We study an expert-advice problem with career concerns and a continuous private signal about a binary payoff. A principal implements safe advice for sure and risky advice with a probability that increases with the expert's reputation; realized outcomes, when available, update reputation. We show equilibrium advice follows a cutoff in the signal and, under a relative-diagnosticity condition, the cutoff increases with reputation (reputational conservatism). Comparative statics are transparent, and simple policy levers - a success-contingent bonus and gatekeeping - provide implementable control of experimentation.
Lukyanov et al. (Thu,) studied this question.