Probabilistic Counterfactual Inference for Discrete Outcomes in Gaussian-Process Causal Models
This work creates a unified probabilistic framework for counterfactual inference in Gaussian-process structural causal models (GP-SCMs) with different variable types. It develops exact noise-abduction procedures for binary, categorical, and ordinal outcomes, proving they accurately reproduce the model's observed and interventional distributions. The key finding is that the choice of coupling mechanism based on variable type is more important than fitting accuracy.