Closed rkothari19 closed 6 years ago
We have to be careful about this, as a change in the median sample for any particular measure might not indicate a regression (since there are so many causes of natural variation for these numbers). But yes, for unambiguous cases we should do this (or something like it).
@mreid suggests maybe looking at cerberus for inspiration on how to detect that there is a significant, sustained change.
Right now we use color coding on the text in the main page to show significant (> 25% changes) in the mean for any particular measure relative to the previous release. I'm pretty happy with the way this turned out -- it gives a hint on things to look at without blazing alarm bells.
I'm going to close this for now. We may well want to tweak this further but I think we can open a new issues for that.
Color coding to show whether “latest version” is better or worse than “previous version (avg)” https://data-missioncontrol.dev.mozaws.net/#/release/windows