Setting up inaccuracies and errors in the analysis module. • page 1/1 • Lichess Feedback • lichess.org

Hello,

I really appreciate Lichess's analysis system.

You play, you analyze, and you can replay the missed moves based on the AI's analysis. Everyone knows that.

However, I'm sometimes frustrated that some moves aren't included among the inaccuracies, and some inaccuracies aren't included among the errors.
The result is that I have to redo the entire analysis myself. This isn't a problem in itself, but it makes the automatic analysis useless.

Yet I find it very useful and would like to be able to fully utilize it.

For example:

e4 Nf6 2. e5 Nd5 3. c4 Nb6 4. d4 d6 5. f4 g6 6. Nf3 Bg7 7. a4 a5 8. Be2 O-O 9. O-O Be6 10. Qc2 c5 11. d5 Bf5 12. Qc3 N8d7 13. Nh4 Bxb1 14. Rxb1 dxe5 15. fxe5 Nxe5 16. g3 Nexc4 17. Qf3 Ne5 18. Qa3 Qxd5 19. Rd1 Qc6 20. Bf4 Nec4
a4 and a5 should at least be considered inaccuracies, if not errors (delta 0.7).

10... c5, 12. Qc3 and 13... Bxb1, should considered as errors (delta 0.9).

THE QUESTION IS:
Is there a way to parameterize the analysis to make it more refined and stricter in order to rework a larger number of errors?

Or add the inaccuracies to the moves to be replayed?

Thank you and have a good day.

Hello, I really appreciate Lichess's analysis system. You play, you analyze, and you can replay the missed moves based on the AI's analysis. Everyone knows that. However, I'm sometimes frustrated that some moves aren't included among the inaccuracies, and some inaccuracies aren't included among the errors. The result is that I have to redo the entire analysis myself. This isn't a problem in itself, but it makes the automatic analysis useless. Yet I find it very useful and would like to be able to fully utilize it. For example: 1. e4 Nf6 2. e5 Nd5 3. c4 Nb6 4. d4 d6 5. f4 g6 6. Nf3 Bg7 7. a4 a5 8. Be2 O-O 9. O-O Be6 10. Qc2 c5 11. d5 Bf5 12. Qc3 N8d7 13. Nh4 Bxb1 14. Rxb1 dxe5 15. fxe5 Nxe5 16. g3 Nexc4 17. Qf3 Ne5 18. Qa3 Qxd5 19. Rd1 Qc6 20. Bf4 Nec4 7. a4 and a5 should at least be considered inaccuracies, if not errors (delta 0.7). 10... c5, 12. Qc3 and 13... Bxb1, should considered as errors (delta 0.9). THE QUESTION IS: Is there a way to parameterize the analysis to make it more refined and stricter in order to rework a larger number of errors? Or add the inaccuracies to the moves to be replayed? Thank you and have a good day.

Milou_le_Viling

Just now, I had another example.

The only inaccuracy of my game was a move rated at 1,7 when the best move is 1,8. It's not an inaccuracy.
On the other hand, a move rated 1,3 when the best move was rated 1.9 was not recorded.

I think there is something to improve here...

Just now, I had another example. The only inaccuracy of my game was a move rated at 1,7 when the best move is 1,8. It's not an inaccuracy. On the other hand, a move rated 1,3 when the best move was rated 1.9 was not recorded. I think there is something to improve here...

ExertC

You can use the Live Engine Analysis setting, which decides if the move is an inaccuracy, mistake, blunder when you're on the move. It seems to be a lot more accurate than requesting computer analysis. However, it seems to be broken for me right now.

gotwic edited

I concur, in some games the eval goes up and up without a single critical moment or deflection point according to analysis.

And thus, with no mistakes the accuracy metric gets inflated as well.

I concur, in some games the eval goes up and up without a single critical moment or deflection point according to analysis. And thus, with no mistakes the accuracy metric gets inflated as well.

gamblej

This is valid criticism. There's often no good way to programmatically identify mistakes at discrete moves when the eval changes are gradual. It's why computer analysis cannot replace your own.