Schema harness achieves ~99% on ARC-AGI-3 public set

The Schema harness reportedly achieves 99% on the ARC-AGI-3 public benchmark using models like Fable+4.8 or GPT 5.6 Sol. The result applies only to the public set, not the held-out set.