chc.causal_bench

Stability tier: experimental.

A causal-methods leaderboard: each modern estimator vs the naive built-in it is meant to beat.

Every method in the causal frontier (chc.did, chc.scm, chc.estimators, chc.gmethods) ships with a test proving it recovers a known effect where a naive baseline is biased. This consolidates those into one table: for each method it draws a self-contained synthetic DGP with a ground-truth effect, runs the method and its naive baseline, and reports both biases so the win is visible side by side. NumPy orchestration; the R-learner row builds a JAX payload internally.

CausalBenchRow dataclass

One method-vs-baseline comparison on a DGP with a known effect.

run_causal_bench(seed=0)

Run every frontier causal method vs its naive baseline on a known-effect synthetic DGP.

causal_bench_report(rows)

Format the causal-methods leaderboard: method and naive baseline bias, side by side.