cluster-randomization

Zero-to-hero notebooks on causal inference and experimentation for global LLM rollouts: synthetic control, difference-in-differences, propensity scores, regression discontinuity, cluster randomization.

Pre-registered, version-frozen protocol package for AI-for-Science validation experiments. ITT accounting + decision-anchored MRD + named adjudication. The validation bottleneck, solved.

Experimentation framework for synthetic accounting data with seven statistical designs, reproducible assignments, causal analysis, guardrails, power checks, and decision reporting.

Measures how badly a per-event significance test misreads a per-user experiment. Over 2000 A/A replicates the naive test claims an effect in 0.191 of them against a nominal 0.05. Prices what fixing it costs in power. Offline, seeded, reproducible in one command.