It's pretty big - between a few hundred GB to maybe 1 TB depending on how far back you want to go - but it would make the generation process much faster for users who downloaded it because there wouldn't be any API rate limiting. It would also allow you to do the multi-move lookahead search ("The most obvious improvement would be something which looks deeper into the tree so that we calculate the win rate assuming we play only the lines we play (right now the win rate includes lines we don’t intend to play). That said I couldn’t figure out how to do this without hitting API limits")
Lichess allows you to download their databases from https://database.lichess.org/
It's pretty big - between a few hundred GB to maybe 1 TB depending on how far back you want to go - but it would make the generation process much faster for users who downloaded it because there wouldn't be any API rate limiting. It would also allow you to do the multi-move lookahead search ("The most obvious improvement would be something which looks deeper into the tree so that we calculate the win rate assuming we play only the lines we play (right now the win rate includes lines we don’t intend to play). That said I couldn’t figure out how to do this without hitting API limits")