You can run this alone. Prop the phone up, tap Record, and you get 8 seconds and a spoken countdown to walk into frame. The app talks to you: it says when to move closer or back, beeps when recording starts and stops, and tells you whether the take was kept. You never need to read the screen while you are the subject.

Each condition records for 20 seconds. Stand the same distance every time, in the same light: the app checks both and will refuse to compare recordings that do not match, because distance and brightness change detection on their own. The printed garment needs three takes, and the worst one counts.

Sessions are kept in this browser and survive a refresh. Nothing is uploaded. Newest first.

The rig's thresholds, all tunable. Hover the ? on any control for what it does. The detect line re-scores every take you have recorded; a value that would make a comparison unfalsifiable is refused, and the refusal says why.

X-Ray shows where the model responds. Occlusion covers the subject patch by patch and measures how far the score falls, which says which part the model was actually using. They look alike and mean different things. Occlusion is 132 passes over one frame, so it runs on demand.

Every garment measured on this device, worst take governing, ranked by how its takes compared with its own plain control.