lichess.org
Donate

Major Errors in Cloud Analysis

It's possible this error is on my end but as its about cloud analysis and I've replicated it on a second computer it seems unlikely.

I've been noticing (only in the past week or so) that cloud analysis has been tremendously inaccurate. For instance in this screenshot imgur.com/a/zwr3KW1 after 1.e4 e5 2.Nf3 Nc6 3.Nc3 Nf6 4.Bb5 Bc5 5.0-0 0-0 6.Nxe5 Re8 7.Nxc6 following Paulsen-Morphy the ridiculous 7...Bxf2+?? is evaluated as only slightly better for white (+0.7). Make the move and the actual engine turns on immediately evaluating it as around +8 and realizing that 8.Rxf2 is better than the listed 8.Kxf2. I'd never seen the cloud have extreme errors on simple moves like this until recently so my guess is something has changed behind the scenes with weird ramifications?

Lichess is wonderful, thank you.
I've been noticing that too. The cloud analysis might show a higher depth, but is sometimes less accurate than my local analysis. My guess is that the data was from older stockfish, although the engine was updated they kept all the data evaluation data from old stockfish, thus the discrepancy. Could be a potential reason why we have yet to get stockfish 15.1.

I've input the moves from your post and seen a correct evaluation for your 7...Bxf2+?? move now at depth 36, maybe your local analysis overrided it?

imgur.com/a/sXAbj0N

I love lichess too but i really hope we could see stockfish updated soon
For me, the evaluation of 7...Bxf2+ is very different depending on how many lines I set. Actually with three lines I get shown a depth of 36, while the op got shown a depth of 29 there.
______________________________________________________________________
With three lines I get
-0.2
Stockfish 14+NNUEDepth 36Cloud
-0.2
7...dxc68.Be2Nxe49.Nxe4Rxe410.d3Re811.Bd2Qd612.Bc3
+0.9
7...bxc68.Bc4d59.d4Bb610.exd5cxd511.Bb5Bg412.f3
+7.9
7...Bxf2+8.Rxf2bxc69.Be2d510.d4dxe411.Bg5Rb812.b3
______________________________________________________________________
With four lines I get
-0.2
Stockfish 14+NNUEDepth 36Cloud
-0.2
7...dxc68.Be2Nxe49.Nxe4Rxe410.d3Re811.Bd2Qd612.Bc3
+0.9
7...bxc68.Bc4d59.d4Bb610.exd5cxd511.Bb5Bg412.f3
+7.9
7...Bxf2+8.Rxf2bxc69.Be2d510.d4dxe411.Bg5Rb812.b3
+12.5
7...Nxe48.Nxd8Nxc39.Qh5Rxd810.Qxc5Nxb511.Qg5Re812.Qxb5
______________________________________________________________________
And with five lines I get
-0.3
Stockfish 14+NNUEDepth 29Cloud
-0.3
7...dxc68.Be2Nxe49.Nxe4Rxe410.c3Bd611.d3Re812.Be3
+0.7
7...Bxf2+8.Kxf2dxc69.Bd3Qd4+10.Ke1Bg411.Ne2Qd612.h3
+0.7
7...bxc68.d4Bb49.Bd3Bxc310.bxc3Nxe411.Qf3d512.Re1
+12.5
7...Nxe48.Nxd8Nxc39.Qh5Rxd810.Qxc5Nxb511.Qxb5d512.Qc5
+12.8
7...d58.Nxd8Rxd89.Nxd5Nxd510.exd5Rxd511.Bc4Rf512.Bd3
______________________________________________________________________
So it seems to make a big difference whether the depth is 36 or 29. But I have no idea why the number of shown lines leads to diffent evaluations at all.

This topic has been archived and can no longer be replied to.