Hunch Lab: Methodology

Method note for Hunch Lab (hunchleads.com/lab), which publishes descriptive findings of the form "companies that had event A are followed by event B more (or less) often than comparable companies", based on public registers covering about 1.14 million Belgian companies. The note describes the data and inclusion criteria for event types, the retrospective cohort design with two measurements in separate periods, indirect standardisation within strata of sector, size and age, the three comparison groups (active comparator, economically active companies, full register) and the agreement rule, the statistics (exact Poisson intervals, exact Poisson test, Benjamini-Hochberg false discovery rate control, absolute differences), the status rules, the automated two-round review with code-executed robustness tests (shifted windows, placebo events, negative controls, E-values, stratum exclusion), publication rules, limitations, conflicts of interest, the use of language models, and the version history of the method.