So you want to repeatedly sample from some collection, but you want each sample's size to be randomly determined. For example, you might want sample sizes to range from 1 to 100, uniformly distributed, as you collect confidence interval widths.
The solution is to use an expression in the "until" box of the Sample panel. The desired expression for the situation described above is:
If you use the expression
you will get sample sizes whose values are drawn from a geometric distribution whose mean is
You can generalize this idea to work with any probability distribution, p(n) = the probability of choosing the number n. If we have already chosen n - 1 cases, then the probability of stopping is