More reliable user stats

I originally posted this for Site Feedback on ask.fp.o but am posting it here instead since it’s relevant to both and the new home for it will be here soon.

This might or might not be worth fixing, but the weekly user stats seem to run some time daily roughly between 9am-2pm PST. Sometimes it runs close to the day before, sometimes it doesn’t. This inconsistency means that if a user consistently logs in at noon and does meaningful activity, it may or may not count for metrics for that day, which can impact TL levels and metric accuracy in general. Another scenario is that someone contributes in the morning one day, in the evening the next, and then in the morning the third day. They have, from the user perspective, been active on all three days, but the second day won’t have been counted as that activity is grouped with the third day collection.

Is it possible to #1 have the metrics count daily and account for the user’s time zone (ideal, but probably most complex) or #2 at least have the weekly metrics run at the exact same time every day? I’m not sure if most users are in the US or not, but having this collection happen at some point between mid-morning to early-afternoon seems to particularly disadvantage metrics for US users, though I don’t want to similarly penalize users in other localities and would rather find a way to make the metrics more accurate in general, especially the weekly one since the inconsistency can adversely impact TL levels.

EDIT - The times I gave were what I observed over the past year. Today, however, ask.fp.o aggregated at 9:42am (PST) and discussion aggregated at 6:04am (PST), so when it fires during the day seems to be getting less consistent as well.

1 Like

Hmmm. I can ask about this upstream. It seems like even if there’s a reason to run at random times daily, it could at least cover a consistent timeperiod, which I think would address that problem.

It’s mostly “days visited” that this impacts, yeah?

Yes. That’s the one that I think matters the most because it potentially impacts certain badges and trust levels. As a bonus, more consistency here would also mean better usage metrics in general.

1 Like

Discourse staff tell me that each run actually looks back over the last ten days, so if there is a discrepancy due to timing one day, it should get fixed eventually.


Thanks for checking on that!