Makes sense. I think that the version where you have a 1/n chance of checking all potential values of r runs into some practical problems, in that very long single instance running time might alter A’s evaluation of the situation (if B is a human they might get tired and make mistakes in their evaluation, there could be a high chance of some external event disrupting the evaluation, or the running time could be so long that B believes that A could not credibly carry out the evaluation). But it seems simpler to use in analyzing the other features of the problem.