The first game is one of the Master games, GoGoD 2017-01-03a. White was Yang Dingxin, 5p.
Yang's big mistake, says Elf, is , a play that we humans had thought was good. As an extension it is a little far from the stone in the bottom left corner, but it aims at an extension to a, which makes a base and threatens the thin Black corner. (White actually extended to a in the game.) Elf reckons as a hefty 19% point loss. Elf normally devalues early plays on a side, but not to this extent. The reason, I think, is that White had the alternative of the solid connection at 11. If White plays there Elf estimates Black's winrate at 48% with 120.2k playouts. was not on Elf's radar at all, with 0 playouts. So Elf estimates Black's winrate after at 67% with 28.9k playouts.
After the solid connection, Elf as White would extend on the 4th line. Often bots play elsewhere after the solid connection, but not in this case.
To get to a White winrate estimate of only 10½%, after White made a number of small errors.
The largest of these was the jump attachment, , instead of the nose attachment at a. The nose attachment has a Black winrate estimate of 68½% with 56.7k playouts. was scarcely on Elf's radar, with only 34 playouts. Elf awarded it the Black winrate estimate after of 77% with 29.9k playouts. lost 8½%.
=
This is Elf's main line after the nose attachment. White forms a small group on the top side and makes shinogi in the bottom left corner. :)