded completely and passed the local delivery checks. The next full-length tests did not produce a film.

Maintenance had repaired genuine parts of the route. Native multiview rendered the selected physical cameras together. A comparable six-camera excerpt became substantially faster. The seventh-camera test then caught a numerical rounding issue at the cable carriage’s travel boundary, repaired without moving the camera.

The successful seven-camera excerpt retained fifty-six pictures, eight from each camera. Its recording took about thirty-four seconds. The owner and readiness records in this equipment fixture were explicitly simulated. It established the exercised route over that sample, rather than a complete supplied-script crew performance.

The short sample underestimated the take

The full test started with assembly and physics baking on a 180-second equipment clock. Preparation finished within its then-current ceiling. A warmed sample supplied a fixed recording allowance with the required margin, using the measured short render work to estimate the time for every camera exposure across the complete three-minute equipment clock.

Actual continuous capture ran more slowly. The attempt stopped at its unchanged deadline with 429 of 720 complete seven-camera frames, reaching 107 seconds of the intended movie. There was no finished recording to edit.

Root first investigated whether the probe had sampled too narrow a part of the scene. A later sample covered positions across the whole timeline. It still rendered some comparable poses more quickly than the continuous take had.

That observation narrowed the explanation. Spreading short samples across the scene did not by itself account for the sustained slowdown. Process samples located expensive rendering work, but did not isolate one cause for its changing cost.

The next full fixture completed its assembly and bake, passed selected reopened checks, then used the remaining preparation time in the timing probe. It never started recording. Calling that result a failed bake would misstate what the record showed.

The actual picture was smaller

Another defect emerged when PNG headers were inspected. The inherited scene used a central render border with crop enabled. A nominal 960-by-540 test had produced 576-by-324 pictures. The earlier nominal 480-by-270 passing excerpt had produced 288-by-162 pictures.

The tests had measured those cropped images. Their configured dimensions did not describe the delivered frames, and the profile had omitted the crop settings.

A later retained recipe made those settings explicit:

    "use_border": false,
    "use_crop_to_border": false

Public source excerpt. Repository snapshot, requiring repository access.

This is implemented configuration from the subsequent equipment profile. It records an uncropped frame. It does not retrospectively enlarge the earlier images or make their timings full-frame measurements.

The crop affected more than a dimension label. The scene needed projected words, readable action and appropriate views of its physical hands. Passing a parentage or control-coordinate check could not establish how those appeared in a reduced image.

The two failed full-length drivers together consumed about forty-four minutes and forty-five seconds, excluding intervening probes and earlier work. Their cutoffs bounded the failure, and their records stayed closed. Neither reached post-production.

The next useful maintenance route began from the saved, already-baked scene to investigate rendering without paying for another complete assembly each time, while measuring picture settings, actual dimensions and sustained costs together.

The small passing film remained valid evidence of a small working route. The larger failures explained which additional costs and conditions had to enter qualification before another whole-show claim was earned.