How this scoreboard works: receipts first, no dunking

Every graded entry on this scoreboard is a receipt: a dated claim, a named predictor, a stated rubric, and archived evidence — source URLs, exact quotes, and SHA-256 hashes pinned at fetch time, including the receipt that proves the predictor made the claim in the first place. Outcomes are produced by a two-model, blind-structured pipeline in which one model grades and a second verifies against the same hashed evidence without seeing the first model's reasoning; when the two disagree the entry is published as contested rather than graded, so the disagreement stays on the record instead of being resolved out of sight. Each prediction is assessed strictly against its own rubric and target date, so a CEO's headline claim and an anonymous forum post are read through the same process. The design goal is a fair, auditable record of who said what and when — receipts over dunks.