At Paddy Power--Ireland's largest bookmaker--teams of quants and risk analysts set the odds on 12,000 to 15,000 events a week--everything from horse races and other sporting events to speculation on the name of Beyoncé's unborn child. Within these events, there are 60,000-70,000 individual bets, or "markets," to be made. And every market needs a set of odds--some kind of calculation of the probability that a specific outcome might occur, based on available data. But how does a bookmaker know what data is good and what data is bad? How can it build safeguards into predictive systems so it doesn't get burned?