Chess evaluation drops suddenly, while following the best sequence !

Title. 

Let's take my last game as example : https://www.chess.com/analysis/game/live/12679093919 

On move 51, I should go b5, and then it gives me an edge of +5 (or so it says) 

But then, following this sequence, it drops to +2 to end up to a draw, while this should be clearly winning (only talking about the evaluation number)

I don't understand why evaluation drops, it should stay more or less the same when both players play best moves, on main sequence. 

well at the second moment it saw something it didnt see at the first and that made it a draw. why? because the first moment was calculated 1 move before so there were much more possible move sequences

I see, it's an engine depth problem then 

I didn't see anything wrong. after b5 you have a slight advantage, around +2, and if both players play logically your advantage grows.

I fixed it my using personal analysis (my cpu) 

It gives more depth options so I set it to 22 and now it sees the draw from move 51. It was indeed about engine depth. 

I don't think that's correct but gz

They call this the "horizon effect". For all chess players, their understanding of a position is limited by how far ahead they see. Usually positional factors and short range tactics are a good guide, but occasionally, just beyond the horizon something happens that affects the evaluation in a big way. This can become visible as the moves go by (because the horizon is moving forwards in the future of the potential game lines).

In the beginning of computer chess, CPUs were still so slow, that computer programs could look only a few moves ahead, and the horizon effect was much more pronounced then. It was possible for an ordinary human player to outwit the computer by analyzing a combination to a deeper level than the program could and see the refutation that the computer did not yet see. The computer's evaluation for the game often changed from "won" to "lost" between a move and the next.

Analysis run by latest Stockfish 14 devlopemental vesion engines, and latest Leela on T60 network 

Both engines say, 51. d5 was best move, and eval around +4

After d5, evaluation are more or less the same.

So, current best top two engines disagreed your claim.