0xNagato

confound-check

Stress-test causal claims for confounds before anyone believes them. Use when an analysis concludes that one thing caused, drove, lifted, hurt, or explained another; when reviewing A/B test results, observational analyses, backtests, or metric-movement postmortems; when designing a study, experiment, or backtest so controls get decided upfront instead of defended afterward; and whenever causal language ("drove", "because of", "the impact of", "X% lift from", "attributable to") appears over data that was not randomized. Also use to audit queries, notebooks, and pipelines for structural bugs that manufacture confounds — lookahead bias, survivorship, mix shift, post-treatment controls.