Statistical modeling is the art of connecting abstract models with scientific or practical questions, so that the models’ probabilities throw light on these questions. Standard statistical theory summarizes the results with p-values. Unfortunately, p-values are widely misunderstood. Game-theoretic statistics replace them with e-values, also called betting scores. The model makes a forecast, the statistician makes an imaginary bet against it, and the value is the factor by which she multiplies her stake.