[grugagag]

Why does it have to be a rate? Failures counts and failure types are good enough to extrapolate whatever else you need

[afavour]

They really aren't. If a package is downloaded 100k times and fails 1k times it's a very different issue than if it's downloaded 1k times and fails 1k times.

[storoj]

IMO installation failures are equally bad, no matter if they happen in 1% of cases or in 100%.

[afavour]

Well, you're wrong. If you have 10,000 packages and 1,000 of them are failing it's clearly useful to be able to prioritize which to fix first.

[cratermoon]

https://www.trevorbragdon.com/when-data-gives-the-wrong-solu...