ResearchBenchmarksTask Resolution on τ2-bench (test)Follow38.3Success Rate (Airline)ERL36.63637.06837.537.932Mar 25, 2026Evaluation ResultsMethodMethodLinksSuccess Rate (Airline)Success Rate (Retail)Success Rate (Telecom)Success Rate (Overall)pass@3 (Airline)pass@3 (Retail)pass@3 (Telecom)pass@3 (Overall)ERLHeuristics=from traini...Heuristics=from training split, Trials=32026.0338.347.528.3382517.5014.2BaselineScaffold=ReAct, Trials=3Scaffold=ReAct, Trials=32026.0336.743.33036.72012.51014.2