ELF OpenGo weirdness

Here’s another position from 1. r16 d17 where ELF would play 2. c16 even at kind of low playouts. It does it again in the LR corner (different order), encloses micro in the UR corner, kicks the high approach in the LL corner, then goes all AlphaGo with move 25. New meta or neural network gone wild? You decide.

EDIT: after 400k playouts the winrate gradually drops 4% and ELF starts to prefer 2. d3 without the c16/c17 exchange. It’ll still play it afterwards, however.

EDIT 2: 1 million playouts. c16 became the top choice again.