This paper studies an agency framework in which a principal hires a supervisor to monitor the agent's productive effort. We consider several monitoring technologies which differ in the quantity (frequency) and the quality (accuracy) of the information they deliver. We show that the frequency of monitoring is irrelevant if the supervisor is honest or if the supervisor colludes with the agent but monitoring evidence can only be concealed and not forged. In either case, a first-best can be achieved if monitoring is sufficiently precise even though unbounded punishments are not feasible. Finally, if monitoring evidence can be falsified, the principal benefits both from the frequency and the accuracy of the supervisor's observations. This is the only case in which collusion imposes an additional cost on the relationship. The findings suggest that it is strictly better for the principal to monitor the agent's action rather than testing for an unknown ability or technology parameter.
