The game starts out innocently enough. (1.c4 and 2.b3)

It's only upon really absorbing the position that the underlying themes become apparent. Rf4 will gain a tempo on the queen, allowing the other rook to join in on the game. Qh5+ is possible, and the pressure on f6 makes ...Kg8 difficult. Thus black's king is stranded in the centre. Now, the possibility of the white e- and g-pawns running up the board makes the kingside rather hazardous, so where is the black king to live? Velimirovic opted for the queenside, only to meet with Tal's 21.b4!

Instead of 20...Kd6!, Kasparov points out Black could have played 20...Kf7.

Also, after 22.Rac1, Dvoretsky suggests 22...Re8 instead of the move played, 22....Rc8.

Even so, after 23.Rf5, Kasparov indicates that after 23...Nd7! (instead of the move played, 23...Qg4?!) 24.bxc5 Kb8 "the Black King would have been assured of escaping from all dangers, and the extra Knight would have told."

Mikhail Tal - Dragoljub Velimirovic 1-0 0.0, YUG-URS Teslic BIH 1979

click for larger view

Analysis by Stockfish 14.1 3 cores only:

Something must of went pretty wrong if white can be both better developed and arguably better in the center even after spending the time to double fianchetto.

After the sac, it's very, very likely going to be either a won or lost game. The attack either works or a proper defense wins. Of course, SF is going to find a proper defense if it exists. The question is how much is white's compensation or chance black will miss something?

The evaluation for that is the level at move 23...Qg4. Black's up a piece less a pawn. So I'd surmise unless the evaluation is at least around -2.00, that would indicate that white has something for the sac. The more toward 0.00 the more he has. And I would guess the greater chances a human defense would falter.

I don't think SF works any more by counting pieces and then evaluating. But anyway, the eval is about -3.28 at 42 ply at this point. Main line is 23....b6 24.Rf4 Qc6 25.bxc5 Bxc5 26.d4 Ba3 28.Bb2 Qxc1+ emerging with rook and two pieces for the queen.

SF doesn't like Black's opening any more than you (or Kasparov or Dvoretsky, whose notes are quoted extensively in OMGP II). At 12....e4 the evaluation is +1.63 at 38 ply, but the computer prefers 13.Ne1 and 14.d3. It doesn't like most of the next few moves by either side, but it thinks either 15.d6 or 15.f3 Qxg5 16.Nxe4 Nxe4 17.fxe4 with a very weird position would have been better than the knight sacrifice in the game.

Related to computer evaluation, I once received a summary of how Deep Blue came up with its calculation from the machines team. If remembering correctly, there were four or five main factors. Mobility, King Safety, maybe Space, and of course Material. So when there's a game material imbalance, my starting point is that if all other factors being equal, the evaluation should equal the material imbalance. If there's a difference, then I try to figure which other factors are in play to adjust that number.

Now I'm assuming that engines today, except those based on experience Monte Carlo evaluations, are still generally based on the same factors concept for the evaluation function. Though understanding today's versions are probably much more robust and fine tuned.