(Based on joint work with Charlotte Werndl)

This talk questions the significance of common intuitions regarding the importance of ‘use-novelty' for scientific prediction, as cashed out in the ‘no-double-use rule’, which says that data used in constructing (or calibrating) a theory/model cannot also play a role in providing support for the model’s predictions. The strategy is to analyse the intuitions in light of prominent logics of evidential support. Here the focus is Bayesian logic vis-à-vis Classical model-selection logic. On both of these accounts, the use-novelty intuitions are too crude, but for different reasons: one may conclude that either the scope or the simplicity of the intuitive position must be revised. Whether one particular revision should be favoured is a topic for further debate.