Best move according to engine is classified as inaccurate

I was going through some openings study and found something that seemed rather odd to me:

The engine proposes a move to be the best available (I let it run for a few seconds) and then when played, it classifies the move as inaccurate (see attached example; left panel is before move, right panel is after). 

Reading some other posts I thought maybe it was due to the fact that, despite being the best move, the resulting evaluation after the move was worse. But reading the evaluation bar, I see that the evaluation doesn't seem to change after the move: +0.06 vs +0.06 (I'm interpreting the number on the bar as the current evaluation and the number on the engine line as the evaluation after the suggested move/line).

What am I getting wrong? Any thoughts on this apparent inconsistency are appreciated.

A few seconds really isn't that long and you're likely going to get differences. I like to go to at least depth 20 in any casual look. For more serious positions even deeper. 

In really complex positions, it might completely change a ply later, even if you let it go deep.

Thank you Martin.

I ran the engine for quite some time and Bxf7 is still the best move for white (image attached). Anyway, I think the depth setup/time is not the problem here. I presume that the engine result being used to classify the move is the same result that appears in the engine line. 

Something new I discovered was that according to this article, the move classification is based not only on the chess engine but also on the player's rating. I'm not yet convinced this is what I'd like to see since I was counting more on an absolute measure of move quality (at least for "Best move") and not one that varies with player rating. Back to my question and to the point, why would the best move produced by the engine end up tagged as "inaccurate"?

So, the problem is like that at different depths, the evaluation flops between basically even, but over the line from very, very slight advantage for white to a very slight advantage for black.

I'm not certain on the exact definition of inaccuracy being used, but flipping evals like that may be enough to quantify as inaccuracy. Before it's played, it's the top engine choice so it's best, at that point.

However, one additional ply is sometimes enough for the engine to not like a move as much. Part of that is due to the horizon effect, some due to move pruning. So the newer information is being used after the move is on the board. A few ply later it goes back to even.

Huh. I was also wondering about something similar to this that happened a while back.

Also, bump 😁

I guess not