15.O-O Qe8 16.Qh4+ Ke6 17.d5+ Kxd5 18.Nc3+ Ke6 19.Rf6+ Ke7 [19...Nxf6 20.Qxf6#] 20.Nd5+ Kd8 21.Rf8+ Ne7 22.Qxe7#

or

See C37 in Modern Chess Openings for details. The soundness has been proved in several games, following 5...Kxf7 6. Ne5+, as shown below

Instead of 6...Ke6 (clearly the weakest continuation), Black can also choose Ke8, with the following continuation:

7. Qxg4 Nf6 8. Qxf4 d6. White has three major options now:

a) 9. O-O (A Nava vs Carlos Torre, 1927 0-1)

b) 9. Nf3 Qe7 (or ...Nc6, see Staunton vs NN, 1846 1-0) 10. Nc3 Nc6 (N Lelen vs K Marzec, 1991 0-1; 10...Nbd7, see Al vs K K Hantouli Firas, 1996 1-0)

c) 9. Nc4 Be6 (Elliot vs Roper, 1903 1-0), or 9...Qe7 (De Villette vs Maubisson, 1680 1-0), or 9...Rg8 (S Shaw vs P Sokol, 1943 0-1)

http://es.youtube.com/watch?v=LDL02...

Does anyone remember the cover?

It felt good to mete out some pain

In the desert, there's plenty NNs with no game

But there ain't no one who plays chess so lame

If he had lost, or if the combination had turned out to be unsound, then I conjecture that he would *not* have saved it.

As I understand it, the game scores were published by wealthy patrons as instruction material for other chess players. The idea of keeping historical records was not yet firmly established.

So the games that we have on record may or may not have been played for real. And if they were played, we don't know if the moves have been "tidied up" - either to remove superfluous moves or to show a spectacular variation that was not actually reached in the game.

Wikipedia and other sources talk about Greco publishing analysis in the form of games. I don't think we should censure Greco for this - it was the usual custom at the time - but nor should we get too excited about a brilliantly played game by Greco or awful defence by NN.

As <sleepyirv> rightfully says, we will probably never know the truth.