The core result
The Kelly criterion gives the bet fraction that maximises the long-run growth rate of your bankroll. For a bet paying b to 1 with win probability p:
f = (bp - q) / b, where q = 1 - p
Why over-betting is catastrophic
The growth-rate curve is asymmetric. Betting half of Kelly gives you about 75% of the optimal growth rate. Betting twice Kelly gives you a growth rate of exactly zero - you have an edge and you make nothing.
Beyond twice Kelly, an edge becomes a losing strategy. You can be right about the market and still go broke through sizing alone, which is the single most important practical lesson in the topic.
Why practitioners use fractional Kelly
Three reasons, all of which point the same way:
Your edge estimate is wrong. Kelly assumes you know p and b exactly. If you overestimate your edge by a factor of two, full Kelly on the estimate is twice Kelly on reality - zero growth. Halving your size protects against exactly this.
Drawdowns. Full Kelly experiences a 50% drawdown with high probability over a long enough horizon. Very few people can hold a position through that, and forced liquidation destroys the argument.
Log utility is a strong assumption. Kelly optimises the log of wealth. If your actual preferences are more risk-averse - and most people's are - the optimum is lower.
Half Kelly or quarter Kelly is common in practice.
Sizing under uncertainty
The general principle worth carrying: uncertainty about your edge should reduce your size, not increase it. Candidates sometimes reason that an uncertain edge deserves a larger bet to "find out". That is backwards - information gathering and position sizing are separate decisions.
In the interview
When given a favourable bet, name a size and a reason. "I'd bet about 10% of my bankroll - Kelly says 20%, and I'd halve it because I don't trust my probability estimate to better than a few percent" is an excellent answer.
Practise in expected value.