Hacker News Reader

I suppose fitting a decaying function would be better than flat-out ARPU?

Or even a moving weighted average?

We're going to explore some of those approaches in upcoming posts!