Disco

Disco is a hypothesis-free pattern discovery engine for tabular data, built by Leap Laboratories Inc. It fits neural networks to uploaded datasets, applies interpretability methods to extract learned patterns, and returns a ranked list of statistically significant findings—complete with p-values, effect sizes, and citations from academic literature. Each pattern is validated on hold-out data and corrected for multiple testing, surfacing feature interactions, nonlinear thresholds, and subgroup effects that neither humans nor LLMs would find through conventional analysis.