Empirical stuff
An independent framework for turning authoritative public evidence and sector research into commercially credible data and AI opportunities.
Two years of commits to every documented Trino connector, with fleet-wide housekeeping separated from real attention: the lakehouse three absorb half of it, the median connector gets less than one specific commit a month, and two people write a third of all of it.
146 years of House roll-call votes: American politics was often this divided, but it has never been this one-dimensional. The second axis of conflict is dead, and that changes how everything else works.
Alpha growth — the part of revenue growth a firm's environment does not explain — validated on synthetic data, then measured on seventeen years of SEC filings.
A map of 7,135 papers from 2021 to mid-2026 on agri-food robotics: themes, growth, geography, venues, influential work, and the frontier.
Market research on device-neutral whole-person health synthesis: a $16.1B to $68B app layer, the Oura/Whoop threat from above, and the open device-neutral quadrant.
Why voting no longer reaches far enough. A 22,000-word structural analysis of British democratic discontent — through trust data, comparative institutional design, and four policy machinery maps.
Five layers, six definitions of revenue, and a seventy-point accuracy cliff when text-to-SQL meets real enterprise schemas. A field report on federation and the semantic layer in May 2026.
Six benchmarks, nineteen models, and four exploit techniques that score 100% without solving anything. A field report on what LLM evaluation actually looks like in May 2026.