Good point, thanks. The way I modeled the adjustment was by assuming that most usage is captured by Statcounter but there’s movement back and forth to a reservoir that flies under its radar, in bursts, with zero net movement in the long run. So I used a geometric mean of the source data scaled by the square root of their averaged ratio.

Today’s a dear friend’s birthday, and I’ve got a really cool present for them!

Getting an initial impression of some new field I want to learn about. I ask the model for a short summary and links to more in-depth information. This would be more difficult to do on my own when I don’t even know where to start.