Glitch in the chess.com analysis tool?

I've found that when I'm analyzing games using the built in engine on chess.com, that sometimes when a better move is recommended, I then make that move on the analysis board to see where it leads, and sometimes it then says that actually another move was better, e.g.:

so here, if I make the move d4 on the analysis board, it says O-O is better, and if I castle, it says d4 is better.

Sometimes the difference in the rating of the move is more significant than in this picture, but hopefully you get the point. the depth of the engine is the same and everything.

Does anyone know what the issue is here, and if it can be solved?

But the depth is not the same.

The chess.com computers are running tens of thousands of processes simultaneously.

The results can come out somewhat differently.

Analyzing opening variations is particularly difficult.

Thanks for your reply. As far as I can see the Depth is the same, it is 20 in both cases though.

Edit: also, i'ts not just in the opening, for example, the same move, with the same depth can be considered both a blunder or the best move:

The same is happening to me. It's very hard to use the tool with this glitch. And I have to say that this just started to happen recently, after the latest updates.

the same happens to me, regularly - I just consider it a function of the statistical process being used to analyze each move - the understanding I've come to is that when you replay a move over and over and it gives conflicting suggestions, it generally means you're in such a bad position that there really is no solid "best move", just moves that are better than others. also, taking the statistical "best moves" as suggestions, since you're playing against a human with emotions, blind spots, and who may have obvious weaknesses in their game, so a move you make may have been a statistical miscalculation, but actually the best move against that particular opponent.

It depends on the depth. On a lower depth especially, this will happen a lot. You will make a move, engine will show some best move, but when it is on the board, it will see that some other move is better. Free depth 10 analysis will do this a lot.

Even depth 18 will do this sometimes. You have a few options:

1) Use some free offline tool for serious analysis

2) Upgrade to diamond membership so you can analyze at a greater depth
3) Try to adapt and use it as it is

Engines have a hard time with positions that are more complex.  In that position, black has a passed pawn and white has the doubled h-pawns, so the engine thinks that is really good for black.

I'm not the greatest player but to my eyes, it looks 100% drawn unless white just gets careless. Engines are usually pretty bad at things like that. I believe there was a fortress position with Anand vs Carlsen when the engines had one side massively up and as the position progressed, it slowly realized it was draw.

There's nothing particularly wrong with the analysis here, just how they work and their horizons/how deep they can actually go.

Also, sometimes one extra ply is enough to change the evaluation enough that what was best the ply before isn't a ply later. Again, it's a horizon issue and in the opening, such evaluations, that aren't drastically different, can be taken with a grain of salt.