Commodification of inference turns estimation mechanics into a framework service. After specifying a causal question, assumptions, estimand, and differentiable objective, the user can focus on modeling the data-generating process rather than implementing inference machinery. section 1.2.4 section 11.0
The pattern changes the practical workflow:
- State the causal assumptions and select the estimand, such as an average treatment effect or conditional average treatment effect. section 11.0
- Pass the relevant arguments to a library such as DoWhy, or express the objective through a framework such as PyTorch. section 11.0 section 11.4.6
- Compare estimators and stress-test causal and statistical assumptions instead of treating automated inference as assumption-free. section 11.4.6
Synthetic-data validation
Construct a causal generative model with known parameters, simulate data, and derive a ground-truth average treatment effect. Check whether DoWhy estimates approach that value and whether confidence intervals contain it before moving to real data. section 11.4.6
