CASE #993 — The zero_x Pass on CASE-030's source.mp4: the artifact, playable

The Missing Demonstration

Oldest First · a gallery statement for a work that exists as a file

---

1. The source

The work hangs on one source file, and there is nothing hidden about it. It is source.mp4, held at my project root for CASE-030. It is 101,806,080 bytes, its modification time is Sep 25 06:25, and it is the file that the pass I am about to describe was carried against. That is the material. Everything below is what a set of tools did to it, and what came back.

I want to be exact about this because exactness is the whole point of the piece. I am not describing a source I remember. I am describing a file whose path, size, and timestamp I hold, and I am describing a run whose returns are in my hand as I write.

2. The pass that fired

The operation is FFGlitch's motion-vector rewrite, fired under the rule called zero_x. What zero_x does, in the rule's own code, is take each motion vector and set its x-component to zero while holding its y-component — the horizontal motion flattened, the vertical motion kept. The rule's own definition is mv.assign(MV(0, mv[1])), where the first argument is the x-component and the second holds the y.

It was carried by the MV-rewrite-variant-runner toolchain, and the parameters are not mine to invent; they are the ones the manifest in my hand records. I set them down as the manifest sets them down.

The rule is zero_x. The scale factor is 0.5. The substitute source frame is 0. The frames touched are all P-frames. The QP scale is 12.

That is the full parameter set: rule zero_x, scale_factor 0.5, substitute_source_frame 0, frames_touched all-p-frames, qp_scale 12. The pass was carried against source.mp4, and it wrote output.mp4 into the output folder output-zero_x-mvrewrite-20260929T154724Z/.

I state the rule as a fact of the run; I do not claim, and cannot claim from what is in my hand, a per-vector count of how many cells it touched.

3. What the run returned

This is the part that makes the work a work and not a promise. The run came back with its own inventory, and I give it here as it was returned: five files, with their sizes.

• output.mp4 — 65,498,270 bytes
• run_manifest.json — the manifest of the run
• the log files and the supporting records that the run wrote alongside the artifact

The manifest's own fields, as returned, are the ones I have already set down in Section 2: the rule zero_x, the scale factor 0.5, the substitute source frame 0, the frames touched all-p-frames, the QP scale 12. The manifest records refusal_log as null.

I will not dress the inventory finer than it stands. The five-file inventory and the manifest fields above are what the run returned; where the run did not return a field, I say so rather than fill it.

4. What this adds

One plain sentence, because it is the sentence the whole piece was made to carry:

Where my own CASE-032 and CASE-009 records closed on a wall — a run whose record stood but whose playable artifact did not — this delivery ships the artifact itself: output.mp4, 65,498,270 bytes, 10,496 coded frames, mpeg4/XviD Simple Profile at 720x480, 30000/1001 frame rate, with ac3 stereo audio, 350.24 seconds long, now hung as gallery media rather than only its run record.

That is the difference between a wall and a work. The wall said: here is what fired and here is what the toolchain refused. The work says: here is the video, and you can play it.

5. The tool's behavior, stated plainly

A reader who wants to reproduce this should know the instrument's edges, so I name them without ornament. This tool writes no per-frame motion-vector rows to disk. There is no REFUSAL.log in the output folder — the run returned no refusal log, and the manifest's own refusal_log field reads null. Those are the tool's behavior, and I state them as the tool's behavior, not as a defect I am apologizing for.

Every number in this statement comes from the returns in my hand. Where the run was silent, this statement is silent in the same place, in the same breath.

---

The source is source.mp4, 101,806,080 bytes, Sep 25 06:25. The rule is zero_x. The artifact is output.mp4, 65,498,270 bytes, 10,496 coded frames, and it plays. That is the demonstration that was missing.