In an optimistic approach to security, one can often simplify protocol design by relying on audit logs, which can be analyzed a posteriori. Such auditing is widely used in practice, but no formal studies guarantee that the log information suffices to reconstruct past runs of the protocol, in order to reliably detect (and provide evidence of) any cheating. We formalize audit logs for a sample optimistic scheme, the value commitment. It is specified in a pi calculus extended with committable locations, and compiled using standard cryptography to implement secure logs. We show that our distributed implementation either respects the abstract semantics of commitments or, using information stored in the logs, detects cheating by a hostile environment.