A real 5% edge and a lucky few weeks can look identical at first. Both can produce winning bets. Both can feel like proof that a strategy works. The difference only becomes clear with enough volume for a genuine edge, if there is one, to stand out from normal variance.
What "average wait events" actually means
A single-selection bet at 1.90 decimal implies a 52.6% chance. Losing the first attempt tells you very little on its own. At just over a coin flip, losing once is completely normal, and the average wait is roughly two attempts before one lands:
- 1.90 decimal (52.6% raw) - average wait, about 2 events
- 4.00 decimal (25.0% raw) - average wait, about 4.2 events
At 4.00, three losses in a row before a win does not mean the strategy is broken. It is the expected behaviour of a one-in-four event, which will lose several times in a row fairly often. "Average wait" means exactly what it says: the expected number of attempts before a hit, given the probability. It is not a countdown, and a losing run longer than the average is not evidence that something is wrong. Roughly half of all such sequences run longer than the average.
Why a handful of results proves so little
The mathematically meaningful sample size for confirming a modest edge is larger than intuition suggests, and it scales with the square of how small the edge is. A rough way to see why is to consider a flat-staked bet at roughly even odds. The noise around your average result per bet is similar in size to the bet itself, while a modest edge is only a small fraction of that. As a result, the number of bets needed before the edge reliably outweighs the noise grows roughly as one over the edge, squared:
bets needed, roughly ≈ 1 ÷ edge²
A 10% edge needs on the order of 100 bets before it reliably separates from noise. A 5% edge needs on the order of 400. A 3% edge, common for a carefully found value price, needs on the order of 1,100. This is a rough approximation, not a precise confidence calculation, but the relationship matters more than the exact numbers. Cutting the edge in half roughly quadruples the number of trials needed to see it clearly, which is not what most people expect.
The trap this sets
Three, five, or even ten results are not a meaningful sample size in this context. They are simply a small snapshot of variance. A run of losses on a real, positive-EV strategy is not evidence that it is broken. A run of wins on a strategy with no real edge is not evidence that it works. After three to ten bets, both situations can look exactly the same as a strategy that is working as expected or one that is genuinely broken.
The difficult part is that the period when a new approach feels most in need of a verdict, the first couple of weeks, is precisely when there is not enough data to make one.
The same number, read as a bankroll question
An average wait of roughly four events at 25% describes the middle of the distribution, not its extremes. Plenty of genuine 25% shots take six, eight, or more attempts before landing, purely because of variance. There may be nothing wrong with the underlying probability at all.
Bankroll planning that assumes the average wait is close to the actual wait every time is planning for the typical case and hoping the unlucky one does not happen. Keeping individual stakes small enough to comfortably absorb a run well beyond the average, rather than sizing them to just barely survive it, is the practical way to account for variance.
What actually helps
None of this means you should ignore results. A process that keeps losing well beyond what variance alone should produce is worth re-examining, and tracking expected value alongside win rate can help identify that sooner than win rate alone.
The point is narrower: the first handful of results after any change, whether positive or negative, carries much less information than it feels like it does at the time. Treating those results as a verdict rather than as data is a common mistake.
- Track the price you got against your own probability estimate, not just win or lose. Expected value accumulates evidence about whether your estimates are sound long before win rate alone would, because it uses the size of each edge, not just its sign.
- Judge the process, not the run. Whether a bet was well-reasoned at the time it was placed is knowable immediately. Whether it was profitable in expectation is only knowable after enough repetitions - conflating the two is how a good process gets abandoned after a bad month, or a bad one gets kept after a good one.
- Expect the losing streaks a real edge implies. A strategy that never has a bad week at the volumes above is more likely under-sampled than exceptional.
Run the odds you are actually looking at through the feasibility calculator to see the average wait for a specific price, and consider any short run of results in that context before drawing a conclusion either way.