Every analyst-market evaluation returns one of four outcomes: BET (official — enters the pick log, leaderboard, and backtester), LEAN (research commentary, never counted), PASS (evaluated, no qualifying edge), or INVALID_DATA (failed validation, never evaluated). The default is PASS. There is no code path that manufactures an opinion.
Chalk (market confirmation). Consensus win probability of at least 62%, spread juice agreeing with the moneyline favorite, at least 3 reporting books, and a hard price floor of -260 — Chalk never lays a shorter price than that, no matter how likely the favorite.
Steamer (line movement). Real movement above a sport-specific threshold (e.g. 1.0 spread points in football, 1.5 in basketball, 0.5 total runs in baseball, 15-20 cents on moneylines), confirmed within a 72-hour window, with the last move no older than 36 hours. A frozen market is a PASS, always.
Iris (price dispersion). A best-available price at least 1.5 implied-probability points better than consensus, inside a -450/+450 price band, across at least 3 books. No measurable dispersion is a PASS.
These thresholds are documented priors, not backtested constants. The live data record is weeks old — too thin for honest train/validation/test splitting — so the thresholds were chosen from betting-market first principles, frozen in versioned code, and are being validated live-forward. When enough out-of-sample record exists, any tuning will use chronological splits with thresholds frozen before measurement, and in-sample results will never be presented as evidence. Book-count minimums are sized to the live odds feed (3 books today) and rise as coverage grows.
An official bet is recorded once, at first qualification, with the entry price at that moment — entries are never revised as lines move, because closing-line value measured against a continuously updated price is fiction. The closing consensus is stamped at kickoff. Grading uses the actual recorded price (a +118 winner returns 1.18 units, not 0.91); the -110 assumption survives only as an explicitly flagged fallback and is marked with * wherever it appears. Every bet stores the exact signals that qualified it.
Market data is validated before evaluation: alternate markets are rejected from official evaluation, out-of-band lines are quarantined with reasons, and suspended, postponed, or corrupt results never grade. The full validation breakdown is regenerated every build.
Today at the Desk · Pick Log · Leaderboard · Why V1 was frozen