What variance does to a good bet
A real edge still produces long losing runs. Here is the independent-trial arithmetic, its assumptions, and why small edges need large samples.
On this page
Suppose a method is genuinely good: it finds bets at 2.00 that win 53% of
the time. Under those assumptions its expected profit is 6% of each unit
staked. It will still lose about 47 of every 100 bets over the long run, and
that is where almost everybody's intuition breaks.
What a normal bad run looks like
Assume for the moment that every bet is independent and keeps the same 53%
chance. A specified block of eight bets has a 0.47^8, or about 0.24%, chance
of being all losses. Across 500 bets there are many overlapping blocks; exact
run arithmetic puts the chance of at least one eight-loss run at about 47%.
It is neither rare nor guaranteed.
The same arithmetic puts the chance of at least one ten-win run at about 34%. Either streak can occur without the underlying win probability changing.
Why this changes what you should measure
If good weeks and bad weeks are both routine, then reading either as evidence is reading noise. The method has not changed; the sample has.
This is the reason this product publishes the closing line rather than the profit as its primary scoreboard. Whether a bet won is one bit of information arriving when the event settles. Whether it was taken at a better price than the market eventually closed at is measurable on every bet once that market closes, including the ones that lost.
What it does to staking
Variance is also the reason staking is not a detail. A method with a real edge and stakes that are too large for the bankroll it is drawn from can still run out of money before its long-run advantage can dominate short-run variance, and running out of money is final in a way that a bad month is not.
What this does not tell you
None of this establishes that any particular method has an edge. Variance cuts in both directions: it hides real edges under bad runs, and it manufactures fake ones out of good runs. A year of profit is not proof, and neither is a year of loss.
Common questions
How many bets does it take to know if a method works?
It depends on the edge, odds, independence assumptions and evidential threshold. For independent even-money bets, reliably distinguishing 53% from 50% requires a sample in the low thousands under conventional test settings.
Is a losing month evidence that a method is broken?
Not by itself. Its evidential weight depends on the number of bets, their odds, the claimed edge and whether their outcomes can reasonably be treated as independent.
What is a better signal than profit?
Beating the eventual closing price can be a faster pricing diagnostic because it is available for every bet once the market closes, including bets that later lose.