weirdnesses in evaluation reports

When going through the key moments of a game, a "friendly" bot-game https://www.chess.com/a/2K7SNRr4N2QyVx, the following weird thing happened. I looked at my "blunder", 52..Rf7 which was evaluated as -4.75, and tried to find a better move. I tried 52...Rc4, which was also viewed as a blunder but evaluated well below -10 [trying to reproduce this did not work, then the eval came back as -5.80]. The actual "best move" (Rh4) was evaluated at -9.12, so the best move was worse than a blunder?? Afterwards, in the analysis, the eval of 52..Rc4 had changed to -3.98.

My wild-guessing explanation for that situation is that the responses to key moment mistakes are evaluatated by the engine at a different depth than the actual report, most likely a higher depth, but what counts as a blunder or not isn't changed, so one can get a rather inconsistent overall picture. (Why any of those moves is supposed to be a blunder is a mystery in its own right, but that's a different topic.)

Another weirdness I have seen numerous times is that when you are close to mating, found (for example) an M5 move (when there is an M4 move) and that move is listed in your key moments, upon reveal both your move and the better move are listed with the same mate counter (although the better move is 1 shorter). This does not happen in the analysis, just under key moments.

The eval tool recently underwent analysis- 

a lengthy couch session with a marked diagnosis-

Bonkers !