> Popovertarget passes every choice made by the player, for example: x1-o2-x3 means X chooses the first position, O chooses the second and X chooses the third, and so on.

I think this is going to lead to a lot of redundant states, isn't it, because there are multiple paths to get to a particular board layout?

There would be a lot of illegal states in that number.

Of course, that's why I said "maximum". That strengthen my point.

It seems there are only 5478 valid boards.

See my script here: https://news.ycombinator.com/item?id=38266987