OddsRecord

Luck is wider than skill

What 100 bets can—and cannot—prove

A short record can look extraordinary even when the person making it has no edge at all.

43%–62%95% range over 100 bets
52.381%break-even win rate at −110
±9.5 ptsordinary luck over 100 bets
+1.57 ptswin-rate lift for a 3% ROI edge

The 100-bet range

Start with someone whose true win rate is exactly the −110 break-even rate: 110 ÷ 210, or 52.381%. Over 100 independent bets, about 95% of their records fall between 43 wins and 62 wins. The same no-edge process can look terrible or brilliant without changing at all.

That is why a 62% screenshot is not proof. It sits near the high end of an ordinary no-skill range. The matching 43% record is just as compatible with the same underlying ability.

Luck is about nine points; a real edge is about one and a half

A strong 3% return on money risked at −110 corresponds to a 53.952% win rate—only 1.571 percentage points above break-even. Over 100 bets, ordinary random movement is roughly six times wider than that signal.

A forecast record is not a profit promise. Bet selection, changing prices, limits, timing and correlation all matter. A model can publish well-calibrated probabilities and still offer no usable price advantage.

How long does evidence take?

Using a one-sided 5% significance test with 80% power, separating a true 3% return from the −110 break-even rate takes about 6,237 independent bets. That is the scale of evidence needed when the real difference is small.

The arithmetic

At −110, a win returns 100 for every 110 risked. Break-even solves p × 100 = (1 − p) × 110, giving p = 0.5238095. A 3% return solves p × 1.909091 − 1 = 0.03, giving p = 0.5395238. The difference is 0.0157143, or 1.571 percentage points.

The 43%–62% interval is the rounded 95% binomial range around the break-even rate for 100 independent trials. Real betting records can be correlated, which can make the effective sample smaller—not larger.