BY 
I spent six years building projection models before I ever wrote about betting, and one number did more work than all the others combined. Expected value. Once you can compute EV, you can evaluate any bet in about thirty seconds. Until then, you're grading yourself on results, and results lie constantly over small samples.
What expected value actually measures
Expected value is your average profit or loss on a bet if you could replay it thousands of times. The formula fits on an index card:
EV = (win probability x profit) - (loss probability x stake)
Profit is what you collect when the bet wins. Stake is what you lose when it doesn't. Win probability has to be your honest estimate of the true chance, and that's the hard part. The arithmetic is the easy part, so let's do the arithmetic.
Turning a price into a probability
American odds hide an implied probability, and you need it for every EV calculation you'll ever run. At -110 you risk $110 to win $100, so the implied break-even rate is 110 / (110 + 100), which is 52.38%. At +200 you risk $100 to win $200, and 100 / (100 + 200) gives you 33.3%.
Whenever your true probability beats the implied one, the bet is positive EV. That single comparison is the entire job.
The break-even bet, worked out
Suppose a -110 bet has a true win probability of exactly 52.38%. You risk $110 to win $100.
EV = (0.5238 x $100) - (0.4762 x $110)
EV = $52.38 - $52.38 = $0.00
Nothing about that is a coincidence, since 52.38% is precisely the break-even rate at -110. A bettor who wins at that clip forever finishes at zero, minus their time. This is why hitting 52% at standard juice still loses money, slowly and politely.
The same bet with a real edge
Now suppose your handicapping is genuinely good and the true win probability is 55%. Same price, same $110 at risk.
EV = (0.55 x $100) - (0.45 x $110)
EV = $55.00 - $49.50 = +$5.50
You expect $5.50 in profit for every $110 you risk, which works out to a 5% return on stake. That may sound modest. Sustained across hundreds of bets, it's a serious edge, and very few people ever hold one that big.
Notice what just happened, though. A 2.62 point improvement in win probability turned a dead-even bet into a profitable one. Real edges live in slivers that size, which is why sloppy probability estimates cost so much.
A dog that loses 60% of the time and still profits
Underdogs confuse people because losing feels like being wrong. Take a +200 bet with a true win probability of 40%, risking $100 to win $200.
EV = (0.40 x $200) - (0.60 x $100)
EV = $80 - $60 = +$20
This bet loses six times out of ten. It's also worth $20 in expected profit per $100 staked, a 20% return, which is enormous by betting standards.
The price only requires the dog to win 33.3% of the time, and your honest estimate says 40%. You're being paid underdog money for an outcome that happens more often than the market believes. That gap is where the value hides.
Why EV beats results on any single night
On one night, results are close to pure noise. The 40% dog above loses most of the time, and the person betting it made a good decision on every one of those losing nights. Meanwhile someone hammering a -300 favorite with a true 70% win probability cashes constantly and bleeds money, because -300 requires a 75% win rate just to break even.
Results answer the question of what happened. EV answers what you bought. Only the second one was under your control at the moment you clicked, so it's the only fair thing to grade yourself on.
How variance hides EV in small samples
A true 55% bettor at -110 owns real, positive EV, and over a random 20-bet stretch they can still go 8-12 without anything being broken. The edge didn't leave. The sample is just too short for it to surface, the same way a loaded die still rolls plenty of ones.
The reverse is nastier. A pure coin flipper can rip off a 15-5 run and look brilliant, and hot streaks recruit followers far faster than long records do. Variance is loud early and honest late.
My old team kept a jar you paid into if you cited a two-week win rate in a meeting. I paid in exactly once, and I still think about the bet that put me there.
Two traps to avoid
First trap: using the book's implied probability as your true probability. The price already includes the book's margin, so if your estimate merely matches the implied number, you're the product. Your edge has to come from information or modeling the market hasn't fully priced yet.
Second trap: computing EV with a win probability you invented to justify a bet you already wanted. I've watched smart people do this weekly. If your estimates always land conveniently above break-even, they're marketing, and the only cure is writing your number down before you look at the price.
Where to practice EV thinking
Grade every record you encounter by price and sample, never by streaks. That's most of why CAPTRACKER exists. Picks get timestamped and locked after a two-minute grace window, with no edits and no deletes allowed. Results auto-settle against ESPN data, so the record you see is the record that happened.
Browse the free leaderboard, which ranks more than 900 tracked handicapper profiles by units won and ROI, and see whose edge survives a real sample. If any vocabulary here felt slippery, the glossary has plain definitions, and the methodology page shows exactly how grading works.
Try this before your next bet. Write down the price and your honest win probability, then compute the EV. If you can't produce that middle number, you already have your answer.