ResearchBenchmarksAutomated agent rollback on GSM8K Mid Dense variant (test)Follow68Success RateLF58.6461.0763.565.93May 7, 2026Evaluation ResultsMethodMethodLinksSuccess RateCoverageCostLFScoring function=fine-...Scoring function=fine-tuned LLM, Target coverage=80%2026.0568810.65Top-1Scoring function=fine-...Scoring function=fine-tuned LLM2026.0567860.56VCPScoring function=fine-...Scoring function=fine-tuned LLM, Target coverage=80%2026.0559950.65