Verdict 001 / filed 2026-09-20

One photo held him. The wide shots are where it slips.

Subject: this site's own eighteen plates. One reference photo, no trained adapter, seventeen frames rendered off plate one and scored against it.

Use for faces, not for wide shots

0.58 worst frame of the set, own likeness, scale 0 to 1, wide gate 0.55

FILED1 verdict
SUBJECTmy own eighteen plates
SCALEworst frame, 0 to 1
FACE GATE0.60
WIDE GATE0.55
MY OWN WORST0.58

The receipt

Two columns, because they are two different claims. The left column is what the scorer measured. The right column is what I saw. When they disagree the disagreement is the finding, and on this one they disagree twice.

Verdict 001, per cell. Seventeen frames, one reference photo each, measured against plate one. Gates are the ones in force on the filing date.
Cell	Measured	Gate	Observed by eye
Close, eight frames	0.886	0.60	Holds. The scar through the left brow reads in all of them.
Medium, four frames	0.889	0.60	Holds. This is the cell I would publish from.
Body, one frame	0.837	0.55	Holds, and the apron straps are in the right place.
Room, one frame	0.891	0.55	Holds. The lamp is doing the work.
Wide, three frames	0.584	0.55	One of the three is thirty pixels of face. It passes by nothing and I would not publish it.
Tightest crop, one frame	no face found	0.60	The best face on the sheet. The scorer failed, the render did not.

Across the workshop, small in frame at the far end of the long bench under two hanging lamps, tools and pegboard filling the wall, flat warm light, one person only in the room. — A. Worst frame of the set, measured 0.584. A man thirty pixels wide in a room that is the actual subject of the picture. Plate 16

Hard overhead fluorescent light in a plain hardware aisle, cool green cast, face to camera, grey chambray work shirt, canvas apron. — B. Best wide frame, measured 0.901, from the same route and the same reference. Same recipe, four times the face. Plate 18

What broke, and why

Nothing broke in the renders. What broke is the measurement, twice, in opposite directions.

The wide frame that scored 0.584 is not a bad likeness. It is a small one: the face is about thirty pixels across because the room is the subject. The number is honest about pixels and says nothing about identity, and a gate applied blind would have thrown away the best picture on the sheet.

The tightest crop scored nothing at all, because the scorer found no face to measure. That is the same failure wearing the opposite mask. If I published the number without looking, I would have filed my own best frame as a failure.

What I did not do

I did not compare this against a trained adapter of myself, I did not test a second subject, and I did not measure what happens past a hundred frames. Those are three different verdicts and they are not filed yet.

The log

Every verdict filed so far, newest first. There is one. It is about this site, the log starts at one on purpose, and nothing here is backdated.

Verdict log. Score is the worst frame in the cell that decided the verdict.
No.	Filed	Subject	Score	Verdict
001	2026-09-20	This site's own eighteen plates, one reference photo, no trained adapter	0.58	Use for faces, not for wide shots

Open the whole log

Write to the studio

Corrections are the most useful mail this site gets. If a verdict here is wrong, say which cell and why, and it will be re-tested and the reversal filed in the log under its own number.

[email protected]