List your company

Policy Repair Evaluation on Production Episodes 5,000 episodes (Held-out)

108.749RevPARBenchmark policy
98.51748101.17374103.83106.48626Jul 3, 2026

Evaluation results

MethodLinks
Benchmark policy2026.07108.74976.96141.31000
Multi-restart LLM editor2026.07108.47279.63136.210.2660.6091.197
Diagnostic projection2026.07107.90280.2134.540.4850.6851.108
Starting policy2026.07102.75185.631201.1531.1531.153
Tree-search editor2026.0798.91175.82130.450.2140.7660.328