30 New-Years In Hastings

AM back in Hastings for the 30th consecutive time of playing in the congress over New Year.
Spanton (1954) - Holger Witt (1567)
Hastings Masters, Round One
1.d4 e6 2.e4 d5 3.Nc3 Bb4 4.e5 Ne7 5.Bd2 c5
By two minor transpositions, we have reached the starting point of the Bogoljubov Variation of the French Winawer.
6.a3
This virtually forces the win of the bishop-pair, but more popular is 6.Nb5, when the knight might go to d6.
6...Bxc3
Just about possible is 6...Ba5?! but White scores massively with the simple 7.dxc5.
7.Bxc3 cxd4 8.Bxd4
The mainline sees White keeping the bishop-pair with 8.Qxd4 Nbc6 9.Qg4.
8...Nbc6 9.Nf3 Nxd4 10.Nxd4 a6!?
This may be a new move.
Three previous games of mine featured 10...Nc6, with me scoring a win and two draws against lower-rated opposition.
11.Bd3 Nc6 12.Nf3 Bd7 13.0-0 0-0 14.Qe2 b5?!
White must be better anyway, but further shutting-in Black's light-square bishop is very unlikely to be right.
15.b4!?
Clearly double-edged as the c pawn is now weak.
15...Qb6

Has Black allowed a Greek Gift sac?

*****
*****
*****
*****
*****
16.c3
Definitely not 16.Bxh7+?? Kxh7 17.Ng5+ as the Black king can come out at g6 or h6 without putting itself in serious danger. White's rooks are a long way from helping in the attack, and the e5 pawn is weak.
16...h6 17.Rfd1 Rac8 18.Rac1 Rc7
My main analysis engines quite like the odd-looking 18...Be8!? Komodo9 for some time assesses the position as almost equal, but Stockfish10 reckons White has the upper hand. However, the engines' evaluations eventually merge at slightly better for White.
19.Bb1 Rfc8
The engines prefer 19...Ne7, but with White for choice.
After the text, both players' pieces are largely bunched on the queenside, but White can more easily switch to the other wing.
20.Qd3 g6 21.h4 Be8
21...Ne7, 21...Rf8 and 21...h5 are all suggested by the engines, but in each case with White having a substantial advantage.
22.h5 Ne7 23.Nd4
Even stronger seems to be 23.Qd2 Kg7 24.hxg6.
23...Nf5
The only hope, according to the engines.
24.Nxf5?!
I seriously looked at 24.hxg6 with the idea of 24...Nxd4 25.gxf7+, which is indeed very strong, although 24...fxg6 keeps the game going. In the end I settled for a line that is easier to calculate but is almost certainly not so good.
24...gxf5 25.g4
25.Qg3+? does not achieve anything. Indeed, it invites later trouble down the g file.
25...Rc4 26.gxf5 Rf4??
A blunder that loses immediately.
I expected 26...Rg4+ 27.Kf1, when the engines give 27...exf5 with advantage to White after 28.Qxd5 or 28.Qxf5, but still with lots to play for.
27.Qg3+ 1-0