This study reports on how instruction that is based on engaging students in practical experiments can create challenges and opportunities in the teaching of the relationship between a classical a priori and a frequentist model in estimating the probability of random outcomes. Knowledge is assumed to lie in the inferentialist relationships within the game of giving and asking for reasons (GoGAR). We report on dilemmas (challenges vs. opportunities) faced by the teachers and the researchers who co-designed the tasks: (i) whether it is effective to avoid the elicitation of deterministic reasons for random behaviour or to invite students to reflect on the lack of power of such reasons; (ii) whether the GoGAR is best served by accepting any responses from students or by challenging responses in order to clarify what is normative; (iii) whether the sample space that generates random outcomes should be revealed or not.