result-prediction based mistake system

I think the current system that marks moves as inaccuracies, mistakes and blunders is flawed.

In relatively even position it is too easy on bad moves and in one-sided position it marks moves that just take longer as mistakes or even blunders.

For example, a move that drops the evaluation from +9 to +5 is marked as blunder when it is more of an inaccuracy, not really harmful on its own but making the job usually slightly more difficult.
To me, a move that drops the evaluation from 0.4 to -0.4 (marked as inaccuracy by the system) is worse.

I suggest calculating recalculating the pawn unit advantage into a number that estimates how many points one player would usually get out of that pawn unit advantage.
Then mark mistakes based on result-prediction loss and not pawn unit loss.

Hello, I agree with that, the system could use some changes. Recently I played an almost perfect blitz game, but I had a lot of inaccuracies in it, because I couldn't find the fastest checkmate, yet after each of the inaccuracies I still had forced checkmate. So those inaccuracies meant nothing.

However, +9 to +5 should be marked as blunder, because it makes it more likely that you will lose. (if you make another blunder that brings the evaluation to +1 then the game is almost equal).

I see your point, but I disagree.

Say you are losing a piece at +9. At that time, losing the piece doesn't risk your win. At +5, just playing decent enough moves will eventually snowball your advantage into a win.

To me, if you're still clearly winning, it wasn't really a blunder.

Let's look at a move in a drawn endgame. It keeps the position equal but now you will have to find a short series of only moves, after which the draw is secure, or lose, which wasn't necessary previously. By the same logic, the move should be marked as a blunder for risking the result if you make more mistakes.

But engines have difficulties assessing which positions are likely to cause human players to blunder, which is why I think a result-prediction based system taking in the pawn unit advantage would work best.

I see... Well, you need to convince @thibault to change the system.

#1 I agree and made this suggestion years ago. The problem is: what to do with already-analyzed/labeled games?

A blunder doesnt mean you will lose the game.
it just mean you proposed an inferior move.
which can be the case even if you are still winning.
if I miss a mate in 2 and play a mat in 4, I wan't the system to point it to help me improve.
I don't wan't the system to say "yes you're good at chess. don't worry this mate in 2 was not better than your move"