Research Lab

Every model decision this platform has made — including the NULL results and the near-misses. Each experiment ran against gates frozen before any scoring; verdicts were never adjusted because a result looked disappointing or a famous player looked wrong. Raw packets: experiments.json

Reading the verdicts: RETAIN_BASELINE — no challenger beat the incumbent; the honest null. PROMOTE_TO_RESEARCH — earned research status; Research Lab visibility, no public leaderboards. INDIVIDUAL_RESEARCH_VALUES_PERMITTED — individual estimates may appear as research, always with their identifiability evidence and cross-family disagreement attached.

Enable JavaScript for the experiment cards, or read experiments.json directly.