I don't understand the computer eval. in this position

So it's black to move. I'd think Bxe5 would be the best move but the computer lists

a5

Rae8

Rf7

as the top moves. I don't understand why you wouldn't just exchange the bishop for the rook. Appreciate any help understanding.

A complex position. I am not sure I understand why you shouldn't just take the exchange either.

I think it has to do with the e6 pawn being a weakness and black's light square bishop being bad. After winning the exchange, the white queen controls e5.

The pin of the white knight by the queen seems to be restraining white's progress a bit. Also white still has pieces awaiting development.

Thanks. Yeah, I thought it had to do with the e6 and d5 pawns. But the rook is already blocking the e6 pawn from going forward. So I don't see why it's worse for black with the queen blocking it instead. 

Maybe it's because of the danger of the White Queen eventually reaching d6, c7 or b8? If the black queen moves away... and the white queen gets to d6... that could be a big issue for black.

You mentioned the pin... the black bishop could exploit the pin coming to c5. Black would lose that possibility with the exchange.

It does not really matter. Black is completely winning, also after ...Bxe5. It is possible to postpone taking on e5 and playing ...a5 first, as white cannot save his rook: if he moves it, then ...e5 wins a whole piece Nd4 instead of an exchange.

Yeah what tygxc said. Black can delay the capture if he wants as white can’t really save the rook.

More generally, taking the rook isn’t an automatic move in such a situation just because the rook is ‘worth more’ as it does give up the better of your two bishops, and potentially leave you weak on the dark squares (whites Bc1 is now unopposed). 

In this exact position though, taking the rook right now is absolutely fine (even if not the absolute most accurate) as after Qxe5 you can always follow up with something like …Qb8 forcing more trades or evicting his queen allowing you to get …e5 in and freeing your renaming bishop a bit. 

A rationalisation of the failure to capture is that the only reason one would think capturing immediately is necessary is that the rook can retreat. But if the rook retreats, black can push the e-pawn winning an exchange any way (since white has to capture on e5 to save the pinned knight).  This is not quite the right justification, but interposing the move Rf5 (taking advantage of the pinned knight to bolster e5 before pushing) is superior.

Even post-justification of superhuman chess can be really hard, but there is something to be learnt by trying.

I was curious, so I fired up Stockfish locally to see what it was saying.

The difference between a5 and Bxe5 is 0.33 points  White is in a completely won position with almost any move.  My GUI (Nibbler) says that the worst of the top 5 moves, white will win 999/1000.

You'll drive yourself bonkers trying to understand the difference between being up 5.25 vs *only* being up 4.92.

In the top two lines, Stockfish merely postpones Bxe5 by a play.

If you want to give yourself an aneurysm, try playing through the third line that starts Rf6 and convince yourself after a few moves that it's really better than Bxe5.   

Absolutely. I managed to rationalise the first few moves in the sort of lines that a human would be looking at, but in the engine lines you have a 3xxx player playing another 3xxx player in every analysis line, which can be opaque!

No way can you be sure that a 4.95 move is worse than a 5.25 move. Both evaluations indicate a clear win, which will eventually evaluate to +infinity to a chess oracle, so is guaranteed "wrong". All computer assessments except extremely large ones indicate uncertainty. The three results of chess mean that a zero rating can be a certain draw or an uncertain result with balanced probabilities.  A more informative evaluation is the probability of each of the three results with perceived best play. @WBillH's GUI gives the next best thing - the expected score.

Thanks. I seem to get a wider margin using chess.com engine (depth=25). After a5 I get an eval of -4.3. After Bxe5 I get -2.66

The chess.com engine is great for quick looks and blunder checks.

If you're wanting to compare different evaluations from the engine, you need to understand what it's telling you.  What's the difference between depth 24 vs 25?  25 vs 26?  Are the evaluations pretty much staying the same, or are they changing as it continues?  If they're changing, are they converging or diverging?  What other lines are shuffling around in the rankings?

You really can't just look at one number and make any meaningful statements.  And you really need to play through the line, too.  Like Elroch said, he looked at the lines that appeared like a human would consider them.  That will tilt the analysis, too.

LOL!  Correct.  Uhm...black is the new white??

You can play a fantastic game and the computer analysis will tell you that you are making mistakes the whole way through.