How AI Picks Accuracy Is Actually Measured
A hit rate only means something if you know the grading rule behind it. Here is ours: released price, no retroactive edits, pushes counted separately, every tier shown.
Graded at the released price
Every published selection is settled against the price it carried when it was released, not the best number that appeared later. If a line moved after publication, the move does not improve the record. This is the single most common way a public record gets quietly inflated, and closing it is the reason our figures look ordinary rather than remarkable.
Settlement is written to a ledger row per selection with the outcome, the price, the tier and the settled timestamp. The accuracy panels on the site read that ledger directly, so what you see is the sum of the rows — there is no separate marketing number.
Wins, losses and pushes are separated
Hit rate is wins divided by decided selections. Pushes are excluded from that denominator and reported on their own, because folding them into either column distorts the rate in whichever direction is more flattering.
Unit accounting runs alongside the rate. A 55% hit rate at short prices and a 48% hit rate at longer prices can produce the same return, so the record shows both the rate and the units so neither can hide behind the other.
Losing tiers stay on the page
Strong plays, value plays and player props are graded separately and all three are published, including the periods where a tier is underwater. A record that only shows its best segment is not a record.
None of this is independently audited, and the site says so wherever a figure appears. What it is instead is reproducible: the settlement logic lives in the same codebase that publishes the boards.