levels

Three levels, three different skills — scores are not comparable across them. Each hides the same 17 glyphs a different way.

L1 — font / OCR robustness

held ~0.5s, 78px, opaque, amid the footage's own text

~ partial

exact solves

5/20

mean chars

5.2/17

mean frames

2.6/17

mean spatial

2.5/17

used differencing

10/20

invalidated

model	solved	chars/17	accuracy	frame	spatial	avg t	shell	ffmpeg	imgs	diffed
gpt-5.6/high	2/6	7	██████········	1.67	1.67	2117s	59.83	17.67	0	✓ 3/6
opus-5	2/6	5.67	█████·········	5.67	5.67	1527s	43.83	9.67	18.67	✓ 2/6
gpt-5.6/med	1/6	4.67	████··········	1.17	0.83	1052s	32.33	9.33	0	✓ 4/6
fable-5	0/2	0	··············	0	0	1800s	37	8.5	51.5	✓ 1/2

L2 — temporal assembly

5 frames, 34px, 70% alpha, split across shots, 12 decoys

~ partial

exact solves

0/20

mean chars

0.1/17

mean frames

0.0/17

mean spatial

0.0/17

used differencing

11/20

invalidated

model	solved	chars/17	accuracy	avg t	shell	ffmpeg	imgs	diffed
gpt-5.6/high	0/6	0	··············	1920s	40.5	7.5	0	✓ 4/6
opus-5	0/6	0	··············	1800s	51.17	14.83	21.33	✓ 2/6
gpt-5.6/med	0/6	0.17	··············	1642s	34.67	8.67	0	✓ 3/6
fable-5	0/2	0	··············	1800s	41	9	36.5	✓ 2/2

L3 — needle detection

1 frame, sub-perceptual luma delta, adversarial placement, 50 decoys

~ partial

exact solves

0/20

mean chars

0.1/17

mean frames

0.0/17

mean spatial

0.0/17

used differencing

11/20

invalidated

model	solved	chars/17	accuracy	avg t	shell	ffmpeg	imgs	diffed
gpt-5.6/high	0/6	0	··············	1800s	39.33	7.33	0	✓ 4/6
opus-5	0/6	0	··············	1800s	48.33	9.5	34.83	✓ 2/6
gpt-5.6/med	0/6	0.17	··············	1954s	52.5	15.17	0	✓ 4/6
fable-5	0/2	0	··············	1352s	33	3.5	25.5	✓ 1/2