a3 is nowadays the main line at high level and almost refutes 3...c5 in my opinion...
For example :
White has won quite a few games in the 5...a3 lines because the ideas for black are still being developed and it's easy for black to go slightly wrong if you aren't up to date on the latest theory. For example, 5...Nc6 is probably a slight improvement over Grishchenko's 5....Bf5. The opening is far from being close to refuted.
a3 is nowadays the main line at high level and almost refutes 3...c5 in my opinion...
For example :
According to stockfish 9, Bf8. is his choice at 400M nodes. I will check later what is Leela's choice about this position).
If Bf8 is Black's choice in that position, I don't much care for Black. And no, I don't trust the comp's eval of the opening much either.
Btw, let's make sure we're all talking about the same variation. Poucin and I were discussing 5.a3 in response to 1.e4 c6 2.d4 d5 3.e5 c5!?/?! 4.dxc5 e6 NOT 4...Nc6.
Btw, White doesn't have to play 6.b4. He can wait to play it and instead play a normal developing move and chase away the B later. Most top players (Karjakin, Jakovenko, Shankland among others) play 6.Nf3 and hold the threat of b4 for later.
I'm sure an engine will be able to draw against 3...dxc
But from the game Poucin posted it looks like such an effort would be hugely impractical for a human. White gets easy moves while black's position looks endlessly unpleasant. If you have to work 10x harder on every move of the game than your opponent just to draw it's time to change openings!
If Bf8 is Black's choice in that position, I don't much care for Black. And no, I don't trust the comp's eval of the opening much either.
Btw, let's make sure we're all talking about the same variation. Poucin and I were discussing 5.a3 in response to 1.e4 c6 2.d4 d5 3.e5 c5!?/?! 4.dxc5 e6 NOT 4...Nc6.
Btw, White doesn't have to play 6.b4. He can wait to play it and instead play a normal developing move and chase away the B later. Most top players (Karjakin, Jakovenko, Shankland among others) play 6.Nf3 and hold the threat of b4 for later.
Yep, 6. Nf3 is Stockfish choice as well. SF will reply with 6....Ne7 at 200 million nodes.
0.00 evaluation.
Please note Stockfish eval are generally reliable once SF hit 1 billion nodes( that usually take 3 mins of analysis on 4 cores desktops).I will have a look at brainfishdatabase whereas Thomas analysed every position for ??? 10-20 billion nodes( not sure how much he analysed) .
I hate caro kan
I know
a3 is nowadays the main line at high level and almost refutes 3...c5 in my opinion...
For example :
White seems to get a long lasting pressure but I don't think Black's resources have been exhausted. Many moves are still untested in this line.
Isn't it a bit premature to talk for refutation?
of course it is premature, it was a bit provocative! 
But it is my opinion : this way to play as white seems to set troubles for black and it is not surprising we don't see 3...c5 often at high level.
I discovered this line when i saw this recent top level game live, Topalov being crushed by Karjakin :
I don't remember seing 3...c5 at top level after this game.
7. ..........Ng6 both Stockfish and Leela agreed!
8.--- Nc6 ? both disagreed! It favour white to get b4 and b5 freely.
Instead both engines agreed
8. Qc7 and a6 in either order
9. .... Bb6 is disagreed by both engines. 9.----Be7 was suggested by both engines.
12. ----Nxf3 was disagreed by both engines. Instead play 12..... Bc7. After 12.... Nxd3 that, the game is unfavourable for black! White is much better,
Conclusion, At move 12, White is much better , black is unfavourable to play anymore!! Both engines agreed three inaccurancies!
I hate caro kan
I know
Stop
I know
Engines can whine about moves, but sometimes they can't find anything better.
Put 10 or 20 engine moves on the board and see if the eval is close to zero. I wouldn't be surprised if the engine slowly realizes white's better and better.
But even if it's equal, it's not practical to put yourself in a situation where you're required to find only moves just to survive.
Engines can whine about moves, but sometimes they can't find anything better.
Put 10 or 20 engine moves on the board and see if the eval is close to zero. I wouldn't be surprised if the engine slowly realizes white's better and better.
But even if it's equal, it's not practical to put yourself in a situation where you're required to find only moves just to survive.
Engine lines are hard to follow. However if you analyse a certain position back and forth several times, you will finally see a reason why engines say their evaluation, and why engines refute the other lines. e.g SF evaluation 0.00 means he is not evaluating in this current positon but at the end of the position of his principal variation (a position at depth 30-40 that your normally dont know where ). To see why SF evaluate like that, i have to stop the analysis, copy his PV and follow through the first prinical variaiton for 5-10 moves, analyse again and stop again to reach to the end of his principal variation. Unless the first move (the root ) is not changing, then 0.00 or 0.20, 0.30 etc doesnt matter. However, if there is change in the root, then there is FAIL high or FAIL low in Stockfish search,. Only then , I need to give SF 2-3 mins to let him search for his new variation again. In general, if you let SF search for 3 mins in 4 cores desktop, SF searches about 1 billion positions, and it is quite uncommon to see SF , (Fail high or Fail low ) or change in PV, then i accept SF choice as a good reliable one.
I play Caro Kan in this varation and "a3" move is unfamiliar to me. In other common varaitons , I will play exactly identical SF moves in my bullet/blitz games (sometime up to 12, sometimes more than 15) cos I played thousands of games with this same Caro Kan variation. I analyse, follow, analyse again in new variation again etc.
As a side note: If you intend on playing the ...c5 line, I'd encourage you to look through the Tal-Botvinnik match of 1960. Botvinnik played that line the remainder of the tournament after losing his first game in a French Winawer.
3...Bf5 is superior to 3...c5. Why don`t these guys just simply buy a book or two on the C-K?
Ok, I looked some with stockfish.
Seems one point of Be7 is a5 and black is getting some play right away. Also like you said Qc7 instead of Nc6. This way of playing doesn't seem so uncomfortable for black. I can see things like the e5 pawn, the c file, the a5 break, and the position seems to make sense.
Engines evaluation make sense when you can see all major lines and , the refuted lines. In worst scenario, SF may say 0.00 in which only 1 out of 100 move is playable to reach 0.00. But the other 99% are losing move and you may fail. 
SF dont care if there is 1 in one million line that can lead to draw , but we care! 
Engines evaluation make sense when you can see all major lines and , the refuted lines. In worst scenario, SF may say 0.00 in which only 1 out of 100 move is playable to reach 0.00. But the other 99% are losing move and you may fail.
SF dont care if there is 1 one million line that can lead to draw , but we care!
There really are players out there, who after they lose complain, "but the position was a draw!"
As a side note: If you intend on playing the ...c5 line, I'd encourage you to look through the Tal-Botvinnik match of 1960. Botvinnik played that line the remainder of the tournament after losing his first game in a French Winawer.