Those are two completely different lines. It isn't saying that 3.c4 is worse than 3.a3 e6 4.c4, it's saying that 3.c4 e6 4.g3 .... is worse than 3.a3 e6 4.c4 ..... Given more time, I am pretty sure it'd alter that conclusion, as well as determine either that a3 or g3 can be played with c4 in any order to reach the strongest position.
I don't know whether you know how engines actually calculate variations, but what you're seeing is neither surprising nor unique to Stockfish.
What does the second number mean? For example, 36/49, what does the 49 mean?
What does the second number mean? For example, 36/49, what does the 49 mean?
The second number is the selective search depth. Here's an article by Steve Lopez that explains it better. (He's talking about Fritz, but the same idea also applies in a generic manner.)
https://uscfsales.wordpress.com/2011/07/08/understanding-chess-engine-evaluations/
"Notice, though, that the value after “Depth=” in the illustration above consists of two numbers separated by a slash. The number to the left of the slash is the regular depth value, in which Fritz is looking 19 plies ahead. The number to the right of the slash is called a “selective search” value, in which a chess engine is able to look far more deeply in certain forcing variations which contain tactics, exchanges, checks, etc. In the present position, Fritz can look a whopping 41 plies ahead in these forcing or tactical lines."
So what's going on here? This isn't the first time that I have noticed something like this. After 1. d4 Nf6 2. Nf3 b6 Stockfish thinks 3. a3 is white's best move and after 3. ... e6, 4. c4. However, it evaluates 3. c4 e6 4. g3 as being just slightly worse; although surely white can just play 4. a3 and have the exact same position? What is going on? Is there a flaw in the program (do other engines behave like this?)? Or is this just a facet that humans grasp instantly while machines don't account for? Thoughts?