Finding a good bet is only half the job. The other half is how much to stake. Bet too little and you leave growth on the table; bet too much and one bad run wipes you out. The Kelly criterion is the staking rule that maximises long-run growth given your edge.
The intuition
Kelly ties your stake to your edge and the odds. Bigger edge → bigger stake. Shorter-priced outcome (more likely) → you can risk more of the bankroll; longer-priced → risk less. The output is a fraction of your bankroll, never a fixed amount and never everything.
For a simple two-way bet the fraction is roughly (edge ÷ net odds). A 10% edge at even money says stake ~10% of the bankroll; a thin 2% edge says stake ~2%. No edge, no stake.
Why not just bet the same amount every time?
Flat staking ignores how good each bet is — you'd bet the same on a coin-flip edge as on a screaming one. And betting a fixed large amount invites ruin during the losing streaks that variance guarantees. Kelly scales with conviction and protects the bankroll at the same time.
Fractional Kelly — the safety margin
Full Kelly is aggressive and assumes your probabilities are exactly right. Since no model is perfect, most disciplined bettors use a fraction of Kelly (e.g. half). It gives up a little growth for much lower swings — a sensible trade when your edge estimate has uncertainty.
How Momus uses it
Momus sizes every bet with a fractional Kelly on its measured edge, then caps it — a maximum per bet and a maximum total exposure. Big edge, bigger stake; thin edge, small stake or a pass. You can see the sizing on each call on the track record.