Strange accuracy score

Sometimes when I analyze games here on chess.com, the accuracy score seems incorrect. For instance, I recently played a blitz game with this result (edit: I won on time by the way):

My opponent had 9 bad moves whereas I had only 4. I have no blunders, my opponent did have one. I do have 1 extra inaccuracy, and 1 fewer "best move". But overall, it seems I have significantly more green compared to red/yellow.

So, I decided to get a sort of second opinion. When I upload the PGN into Lichess, it gives

Opponent
3 inaccuracies
2 mistakes
4 blunders
72 Average centipawn loss
Me
6 inaccuracies
2 mistakes
1 blunder
47 Average centipawn loss

A lower centipawn loss is better, so Lichess's stockfish seems to agree that I played more accurately.

How can these 2 platforms, both using stockfish, come to such different results?

Source: game 28657261317

Stockfish doesn't do anything other than provide the evaluation. The sites decide how to categorize those evaluations and that is a subjective decision and different sites can quantify how that works.

Makes sense, but "average centipawn loss" and "accuracy" should be a very similar metric right? They both express how much worse your moves were than the engine's recommended ones. 100% accuracy = you played every top engine move, which is the same as 0 centipawn loss.

And since it's the same underlying engine as well, I'd assume there wouldn't be such big gaps.

64% vs 58.5% means: slightly better by white

72 vs 47 means: significantly better by black

I just can't see how the result is so far apart without one of the scoring mechanisms being poorly implemented.