Ask the analysts
A question in plain language is routed to the research questions it matches, answered by the specialist agents, then recomputed independently by a critic before it is shown. Numbers come from the statistical engine — never from a language model.
Verified findings
Agent trace
idleSearch the corpus
The ten questions
What happens when you press run
The question is matched to research questions on lexical evidence, and the routing is shown so you can disagree with it. A tagger audits label completeness. The owning specialist computes the statistic. The critic then recomputes the headline number from the raw records on a separate code path, checks that table counts match the denominator, and rejects causal wording on association results. A rejection sends the question back for one corrective pass.
This page is the application, not a summary of one. The written report, the figures and the codebook remain available at /report/.