ResearchBenchmarksDecision classification on Damage-audit (40-case held-out)Follow76.4Action Accuracyv175.04875.39975.7576.101Jun 9, 2026Evaluation ResultsMethodMethodLinksAction AccuracyCategory Accuracyv1Executor regime=Strong...Executor regime=Strong 5-model avg2026.0676.436.4Auto-EISRExecutor regime=Strong...Executor regime=Strong 5-model avg, seeds=3, regression gate=disabled2026.0676.438.8Human-v8Executor regime=Strong...Executor regime=Strong 5-model avg2026.0675.152.2