ResearchBenchmarksGroup Booking with failures (Grand Rollback) on τ2-bench+ / REALM-bench+ P14 v1 (test)Follow2Result Success RateRAC-0.080.4611.54May 5, 2026Evaluation ResultsMethodMethodLinksResult Success RateTokens Consumed (k)Time Taken (s)Completion CountRACFramework=RACFramework=RAC2026.052189740SagaLLMFramework=SagaLLMFramework=SagaLLM2026.051951732LGFramework=LGFramework=LG2026.050—180LG(PE)Framework=LG(PE)Framework=LG(PE)2026.050—170