Answering your last question first, yes, ratings are more uncertain at any robustness when FargoRate doesn't know much about your opponents.
But in "typical" situations, where we know most of your opponents but some are more unknown, the situation is about as follows:
Let's say your rating is 500, and it is based on X games. We want to say there is a two-thirds chance your actual skill is within U points of 500
U is about 283/sqrt(X)
So when X=200, U=20 points
When X=800, U=10 points
When X=50, U=40 points
The players in my local group had a little over 100 games on average, and we expect the preliminary ratings are probably within 30 points. And the graph relating to my judgment kind of aligns with this.