Know where it breaks.
Falsification for systems that make decisions.
Run a stochastic system twice and the outcome can diverge. A single trial is one draw from a distribution; it carries no mathematical claim about reliability.
Monte maps the failure boundary. It sweeps environments, aggregates trials, and calculates the exact statistical surface where a system works and where it collapses.
Results are durable. Every condition caches its evidence. Every boundary shift is tested for statistical significance.
Reinforcement learning and robotics policies. LLM agents. LLM pipelines.
A quadruped locomotion policy, mapped across 400 worlds and retrained against what the map exposed. 60 conditions became significantly safer, none significantly worse, for $42 of compute. The map is live.