Loading…
Archive · nothing edited after the fact
The archive is the product's receipts. One beam per bout — where the model stood, where the market closed, and where the result landed. Green called it; oxblood missed it. Nothing is edited after the fact. See the rolling calibration record →
Across 10 graded cards, the model called 65 of 91 winners, trailing the market by 3.9 pp where a line existed. Every read was locked before the event and graded against the result — nothing edited after the fact.
Brier 0.163 · last 4 events
· 10 fights had no line and no model read
Correction (2026-07-17): 16 historical rows carried a market snapshot captured after their bouts had started. Those snapshots are now excluded from market comparisons; every model read and result remains on the record unchanged.
Every bout the model graded, newest card first. Each beam shows the locked model probability, the closing market where one was captured — otherwise the market when the read locked — and where the result landed. “Called” counts every graded bout; the points-ahead figure compares model and market only on bouts that had a line.
Forecasts were locked before each event and graded against the official result. Nothing edited or removed. Analytical only — not a betting claim. 21+. Problem gambling? Call 1-800-GAMBLER.