Comparing: accuracy by game phase

There is always someone telling someone to play better. Well for the fun of it, I'm going to compare how much better I need to play.

My accuracy by game phase:
Opening: 73.8%
Middlegame: 56.5%
Endgame 66.9%
https://lichess.org/insights/Toscani/accuracy/phase

Well for starters, it sure looks like I need to improve my middle games ...
Opening: 22.7%
Middlegame: 30.1%
Endgame 21.6%

How did I discover by how much I needed to improve? I compared my "Accuracy by game phase" with the info found about the World Grand Master, that everyone should know by now.

Ding Liren accuracy by game phase:
Opening: 96.5%
Middlegame: 86.6%
Endgame 88.5%
https://www.chess.com/insights/chefshouse#phases

Besides the bots and assistant users ... is there a player on lichess that gets even close to the "accuracy by game phase" percentages of the new world grandmaster?

engine point of view of accuracy.. should i remind, that chess is still not known. We have no clue if the engine are actually accurate themselves. but hey they beat us. and beat each other.. so... I just want to remind that these accuracies are per certain engine (and maybe the engines used and their versions should be mentioned, if one day, some metric were to appear about all engine accuracies relation which each other across versions).

I am not saying it might not be close to complete chess accuracy, just that we don't have any way to prove it or measure it.

So the accuracy of what one site uses as an engine does not mean the other site uses the same engine or it's algorithms.
So I guess the accuracy by phase cannot be compared accurately between sites.

Still, for the fun, please continue comparing.

I have (for rapid only):
Opening: 88.6%
Middlegame: 72.4%
Endgame: 80.2%

Interesting, perhaps as long as we respect the principles and have a general idea of what an okay position is, the opening will be higher because there is less calculation in it.

probably depends on how far from book players have been in game.. it might be about being strong of repertoire, and same repertoire..

perhaps filtering by rare popularity and then looking at accuracy..

i think the engine does not have book when computing scores from which error in cp is computed. so it may also be scoring a very accurately followed opening as your errors..

might want to factor the engine opinions versus opening theory conventions or knowledge out..

I would think middle-game to be where the engine might be bring something to the table in term of per move measure.

but some believe in the engine preferences in openings as possible discoveries.. i think best is testing by yourself and considering both possibilities.. having mutliple answers in parallel is sustainable.. it they might even be more informative together.

I have for rapid here:
Opening 94,4%
Middlegame: 84,4%
Endgame: 87,2%

That is pretty close to the figures of the world champion but with the huge difference that it is comparing blitz with rapid.
For people wondering if this is possible without cheating then I recommend reading my blogarticle http://schaken-brabo.blogspot.com/2023/03/extra-tijd-deel-2.html