In an adverse-selection context, we use an experimental protocol to study the intensity and efficiency of a control policy as filtering instrument in the long-term financing relationship between an investor (principal) and an entrepreneur (agent). We find three control effects : a disciplining effect, an educative effect, and a filtering effect. While the first is expected, the educative effect of control gives way to a crowding-out effect already observed in other tests but in a moral-hazard context. Lastly, our experiment shows that the principal restores a filtering effect through a trade-off between control efficiency and control intensity : it is better to control little in order to detect a little, rather than control a lot and detect nothing.