I read the docs on setting capped values, but I’d love to know how people approach setting a cap. Are you reviewing the distribution of values per user and setting based on the 95/99% value, for example? Many of our metrics are skewed, with most values low and a long power law tail to the right.

Yes, it's usually done with percentiles. 99% is pretty common

Thanks, that makes sense. We produce quantiles of a few of these metrics so I’ll take a look at that and start with 99%.