A random thought on download karma... Merging versions with downloads is a bit tricky, right ? I mean, when you push something to extras, you will certainly see a surge in downloads from people who are updating in addition to those who are installing for the first time. This means projects updating more often will certainly see a higher number of downloads even with the same userbase...

... and now, the kicker...

What if we use these spikes as an additional metrics in some form ? After all, the volume of the spike is the approximate numbers who still have it installed. So it is possible to measure uninstalls, even if in an indirect form. Yes, I know, some people never update, but the point is that the metric will be equally wrong on all projects, so you do get a realistic comparison of userbases. Note that this wont work if you update software TOO often as the update dispersal can take a while.

See this graph from maemo mapper - you see a steady stream of downloads, but the spikes containing the updates are those which show the real userbase increase.

EDIT: I'm not saying this should replace download count, but could be something like the t.m.o. posts vs thanks karma. Download karma (=posts) is cheap, update karma (=thanks) is value.