Morphing The French XI

Spanton (1951/168) - Seyon Rajeev-Lazar (1300/80)
Coulsdon Christmas U2000/170 Round 2
French Exchange
1.e4 e6 2.d4 d5 3.exd5
My 11th attempt at playing against The French Defence in the style of Paul Morphy.
3...exd5 4.Nf3 Nf6 5.Bd3 Bd6 6.0-0 0-0 7.Nc3 c6 8.Bg5 Bg4 9.h3 Bh5
For 9...Be6, see my game against Alan M Brown (1991/186): https://beauchess.blogspot.com/2019/09/morphing-french.html
For 9...Bxf3, see Morphy - Sicre: https://beauchess.blogspot.com/2019/07/morphy-v-french-part-six.html
10.g4 Bg6 11.Ne5 Nbd7
For 11...Re8, see my game earlier this month against Shlok Verma: https://beauchess.blogspot.com/2019/12/morphing-french-x.html
12.Nxd7 Qxd7 13.Bxf6 gxf6

How should White proceed?

*****
*****
*****
*****
*****
14.f4
This was James Mason's choice in a draw with Adolf Schwarz at Vienna 1882.
Stockfish10 and Komodo10 prefer 14.Qf3!?, which was played by Mikhail Chigorin in a loss to Max Weiss in a later round of the same tournament.
14...f5 15.g5?!
15.gxf5 Bxf5 16.Bxf5 Qxf5 17.Qg4+ Qg6 is also a little better for Black, according to the engines.
15...Qc7
Mason - Schwarz saw 15...f6 16.h4 fxg5 17.hxg5, which the engines reckon is good for Black, especially after their continuation of 17...Qg7 (Schwarz played 17...Rae8), eg 18.Ne2 (or 18.Qf3 Qxd4+ 19.Kg2 Rae8) Bh5 19.Qd2 h6.
16.Qf3 Qb6 17.Ne2 Rab8?!
Played so b7 is not hanging after ...Qxb2, Rfb1, but it is slow. Black now gets a kingside initiative.
The engines prefer 17...f6 18.h4 fxg5 19.hxg5 Rae8.
18.h4 f6?!
18...h5 looks better.
19.h5 Be8 20.Qg3?
The simple 20.Bxf5 is strong, as is 20.g6, eg 20...hxg6 21.Qg2 Kg7 22.hxg6, when 22...Bxg6 loses to 23.Bxf5.
20...Kh8?
Much better is 20...Bxh5 21.gxf6+ Bg6.
21.g6
Black is lost.
21...Rg8
Or 21...hxg6 22.hxg6 Bxg6 23.Kf2 etc.
22.Bxf5 Qxb2
This does not help, but the engines cannot find a decent move for Black.
23.Kh1 Qa3 24.Qh4 Be7 25.gxh7 Rf8
Or 25...Rg7 26.h6 Rxh7 27.Rae1 with a huge attack.
26.Qg4 c5 27.Rg1
The game finished:
27...Qh3+ 28.Qxh3 Bd7 29.Qg4 Rf7 30.Qg8+ Rxg8 31.Rxg8#
Addendum 24/12/19: My opponent's 1300 rating turns out to be a converted ECF.