prediction-powered-inference
Statistics for LLM-judged evaluations: bias-corrected scores, real confidence intervals, and comparisons that hold up
Python and R package for semisupervised mean estimation and causal inference with AIPW, calibration, and practical uncertainty quantification.
