Expected value (EV) is the average profit per unit wagered if your probability estimate is correct. Positive EV means the model believes the market mispriced the bet. Here's the math + why it's the only metric that matters long-term.
EV = (probability × decimal_odds) - 1
Or expressed as percent: EV % = (prob × decimal - 1) × 100
If a pick has 60% probability at -110 odds (decimal 1.909), EV = (0.60 × 1.909) - 1 = 0.145 = +14.5% per $1 wagered.
Short-term wins and losses are noise. A coin flipper hits 60% of his calls in any given week 25% of the time by pure variance. To know whether you're actually beating the market, you need a metric that smooths variance.
EV is that metric. Over a large sample of +EV bets, your bankroll grows. Over a sample of -EV bets, it shrinks — even if you have winning streaks. Long-term ROI converges to mean EV.
EV and edge are related but distinct.
Edge measures probability disagreement; EV measures expected return per dollar. A pick with 5% edge at -300 odds has different EV than 5% edge at +200 — though the edge is identical, the payout structure changes EV.
You need a probability estimate that disagrees with the market in a positive direction. Three ways:
For every pick we display, EV is computed against the de-vigged market closing implied probability. Picks with positive EV get a recommended stake; picks with negative EV get 0u (Kelly says "don't bet"). The display always shows the EV honestly — even when -10% — so you know which picks the model believes have value vs which don't.
+EV doesn't mean a single bet wins. It means over hundreds or thousands of similar bets, you make money. Picking a +5% EV bet that loses tonight isn't a model failure — it's variance. The same +5% EV bet over 1000 plays expects to return $50 per $1000 staked. That's the math; the variance is just noise.