Interesting. I haven’t read the article too closely, but the way it converges is interesting.
Clearly I was still stuck in the ye-olde days mindset of ELO where losses essentially had to be directly offset by wins. This seems much more robust, especially since losses from past restarts are (presumably?) no longer in the rating period, and it can just converge from where people’s scores are. (even if I’m still traumatized by accumulating upwards of 20 losses to oracle + zero alone this week )