How to use the Kelly criterion

How much of your bankroll should one bet be?

Kelly gives the stake that maximises long-run growth given the true probability. You do not have the true probability — you have a devigged estimate with an error bar.

That asymmetry decides everything. Betting twice the correct fraction has negative growth; betting half has about three-quarters of the growth at a quarter of the variance. Overestimating an edge costs far more than underestimating it, and devigged estimates are exactly the kind that get overestimated.

So: fractional Kelly, and size on the least favourable devig rather than the friendliest. Cap any single bet regardless of what the formula says — Kelly on a genuine 30% edge and on a stale line ask for the same number, and the cap is what makes the difference survivable.

One more trap: Kelly assumes bets resolve one at a time. Twelve overs on one game script is one undiversified position wearing twelve hats. Size a bet →