I analyzed it some more.... lol
If Kc7 is met with b6+, leaving the white Bishop on a7, white can draw with Bb8 later.
Then you'll have this position
With Bb8, it protects the pawn the Rook would have captured to win. Black can't win without the Rook, so sacrificing it to make a pass pawn doesn't work. Tf the white Bishop is captured by the rook, then we have a position as in the original video where the King makes it to h1 and Black cannot make progress without stalemate. There's also no way to leave the King in a position of capturing the white Bishop, because there are limited squares for mobilizing the bishop and rook. Doing it any other way gives tempi to White, allowing the King to get to h1. Furthermore, you can't park the Rook on g2 in hopes of distracting the white Bishop later on, or forcing zugzwang, since it has too many outposts along the h2-b8 diagonal to hang out on.
b6 is the really important move, locking all the pawns.
Turns out it's a stalemate after all! Though I wish Simon would have analyzed that line.
I just watched the second video of Simon Williams' series on how computer misunderstand certain positions and evaluate them incorrectly. The second video deals with more complex positions involving blockades where none of the attacking pieces can penetrate the opponent's position, or there are stalemate tricks that the computer cannot see.
One position Williams says he discovered through a Chess.com member's suggestion. It's as follows:
Williams says it appears that black is winning with the following strategy: move the bishop (Bxg4) and move the King so the Rook can enter the game on g2 via g8. The 'winning' move by white, which is actually the drawing move is Ba7+ followed by b6. Williams describes that either the King or the Rook can capture, and while black executes the plan as described above, the white King runs over to the h1 square. By the time the Rook gets to g8, the white King will be trapped on h1 ending in a stalemate.
However, capturing is not necessary. The black King can dodge the check with Kc7, and even gain a tempo in executing the plan. The Rook makes it to the G file in time to cut the white King off from the h1 square in this line.
I couldn't really analyze it properly with an engine, because I don't trust that the engine understands the position, which is what the whole video series is trying to teach (c7 was its third option after capturing with Rook or King) and so I was left to analyze it on my own. But since I'm an amateur player, I thought I'd post it on here and get other people's opinions. I don't think this is one of those complex blockades that computers don't understand. I think it really is winning for white.
Let me know what you think.