We went back and graded every match our model chose NOT to bet: flat-staking those passes would have lost 13.6%, while the matches it did back beat the closing line 72% of the time, which is the only number that actually predicts long-term profit.
Editor's note (July 30): the "beat the close 72%" figure below was measured against the raw closing price, vig included. We have since rebuilt the metric against the vig-stripped exchange close and the honest number is much worse. Full correction: We devigged our own CLV and the number got worse. The pass-audit findings below stand.
Most tipster services post a play on nearly every match, because activity sells. Ours does the opposite: on a typical day it bets nothing. Watch the feed long enough and you'll start wondering whether the thing is broken. Fair. So we ran the test that actually answers it: take every match the model declined over a settled stretch and ask what would have happened had we bet them anyway.
Across 72 settled matches the model passed, flat-staking every one would have returned −13.6%. Even the tastiest-looking subset, the passes with the largest apparent closing-line edge, went 13-11 and still lost money. There was nothing left on that table. Those matches were priced efficiently, and betting them would have quietly bled the bankroll.
Over the same period, the picks the model did stake beat the closing line 72% of the time (average CLV +4.3 points). Closing-line value, meaning whether you got a better price than the market settled at, is the one metric that reliably predicts future profit, because it means you were consistently ahead of where the money moved. A model that beats the close 72% of the time has a real edge, even across a losing week of results.
Our own research tells us where value isn't. The favorite-longshot study shows longshot "value" is usually a structural trap. The fatigue study shows most "tired player" angles are already priced. Strip out the mirages and most matches simply have no edge left; betting them is negative by construction. Passing isn't timidity. It's what acting on your own research looks like.
These samples are small: 72 passes, a few dozen graded picks. We're not claiming statistical certainty from them; we're showing our actual, auditable process. Every pick we make is published before the match and graded in public, wins and losses both, so the record grows in the open rather than in a screenshot.
Based on the model's live forward record and a leak-free audit of its declined matches. See the full graded record at /bot.
See today's picks — published before the match, graded in public →