The puzzle of the self-torturer raises intriguing questions concerning rationality, cyclic preferences, and resoluteness. Interestingly, what makes the case puzzling has not been clearly pinpointed. The puzzle, it seems, is that a series of rational choices foreseeably leads the self-torturer to an option that serves his preferences worse than the one with which he started. But this is a very misleading way of casting the puzzle. I pinpoint the real puzzle of the self-torturer and, in the process, reveal a neglected but crucial dimension of instrumental rationality.

The justification of the reasonable doubt standard has been hotly debated in recent years. Deontologists—including retributivists—have generally defended the standard, whilst consequentialists have generally argued for a lower standard of proof. Captivating arguments have been produced from both sides. The paper narrates this debate through a dialogue between ideal representatives of these different camps. In doing so, it recasts—and, hopefully, improves—some of the arguments presented thus far. Then, the paper introduces a new participant in the debate, the Intermediary. The Intermediary is under the impression that the debate has reached an impasse, due to fundamental moral disagreements between the parties involved. Therefore, she presents them with a challenge: to find a common ground that allows the parties to justify to each other the choice of a standard of proof, notwithstanding their different basic moral commitments. The Intermediary takes up this challenge, and provides a justification for the reasonable doubt standard based on the value of respecting defendants and on rules of instrumental rationality.

Kant famously distinguishes between the categorical imperative (CI) – the fundamental principle of morality – and hypothetical imperatives (HIs), which are instrumental norms. On the standard reading, Kant subscribes to the ‘disjunctive reading’ of HIs, which takes HIs to be consistency requirements that bind agents in exactly the same way whether or not agents are subject to CI and whether or not they conform their choices to CI. I argue that this reading cannot be squared with Kant's account of an agent's disposition, in particular his claim that cognition of CI is a necessary condition of willing a maxim. I further argue that Kant could not accept an account of HIs as consistency requirements. Finally, I outline Kant's conception of HIs as non-disjunctive requirements that arise when and only when agents will permissible ends. This account can help recapture Kant's conception of the unity of rational norms.