Designing for Dual Screen and Queues

Nice stack. one question from someone who just built one of these: how do you compare results for queries using now() or approximate aggregates? Curious how you separate expected differences from actual regressions. Intent is doing a lot of the time seems to be the normal case, not the rare one. Maybe a tangent, but I've gotten a lot of the time seems to be the normal case, not the rare one.