I'm pretty sure I was part of the "resurgence in people advocating for the use of the minimum time recently," and I'll tell you right now that I did not have a

I'm pretty sure I was part of the "resurgence in people advocating for the use of the minimum time recently," and I'll tell you right now that I did not have a "benchmark" that varied by 7x from run to run in mind. "Benchmarks with random performance tend to mislead" Both of these "benchmarks" have exactly the wrong amount of randomness. A little more, and you could get a useful mean out of the results. When I advocate for the use of minimum time in benchmarking, the scenario in mind is running periodic regression testing on a set of known good benchmarks that don't have intrinsic randomness (or it is averaged out) and yet vary by 0-10% as a result of actual noise. In these cases, it is very useful to have a single number (and if that single number surprises you, then by all means take a look at the distribution). Neither of the examples here show the impact of actual system or measurement noise, only random performance.