start the camera, then: raise a hand and lower it · or hit the ding and draw a slow ⭕ in the air
🎞️ the 1fps review artefact (what a reviewer would see)
One frame per second, already blurred — a moving silhouette, not a video of a person.
Gold-ringed frames = the continuation window a reviewer's playback slows down on. The sharp image
never exists anywhere: blur is applied in the same draw call that reads the camera, and the slider
cannot go below the 62% floor.
what this proves
1 · blur-at-capture: the only pixels ever drawn, stored, or analysed are post-blur (the tracker runs on the blurred 96×72 frame itself). 2 · consciousness check: raise-and-lower detected by motion-centroid rise + return — no ML, no identity. 3 · continuation ⭕: after a ding, ~8fps centroid samples for 6s → the harvested path classifier → kind: circle. 4 · the artefact: 1fps strip + a flagged continuation point — review in seconds, judgement aimed where it lives.