Carlos Torre Repetto, 1925
When Adams and Torre played this game in 1920 (or, more likely, when Torre created the conclusion of this casual game from post-mortem analysis with Adams), Torre (born in 1904) was only about 16 years old!
Let's look at moves that don't work. 22.b3 is a very logical try:
After 22.b3?: Black to move
White makes a threat that must be answered, and 22...Qxb3?? loses for the usual reason. But Black can simply reply 22...Qb5!, and I don't see a way to drive the Black queen off the a4-e8 diagonal. For example, 23.Qc4 Qd7! 24.Qg4 Qb5! rewinds the tape of the game, except that White has lost the a-pawn.
22.Qa5?! is a better idea, and it gives White real winning chances:
After 22.Qxa5?!: Black to move
Black's only reply is 22...Qd7, and White can press Black with 23.Qc7! Qb5 24.Rxe8+ (24.Qxb7?? loses for the usual reason: do you see why?) 24...Rxe8 25.Rxe8+ Qxe8 26.Qxb7, with excellent winning chances in the endgame. (The helpful knight on f3 guards the e1 square, so Black has no back rank mate.) But White has better....
After 22.Re4!!: Black to move
22...Qxe4 loses to a version of a trick we saw in the Bernstein-Capablanca game. 23.Rxe4! and Black can't take both unprotected pieces at once. So Black must return the queen to b5:
After 22...Qb5: White to play and win
There are a couple moves that win, but there is only one winning idea!