Kelly tells you the bet size that grows a bankroll fastest over time. It is also far too aggressive to actually live with, which is why serious bettors take a fraction of it. This one runs a quarter and caps at 3%.
Kelly asks a narrow question: given this price and this win rate, what fraction of the bankroll maximizes long-run growth? The answer is exact, and it is almost always bigger than you want. Full Kelly on a genuine 5-point edge can mean 10% of your roll on one game.
The reason to take a quarter is that Kelly assumes your win probability is correct. It isn't — it's an estimate from a model, and the error in that estimate is the thing that ruins people. Betting a quarter of Kelly costs a little long-run growth and buys a lot of protection against being wrong about your own edge.
The 3% cap on top is not part of the formula. It's a hard ceiling that catches the case where a bad probability estimate produces an absurd recommendation. If a model tells you to put 9% on a college football game, the model is wrong before the game is.
Feed this the fair win probability, not one you talked yourself into. If your number came out of a model, use the model's number. If it came out of a feeling, this calculator will happily size a bet on a feeling, and it will be precise about it.
Math, not advice. 21+ where legal. If gambling stops being entertainment, call 1-800-GAMBLER.