Yingfei Wang

CALM Cuts Trial-Estimate Variance While Preserving Validity

LLM counterfactual predictions are calibrated and cross-fitted into causal estimators, lowering variance across 300 BRIGHTEN-based simulation runs.