Oh! I forgot to mention my favorite Gambit. The Traxler-Counterattack.
Don't play this unless you know literally everything White can throw at you. This opening seldom leads to draws.
On the Latvian, it really just depends on what you mean by "unsound".
Is it a forced loss? Maybe not (but also maybe, and I wouldn't be surprised if it is simply refuted eventually).
So if "unsound" means "forced loss", then the Latvian's probably not unsound.
However, if you mean something like having a very poor evaluation in very deep engine analysis or performing much more poorly than the alternatives in practice, then probably yes, it is unsound (although that will depend on your threshold for each metric).
For the former, you're looking at being -1 or worse straight from the opening.
For the latter, the performance rating of players playing the Latvian as black is over 100 points below their nominal rating.
Now, players playing black on average have a performance rating below their nominal rating even in completely sound openings; this we more or less expect. It's just easier to play white in general thanks to the extra tempo.
However, for "normal" openings this rating gap is kept somewhere in the 30-50 rating point range, so having a gap >100 points for the Latvian is a substantial difference.
In engine-engine play I imagine the difference would be markedly more pronounced (I'll probably try to track that down later, or run a tournament of my own).
Of course, the more important question is whether you enjoy playing the position; I regularly play a couple openings that I know full well are objectively inferior, but I just enjoy playing those positions.
Chess is for fun, after all :)
EDIT: Also, that Sokolov game is NOT a correspondence game. It was played in the European team championship, an OTB tournament.
People look at it in a way too scientific way. In correspondence chess (with engine assistance) people are much braver in playing "unsound" lines. Over the board, with rating and money at stake most grandmasters are over-prudent these days. The gambit player often has a huge advantage in knowledge. How many times has the white player ever faced the Latvian Gambit? If he has ever studied it, how long is that ago? What if any does he remember? Does he know about recent games or analysis in it? Meanwhile his clock is ticking and he has to play a move.
But in CC both sides can use engine assistance and unsound line player gets busted.
No in correspondence chess players prove that lines generally considered unsound are in fact quite playable. Most pronunciations of "unsound" are just dogma.
Not sure if it's been played by grandmasters, but it fits your other criteria and is definitely sounder than some of the other gambits mentioned in this thread.
In the lichess masters database, white wins 60% of Latvian gambit games and draws 20% more.
I suppose some brave soul could consider this playable, but not me.
www.youtube.com/watch?v=f_Y6h7JOuiU&tChristof Sielecki seems to think it's refutable.
Hey guys. I'd just like to mention I think there's a big difference between gambits per se and theoretical pawn sacs that are known to be good.
For example, the main line of the Catalan nowadays starts 1. d4 d5 2. c4 e6 3. Nf3 Nf6 4. g3 Be7 5. Bg2 0-0 6. 0-0 dxc4 7. Ne5 Nc6 and one line is 8. Bxc6 bxc6 9. Nxc6 Qe8 10. Nxe7+ Qxe7 11. Qa4 e5 12. dxe5 Qxe5 13. Qxc4. In this line black is considered to be perfectly fine and it has been subjected to a lot of testing at the super GM level. However, no one has ever called it the "anti-Catalan Gambit" or anything like that. It's just a question of names and terminology.
I think the Marshall is the most sound gambit in chess, Queen's Gambit and Réti Gambit not included. Blumenfeld Gambit, Budapest Gambit, Albin, Benko, and other such gambits are probably a little dubious.
I also don't think you should be trying to make a repertoire of only gambits- it probably won't be great for your chess if that's the only type of position you play!
@lovlas Why not the Evans gambit? I am a much weaker player than you, so I would like to hear your thoughts. My impression is that the Evans is actually relatively sound. And it scores fairly well in the Masters database here.
@Pat6578 My impression is that the gambit is usually accepted because the declined lines are good for white. Is this not the case?