How many traces do you need before you're likely to catch a rare failure once?

Multiply the failure rate by the share of sessions you review. That product is the chance any one session both fails and gets looked at, and it is what sets how long you wait.

The formula is N = ln(1-c) / ln(1-p·s): c is the confidence you want, p is the failure rate, and s is the share of sessions you actually review. As a hypothetical, take a failure rate of 2% and a 5% review sample. Then p·s is 0.001, one session in a thousand, and ln(0.05) / ln(0.999) comes to roughly 3,000 sessions before you’re 95% likely to have caught it once.

Push either number down and the wait gets much longer. Halve the review sample and you need about twice as many sessions for the same confidence. Stratified sampling raises s inside a slice you deliberately target and leaves it where it was everywhere else.