Software

Throughout the material, we ground causal inference and decision-making concepts in a single, concrete decision context drawn from large-scale online retail: improving product data quality at scale. The ORS Online Retail Simulator provides the domain-specific environment in which methods and trade-offs are introduced. The Impact Engine then generalizes these ideas into a reusable ecosystem for measuring impact, evaluating evidence quality, and allocating resources across decision contexts.

Online Retail Simulator GitHub

Generates fully synthetic retail data for end-to-end testing of causal inference workflows. Supports controlled treatment effects, enabling validation of estimators and comparison of causal models against known ground truth.

Impact Engine

The Impact Engine is an open-source Python ecosystem that operationalizes the Learn, Decide, Repeat loop as a production pipeline. Four independently tested and deployable components map directly to the framework stages:

Measure GitHub

Estimates causal impact using a unified, config-driven interface. Wraps multiple estimation methods (SARIMAX, synthetic control, difference-in-differences, propensity score matching, causal forests) behind a single execute() call. Swap the estimation method by changing one line in a YAML config.

Evaluate GitHub

Scores how much to trust each impact estimate based on its measurement design. Supports both an LLM-powered agentic review mode and a lightweight deterministic scorer. The confidence score directly penalizes return estimates downstream, making the allocator conservative where evidence is weak.

Allocate GitHub

Determines where to invest using portfolio optimization under uncertainty. Supports pluggable decision rules (minimax regret, Bayesian weighted-scenario) and respects budget and strategic constraints.

Orchestrator GitHub

Wires all three components into a single config-driven pipeline. Takes a YAML configuration and runs the full measurement-to-allocation loop: config in, decision out.