ASRA Phase 2 — Original ARC Full-Dataset Evaluation Results
Empirical results for the ASRA Observation Engine on the Original ARC corpus (800 tasks): 100% rule-candidate coverage, ~98% cross-demo common-rule consistency, transform-event distributions, training vs evaluation complexity gradient, and branched-per-demo resolution for 17 exception tasks.