Model transparency
A confident model is worthless if its confidence is wrong. This page holds AGI.ML accountable to its own numbers: when it says a pick wins 65% of the time, does it? We plot predicted probability against real, settled outcomes — no cherry-picking.
The reliability curve needs graded picks with recorded predictions before it can mean anything. As today's slate settles against final scores, this page fills in on its own. In the meantime, see the track record for the running win/loss ledger.
Calibration is descriptive of past, settled selections only and is not predictive of future results. Sample sizes in early bands may be small. 21+. Please gamble responsibly.