Experimental Rules-Based Models · Frozen July 28, 2026
Desk V1 was our first generation of automated analysts — Chalk, Steamer, and Iris — each a rules-based model that read live market data and published picks. Every pick it ever made, and every grade against a final score, is preserved unedited in the Pick Log and Leaderboard.
Running V1 in public taught us more than any backtest could, and the honest summary is that it had structural problems we are not willing to paper over:
It forced opinions. Every analyst produced a moneyline, spread, and total call on essentially every game. When no real edge existed, fallback rules invented one — the plus-money side, the cheaper juice, the frozen market. A model that always has an answer is a content generator, not a research desk.
It assumed prices. Spread and total grading leaned on a universal -110 assumption instead of the actual recorded price, which quietly distorts units, ROI, and closing-line value.
It trusted data too much. Market records flowed into pick logic without sport-specific sanity validation, so occasional malformed lines could reach the desk.
Desk V2 launches with a fresh official record and a different philosophy: the default outcome for every market is PASS. An analyst publishes only when a documented, sport-specific signal threshold is met, prices are recorded and graded as structured point-and-price data, and every official bet stores the exact signals that qualified it. V1 results will never be mixed into V2 performance.
Desk V2 is now live with its own fresh record. V1 picks made before the freeze are graded to completion here so this record closes honestly. See the V2 methodology.