Statistical analysis for LLM evaluations, from model and prompt comparisons to inference resilient to LLM judge bias, including at small sample sizes. All defaults battle-tested in Monte Carlo simulations.
