[LtWorf]

Oh what a coincidence. I wrote pypular (https://codeberg.org/ltworf/pypular) to artificially increase download counts on PyPI (talked about it at a MiniDebConf).

I was claiming that there are important incentives to play this sort of games.

Seems OP proved me right!

[kvinogradov]

In your opinion, which metrics could be better than # downloads to evaluate PyPI packages at scale?

[LtWorf]

# of downloads from unique IP would be better for sure.

Also if you have a local cache and just check the version, that won't count.