Told the rules and tested fair, can a frontier AI match our trained model?

BLOCKED: Fable-5 quota was exhausted before the frontier side of the rematch could run. This is the fair rematch Alex explicitly asked for, and it's still owed.

What the measured part found

The two incumbent models WERE measured on the fixed 40-photo sample. The frontier AI side is the piece blocked on quota.

Model	Stripes found	False lines / photo
v26 (production)	96.1%	0.225
deep20	95.3%	0.400
Frontier AI (Fable-5)	Unmeasured — quota blocked

Sample: 40 photos, 127 human-labeled stripes, none used in training any of our models.

Source: frontier-rematch.json.

raw data (JSON) →