Yes, I'd play <39....Rxd5> but just because it's the most logical move to play at hand. We discard an outposted piece for a rook that's otherwise just looking at said piece, we increase the value of our fiancetto'd bishop because that's stronger than the opposing rook.

After 40.exd5 Qxd5+ with d3 Black's winning, but I also saw that <40.Rf1!> is playable forcing <40....Qe6> meaning that from the get go White has a few options.

<<mel gibson> I saw the text move in under 3 seconds however Stockfish is calling it a draw>

I had the same experience. It was obvious to me that 39...Rxd5 was the only way to push the attack and, like <FSR> indicated, in situations like this, "Sac first, think later". So I would have played 39...Rxd5, expecting 40.exd5, without bothering to do any deep (to me) calculations. I simply would have thought that I had no choice and, if 39...Rxd5 led to an eventual loss, so be it.

FWIW I use Stockfish 11, the latest "official" (non-development) version release. I don't know if you also use Stockfish 11. And, FWIW, "my" Stockfish 11 doesn't necessarily agree with "your" Stockfish evaluations, at d=40 it evaluates both 39...Rxd5 and 39...Qe6 at [0.00], with its 3rd best move, 39...Rd7, evaluated at [+0.75], slightly better for White.

But, as <whiteshark> and "your" Stockfish pointed out, 40.exd5 is not White's best reply, the zwischenzug (sorry, I couldn't resist, I get so few opportunities to use that word) 40.Rf1 was better.

After 38...Rxd5 I get different results with "my" Stockfish than you got with "your" Stockfish.

click for larger view

At d=40 "my" Stockfish's top 2 lines (its "top" 3rd move, 40.Qa2 evaluated at [-18.30] need not be considered further since it loses a piece after 40...Rd7):

1. [-0.46]: 40.Rf1 Qe6 41.exd5 Qxd5+ 42.Kh2 e4 43.a6 d3 44.Qe1 d2 45.Qe2 Bd4 46.Qg2 Qa2 47.a7 (I don't see the purpose of this move. Was it that important to divert Black's bishop from the a1-h8 diagonal?) 47...Bxa7 (Material is now effectively equal, R vs. B+2P, but Black's 2 far advanced passed pawns give it the advantage) 48.Rd1 e3 49.Qb7+ Kh8 50.Qe7 Qb2 (50...Qc2 seemed stronger to me but Stockfish indicates that it leads to a draw by repetition after either (51.Qd8+, 51.Qe8+, or 51.Qe6+) 51.Qf8+ Kh7 52.Kg2 Qg7 (Unfortunately for Black 52...e2 loses to 53.Qe7+ Kh6 54.Qxe2 Qxb4 55.Rxd2) 53.Qxg7+ Kxg7 54.Kf3 Bd4 55.Rf1 Kf6 56.Ke2+ Ke5 57.Rf8 Kd5 58.Rc8 Ke4 59.Rb8 Bc3 60.Rb7 (Unfortunately for White the obvious 60.Rxb5 loses to 60...d1Q+ 61.Kxd1 Kf3 and Black's e-pawn cannot be prevented from queening) 60...Bxb4 61.Rxb5 Bc3 (Now 61...d1Q+ fails to 62.Kxd1 Kf3 63.Rxb4 and the queening of Black's e-pawn can be prevented) 62.Rb8 Bd4 63.Rb7 Kd5 64.Rb5+ Kc4 65.Rg5 Kb3 66.Rd5 Kc3 67.Rd7 Bc5 68.Rc7 Kb4 69.Rb7+ Ka3 70.Rb8 Bd4

click for larger view

<mel gibson> Here is Stockfish's evaluation of the game continuation, 40.exd5:

2. [-4.05]: 40.exd5 e4 (Interesting, Stockfish evaluates the advance of the passed pawns as being better than the immediate recapture – with check – 40...Qxd5+. Perhaps a lesson for at least some of us) 41.Qa3 e3 42.Qa2 Qf5 43.Qe2 Qxd5+ (Only now, when White is able to blockade the advanced center pawns but with the pawns further advanced, does Black capture the Pd5) 44.Kg1 Qc4 45.Kf1 (Per Stockfish Black's 3 connected passed pawns after 45.Qxc4 bxc4 are more powerful than White's 2 connected passed pawns, evaluating the resulting positions at best (for White) at [-60.66] following 46.Kg2. And 46.a6 does not help, at d=49 Stockfish indicating mate in 15 or 23 after 46...e2 and 46...Bd4 respectively) 44...Qxb4 46.a6 (This is still not good enough) 46...Qa4 47.g4 (This smacks of the horizon effect at work, White realizes it is lost and tries gain the initiative at all costs) 47...hxg4 48.Qxg4 Qc4+ 49.Qe2 Qc6 50.Qg4 Qf6+ 51.Ke2 Qxa6 (And only now, with Black's pieces in their best positions and White's king more exposed, does Black capture the Pa6. A good chess player this Stockfish) 52.Rg1 b4+ 53.Kf3 Qb7+ 54.Qe4 Qf7+ 55.Qf4 Qa2 (55...Qxf4+ 56.Kxf4 and at d=46 Stockfish evaluates the position as even, [0.00], after either 56...Bf6, 56...b3, or 56...Kh6) 56.Rg2 Qb1 57.h5 (Charge!) 57...Qf1+ 58.Kg3 Qxf4+ 59.Kxf4 gxh5 (Now, as a result of 57.h5, Black picks up an extra pawn so the queen exchange is winning, Stockfish at d=30 evaluates the resulting positions as no worse than [-23.30] after 60.Kf3, with the evals climbing) 60.Kf3 Kh6 61.Rb2 h4 (Too many pawns for White's rook to handle) 62.Rxb4 h3 63.Rb6+ Kh5 64.Rb5+ Kh4 65.Rb2 Be5 66.Rb6 (I'm not sure why not 66.Rb5 immediately) 66...Kg5 67.Rb5 Kf5 68.Rb6 Bf6 69.Rb2 Bg5 70.Rb5+ Kg6 71.Rb1 d3 (Game over, 2 connected passed pawns on the 2nd rank typically beat a rook, even without a supporting bishop) 72.Ke4 d2 73.Kf3 Kf5 74.Rg1 h2 (Like trying to stop a swarm of African army ants) 75.Rh1 Bf4 76.Rd1 Bh6 77.Rh1 Ke5 78.Rb1 Kd4

click for larger view

And restarting the analysis from this position Stockfish at d=35 indicates mate in 16, 11, and 10 after 79.Ke2, 79.Ra1, and 79.Rh1.

<mel gibson> But back to the issue of "your" Stockfish vs. "my" Stockfish. Because of multi-core engine non-determinism it's not unusual, indeed it's the rule, that two Stockfish analyses of the same position carried to the same depth will give different evaluations and even different move rankings (with, of course, different lines) even if they were done on the same computer and the same version of the engine. Sometimes the differences are small but sometimes they are significantly different.

A minor quibble. An evaluation in the range [-0.49, + 0.49] or even [0.00] does not mean that the position is a draw (although that's the most likely result) but that the position offers even chances for both sides. It may be that if the analysis is carried out to a deeper depth that the evaluation might change dramatically, and this doesn't mean that the earlier "equal" evaluation is necessarily wrong. Check my forum's header for my definition of AKC2ML (AylerKupp's Corollary to Murphy's Law) for my tongue-in-cheek description.

I'm definitely with you on this one. The extra superfluous stuff also makes the post larger.

< This is probably a setting in your GUI that you can change and it will probably benefit everyone. >

machine analysis is fine when machines are playing chess. but i can't help inferring a certain element of human intimidation in the follow up to Rxd5, in particular white's indecision on pushing the QR pawn and kneejerk K retreats.

<acg2002>'s mere tactical analysis and <FSR>'s aphoristic summary may be closer to the truth, both because they smile away arduous calculation and because they don't imply black had any plan greater than bluff -- a tactically decent bluff, as <acg2002> points out, but a bluff perhaps premised on the fact that white saw his nice pawn fence destroyed in 5 moves and two lethal pawns left marching his way, and fell into a flop sweat.

Spotting it faster, not because of. Doing a position evaluation at move 39 shows one striking difference: White has the much better minor piece.

Going indepth in the position evaluation, Black's major pieces cannot invade, the central pawns can't move and the bishop's aiming at his own pawns. Meanwhile, White is not facing immediate issues, can push the outward passer and the knight's on an amazing outpost that's immune to the bishop.

RxN comes naturally realizing it removes the minor piece for White vs a slicing bishop, strengthening the central pawns and gaining d5 to boot. At least Black exchanges 5 for 4 in material and a difficult assessable positional advantage. The fact that the resulting position is a drawish is because White himself can improve positionally before taking on d5, levelling strengths when the dust settles.

Another minor quibble. I don't know which GUI you are using but I personally find it hard to follow your posted analyses without move numbers and with the long notation; e.g .♕f7-e6 instead of 40.Qe6). I personally find the figurine notation and extra "from" square distracting, but others may prefer it. At any rate it's your analysis, so you can post it any way you want to. But please consider adding move numbers since it makes the analysis easier to follow. This is probably a setting in your GUI that you can change and it will probably benefit everyone.>

I use Stockfish 11.
The answer you get depends on how long you run the engine. I use Arena as my GUI.

I think that it's you that is missing the point. Analysis is analysis whether it's done by human or machine and, the more accurate it is, the better. But if you think that a superficial analysis made by humans is somehow closer to the "truth" because it was done by humans, OK.

<i also take the indecision and confusion in the machine analyses laid before us as another point of evidence.>

Evidence of what? How did the machines <in this case> show indecision and confusion? (I've seen many other cases in which they did, exhibiting a phenomenon that I call "machine dithering" where the engines move pieces back and forth seemingly aimlessly because they don't seem to know how to make progress. But this was not one of those cases. I thought that Black's play in particular was exemplary, delaying playing obvious moves in order to increase its advantage. And White put up a good defense, it just had a disadvantageous position to start with I don't know why you consider White's indecision on pushing the QR pawn to be indecision; the engine analysis shows that it doesn't change the game's likely outcome.

< after all, what do machines know of intimidation?>

The EKG was "read" by a computer and "it" assigned a certain interpretation to the results. This interpretation is initially called "unconfirmed" in medical terminology, until it is read and approved by a doctor, at which point it is "confirmed".

In very similar fashion, a computer produces "unconfirmed" results, which may be meaningless, they may be correct, but they should be taken "lightly" until reviewed by an expert, who can either "confirm" or further modify the results.

For example, I present the following "unconfirmed" results, after 39...Rxd5 40.Rf1:

click for larger view

Stockfish_20082118_x64_modern:
NNUE evaluation using 20200812-2257.bin enabled.

<56/32 14:38 0.00 40...Qe6 41.exd5 Qxd5+ 42.Kh2 d3> 43.a6 e4 44.Qe1 d2 45.Qe2 Bd4 46.Qg2 Qa2 47.a7 Bxa7 48.Rd1 e3 49.Qb7+ Kh8 50.Qe7 Qb2 51.Kg2 Bd4 52.Kf3 Qb3 53.Qf8+ Kh7 54.Qe7+

I think that it's you that is missing the point. Analysis is analysis whether it's done by human or machine and, the more accurate it is, the better. But if you think that a superficial analysis made by humans is somehow closer to the "truth" because it was done by humans, OK.

<i also take the indecision and confusion in the machine analyses laid before us as another point of evidence.>

Evidence of what? How did the machines <in this case> show indecision and confusion? (I've seen many other cases in which they did, exhibiting a phenomenon that I call "machine dithering" where the engines move pieces back and forth seemingly aimlessly because they don't seem to know how to make progress. But this was not one of those cases. I thought that Black's play in particular was exemplary, delaying playing obvious moves in order to increase its advantage. And White put up a good defense, it just had a disadvantageous position to start with I don't know why you consider White's indecision on pushing the QR pawn to be indecision; the engine analysis shows that it doesn't change the game's likely outcome.

< after all, what do machines know of intimidation?>

I use Arena as my GUI also, but I don't think that affects the results. The point I was trying to make is that even though you run an analysis of a position several times successively, on the same computer, with the same multi-core engine, and to the same depth, you will get different results, even if you clear the engine's hash table before to start each analysis. Not MAY, but WILL. Guaranteed. But don't take my word for it, check it out for yourself.

The only way to reduce (not entirely eliminate) chess engine non-determinism is to run the analysis using MPV = 1. The reasons I've read are somewhat complicated and I don't entirely believe them myself. And it doesn't entirely eliminate the non-determinism but it greatly reduces it. The extra time that the analysis takes doesn't bother me because for "serious" analyses I run multiple engines concurrently overnight and the next morning I average their evaluations to try to eliminate the engines' evaluation biases. Stockfish, for example, is well known to generate evaluations that are typically higher than other top engines like Houdini and Komodo produce. So averaging their evaluations reduces, although it does not eliminate, some of the engines' built-in evaluation biases and generates an evaluation that is closer to the "truth", whatever that might be.

And, like I said, it's your analysis and you can post it anyway you want to. We are all grateful that you take the time to do so. But if your intent is to post analyses so that others can see it, then I think that everyone, including yourself, should post their analyses in a way that it's of most use to those that are intended to benefit from it.>

A longer run provides more evidence that the position is drawn under best play.

click for larger view

Stockfish_20082118_x64_modern:
NNUE evaluation using 20200812-2257.bin enabled.

88/35 18:41:36 0.00 42...e4 43.a6 d3 44.Qe1 d2 45.Qe2 Bd4 46.Qg2 Qa2 47.a7 Bxa7 48.Rd1 e3 49.Qb7+ Kh8 50.Qd7 Qc4 51.Qxa7 Qe2+ 52.Kh3 Qxd1 53.Qb8+ Kg7 54.Qe5+ Kf7 55.Qd5+ Ke7 56.Qe4+ Kd8 57.Qd5+ Kc7 58.Qc5+ Kd7 59.Qd5+

Fair enough. We all visit this site, participate in the discussions, and post information because we like doing it. That's certainly the case with me. And at least in my case no amount of financial inducement would make me change from doing the things I like to do (and the way I like to do them) to doing things I don't like to do and/or doing them in a different way.

And I only mentioned that financial inducement because you brought it up tangentially. I didn't mean to insult you in any way. If I did, I apologize.>