Clearly, Black was looking for complications.

13...Bg4 is in fact a blunder that leads to a losing game. 13...Rg5 is an improvement, which leads to a melee: 14.Bxd4 Rxd4 15.c3 Bxc3 16.bxc3 Rg5 17.Qe3 Qxc3+ 18.Bc2 Qxe3+ 19.fxe3 Rxg2. Black emerges with 4 pawns for a knight, but owing to having to develop his unmoved pieces probably no more than an even game. But Black had possibly an even stronger move - or at least one that offered better prospects for an edge: 13...Nf5(!) with this possibility: 14.Qf3 Re8 15.Bg5 Qe5 15.c3 ... At this point Black might go for the piece sacrifice: 15...Bxc3 16.bxc3 Qxc3+ 17.Bc2 Qa1+ 18.Kd2 Qxa2. This leaves Black with FOUR passed pawns on the queen side for the piece, but a rather under-developed game.

It is possible that Capablanca had assumed White would take the Exchange, here, but after 14.fxe3 Ne2+(! - 14...Bxd1?? 15.Rf1!) 15.Bxe2 Bxe2 16.Re4 Bc5, Black has the bishop pair to match White's rook and knight. But White has better...

Possibly 15...Rxd3 would have given away less than 15...Ne2+ has, but, at this point Black is almost completely busted: exchange down for a pawn. I say 'almost', for White still has to work for the win!

Centralising. Black's bishop is flapping in the breeze, rather, with (for the moment) no really good squares.

The thing with the Q exchange, is that the slight disturbance to Black's Q-side structure might have sweetened the deal for White. Maybe White was hoping (a) for a 'middle game' win, or (b) that there would appear a more favourable moment for exchanging.

Already a large chunk of White's advantage has been squandered. Black has been given a tempo to strengthen his presence in the centre. The move f4-f5 could have waited.

In going for a K-side attack, White has adopted a mistaken plan. Possibly 28...Kh7 wasn't even necessary, as 28...d4 offered prospects of counterplay in the centre. Already Capa must have been confident of at least 'saving the game'.

That White is on the defensive indicates (to me) that his earlier advantage has disappeared altogether.

Now Black goes for a flank attack!

Not so long ago I won a game with just such a tactic...

The game is taking on the features of games with 'opposite wing castling'. Such games can lead to races to see who can get his licks in first.

38.Rd2xd5 was a mistake that handed Black an edge. To stay 'in the race', White probably had to try 28.h4(!?) Bxh4, decoying the bishop momentarily from the dark long diagonal, and then 29.Rxd5. Now Black's attack is looking dangerous.