Last released Jun 3, 2026
Counterfactual replay and causal attribution over LLM-agent trajectories: find which step actually caused a bad outcome by intervening on it.
Supported by
