The ambitions of behavioural science are, in principle, (0) CIRCUMSCRIBED by the very phenomenon they seek to model. Unlike the particles of physics, which submit indifferently to whatever laws happen to govern them, human beings stand in a reflexive relation to the theories proposed their conduct. A well-publicised finding about, say, the effect of default options on pension enrolment does not simply its object: it alters the informational environment within which future enrolments take place, and so reshapes the very it was intended to describe. This is not a marginal complication but a feature of any science whose subjects can read its journals. The methodological that follow are considerable. Replication, normally the final arbiter of a finding's robustness, becomes complicated when the original study has itself changed the population against which replication will be attempted. Behavioural economists now speak openly of *finding decay* — the tendency of once-robust effects to as awareness of them spreads — and the profession has begun, somewhat belatedly, to treat its own influence as a variable rather than a nuisance to be ignored.