Better decisions.
Before deployment.

Evaluate alternative decision policies from logged outcomes using inverse propensity weighting, doubly robust estimation and explicit overlap diagnostics. Inspect the uncertainty before trusting a result.