ClickHouse
Feature request: filter out mirrors data (maybe even by default)?
Is your feature request related to a problem? Please describe.
For packages that are not downloaded much, most downloads come from mirrors and not pip or uv installs. This can give a wrong picture when looking at the numbers themselves or the timeseries of number of downloads per month. For example for skore pip and uv downloads are roughly 30% (i.e. roughly 70% are from mirrors or other installer types), see https://github.com/hugovk/top-pypi-packages/pull/41#issuecomment-2771919920 for more details.
pypistats.org filters out mirrors data from their aggregated numbers, see their FAQ.
Describe the solution you'd like Maybe:
- a check box in the project dashboard to only take into account
pipanduvinstallers (maybe a few others not 100% sure but those are the main 2 I would say) and exclude the rest - or by default downloads from mirrors are excluded and a check-box to add them (this would be my preferred option but more disruptive)
Describe alternatives you've considered Keep the status quo, numbers are higher, which you know for vanity metrics may be considered as a good thing 😜.
Additional context
I am now realising that the "filtering out mirrors" feature was asked in https://github.com/ClickHouse/clickpy/issues/87 and closed with a link to https://github.com/ClickHouse/clickpy/issues/109.
#109 is about filtering out CI downloads (not mirrors). Not sure whether this was closed on purpose, but I am guessing that similar considerations need to be taken into account (backfilling is too expensive) as mentioned in https://github.com/ClickHouse/clickpy/issues/109#issuecomment-2553346597.
@lesteve yes exactly, we can do but only for data moving forward cc @lio-p
