Disco

Connects Claude to Leap Labs' hypothesis-free pattern discovery engine for tabular data. Exposes a `discover()` operation that takes a CSV or DataFrame, runs ML-based pattern extraction with holdout validation and FDR correction, then returns structured findings with effect sizes, p-values, and novelty classifications checked against academic literature. Each pattern includes specific feature conditions (like "humidity 72-89% AND wind speed below 12 km/h increases yield 34%"), not just correlations. Useful when you're staring at a dataset without a clear hypothesis, need to surface interaction effects that standard analysis misses, or want statistically validated subgroup patterns you didn't think to look for. Public runs are free but published; private analysis costs credits. Runs take a few minutes and return both programmatic output and a web report URL.