I don't know if this will help, but I analyze most of my games with both chess.com analysis and the freeware Lucas Chess (with Stockfish 9). Lucas gives you an analysis like this one I finished yesterday (https://www.chess.com/daily/game/192545898) where I played White and the King's Gambit for the first time and won:
This is more accurate than the chess.com analysis for the same game:
where a CAPS of 95.40 indicates about a 2350 rating equivalent and 47 must be around 500.
Considering I missed a tremendous move three moves in a row while concentrating on going up by one pawn, I think the 1828 rating Lucas gave me is closer to reality than the 2350 chess.com says I played at. Note that my chess.com rating at the end of this game is 1797.
Lucas Chess also gives you a list of variations for each move: you just click on the move and it shows the best moves in order. The moves listed in blue are considered best. The blunder (??) indications are mostly because of the difference in pawn equivalents compared to the best move and don't necessarily mean a true blunder:
Every so often in my many bullet games I find that I manage to pull off a surprising mate. I may see it 3 or 4 moves in advance, but the position leading up to it is always intuition. It would be nice to see the computer assign an ELO rating, based on the level of play in that particular game. I'm sure this has been thought of before and maybe it would cause more trouble than it's worth (people crying cheater and such), but I think it would be fun.
As it is the analysis is terrific. I still don't know the difference between a mistake and an inaccurate move, but I still love it.