strange analysis by reckless in a winning position

[Uri Blass]

At keast the evaluation increase when you search deeper but the moves of the pv are strange.
I guess reckless 0.10 is also going to lose against Wasp with big handicap in the competition who mate faster based on this analysis

[d]1nb1kbn1/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQ - 0 1

Reckless 0.10.0_generic:
set Hash to 4096 MB
set Threads to 6
1/1 00:00 318 240k +6.11 e2-e3
2/2 00:00 447 307k +6.27 e2-e3
3/3 00:00 621 417k +6.36 e2-e3
4/4 00:00 2k 774k +7.20 Ng1-f3 h7-h5
5/6 00:00 8k 1,480k +6.84 h2-h4 Nb8-c6 d2-d3 Nc6-a5 Ng1-f3 Na5-c6
6/8 00:00 12k 1,622k +6.85 h2-h4 Nb8-c6 e2-e3 Nc6-e5 b2-b3 Ne5-f3+ g2xf3
7/8 00:00 19k 1,621k +6.95 e2-e3 Nb8-c6 Qd1-e2 Nc6-e5 h2-h4 Ne5-f3+ Ng1xf3
8/12 00:00 43k 1,593k +6.75 Nb1-c3 Nb8-c6 e2-e3 Nc6-b4 Ke1-e2 g7-g5 a2-a3
9/16 00:00 156k 1,613k +6.66 Nb1-c3 Nb8-c6 e2-e3 Nc6-b4 Bf1-c4 d7-d5 Nc3xd5 Ke8-d8 Ke1-f1 Nb4xd5 Bc4xd5 c7-c6
10/15 00:00 281k 1,626k +6.65 e2-e3 Nb8-a6 c2-c3 Ng8-h6 Qd1-e2 Na6-b4 c3xb4
11/19 00:00 513k 1,600k +6.59 e2-e3 Nb8-c6 Bf1-b5 Nc6-b4 Bb5-a4 b7-b5 Ba4-b3 e7-e6 Qd1-e2 Bf8-d6 f2-f4 Nb4-d3+ Ke1-f1 Nd3xc1
12/21 00:00 794k 1,581k +6.63 e2-e3 Nb8-c6 Nb1-c3 Nc6-b4 Qd1-f3 e7-e6 Bf1-b5 c7-c6 Bb5-a4 b7-b5 Ba4-b3 Nb4-d3+ Ke1-f1
13/21 00:00 1,139k 1,546k +6.61 e2-e3 Nb8-c6 Nb1-c3 Nc6-b4 g2-g4 Ng8-f6 Bf1-c4 Nf6-e4 Bc4xf7+ Ke8xf7 Qd1-f3+ Kf7-e8
14/19 00:00 1,454k 1,544k +6.57 e2-e3 Nb8-c6 Nb1-c3 Nc6-b4 Bf1-b5 e7-e6 Qd1-e2 Ng8-f6 Bb5-a4 c7-c6 Ra1-b1 Nb4-d3+ c2xd3 Bf8-d6 f2-f4
15/28 00:01 2,184k 1,529k +6.62 e2-e3 Nb8-c6 Qd1-e2 Ng8-f6 Ng1-f3 Nc6-b4 Nb1-a3 Nb4-d3+ c2xd3 d7-d6 Nf3-g5 Bc8-g4 Na3-b5 Bg4xe2 Nb5xc7+ Ke8-d8 Ke1xe2
16/25 00:01 2,443k 1,533k +6.61 e2-e3 Nb8-c6 Qd1-e2 Ng8-f6 Ng1-f3 Nf6-d5 Nb1-a3 Nc6-d4 Qe2-d3 Nd4-b3 Qd3xb3 c7-c6 Nf3-d4 e7-e6
17/24 00:02 3,376k 1,534k +6.63 e2-e3 Nb8-c6 Qd1-e2 Nc6-b4 Nb1-a3 Ng8-f6 h2-h4 e7-e6 Ng1-h3 Nf6-d5 Rh1-h2 Nb4-d3+ c2xd3 Nd5-b4 Nh3-g5
18/28 00:03 5,972k 1,499k +6.64 e2-e3 Nb8-c6 Ng1-f3 Nc6-b4 Nb1-a3 Ng8-f6 Bf1-c4 d7-d5 Bc4-b5+ c7-c6 Bb5-a4 Nf6-e4 Nf3-d4 Ne4-c5 Rh1-g1 Nc5xa4 Na3-b1 Nb4-d3+ c2xd3 Na4-b6 Qd1-c2
19/28 00:05 7,963k 1,484k +6.66 e2-e3 Nb8-c6 Bf1-b5 Nc6-b4 Ng1-e2 c7-c6 Ne2-d4 c6xb5 Qd1-e2 e7-e6 Nb1-c3 Nb4-c6 Nd4xc6 Bf8-d6 Nc6xa7
20/25 00:06 10,093k 1,469k +6.67 e2-e3 Nb8-c6 Bf1-b5 Nc6-b4 Ng1-e2 c7-c6 Ne2-d4 c6xb5 Qd1-e2 a7-a6 b2-b3 Nb4-c6 Bc1-b2 Nc6xd4 e3xd4 d7-d5 f2-f4
21/29 00:15 22,076k 1,419k +6.79 e2-e3 Nb8-c6 Qd1-e2 g7-g6 Ng1-f3 Bf8-g7 Nb1-c3 Nc6-b4 Qe2-c4 Nb4-d3+ Ke1-e2 c7-c6 c2xd3 d7-d6 Nf3-g5
22/32 00:51 67,402k 1,315k +7.26 e2-e3 Nb8-c6 Ng1-f3 Nc6-b4 Nf3-g5 Ng8-h6 Nb1-a3 e7-e5 Ng5-e6 Bf8-c5 f2-f4 f7xe6 Na3-b5 c7-c6 Nb5-c3 Nb4-d5 f4xe5 Bc5-b6
23/36+ 02:15 166,462k 1,229k +7.68 Nb1-c3
23/36 04:50 342,689k 1,181k +8.33 e2-e4 Nb8-c6 Ng1-f3 e7-e6 Nb1-c3 Bf8-d6 Nf3-g5 Nc6-d4 Nc3-b5 Bd6-f4 Nb5xd4 e6-e5 Nd4-b5 c7-c6 Ng5xh7 c6xb5 Qd1-e2 a7-a6
24/26+ 05:10 365,483k 1,177k +8.48 e2-e4
24/35 07:56 557,385k 1,170k +8.80 e2-e4 Nb8-c6 Ng1-f3 h7-h5 e4-e5 e7-e6 Qd1-e2 g7-g6 d2-d4 h5-h4 d4-d5 Bf8-e7 Nb1-c3 Nc6-b4 Nf3-d4 e6xd5
25/35 12:30 866,862k 1,155k +9.15 e2-e4 e7-e6 Ng1-f3 Bf8-e7 Qd1-e2 b7-b6 Nf3-e5 Ng8-f6 Qe2-f3 h7-h5 Ne5xf7 Bc8-b7 Qf3-h3 Ke8xf7 e4-e5 Nf6-e8 d2-d4
26/37 35:17 2,423,574k 1,144k +9.58 e2-e4 h7-h5 Ng1-f3 Ng8-h6 e4-e5 c7-c6 e5-e6 d7-d6 Nf3-e5 f7-f6 Qd1xh5+ Ke8-d8 Ne5-f7+ Kd8-c7 g2-g4 Kc7-b6 g4-g5 Nh6-g8 Qh5-g4 f6xg5
27/38 1:04:10 4,399,426k 1,143k +9.72 e2-e4 Ng8-h6 Ng1-f3 Nh6-g4 Qd1-e2 e7-e6 Nf3-g5 h7-h5 d2-d4 b7-b6 e4-e5 Nb8-c6 Ng5xe6 d7xe6 c2-c3 Nc6-b8 d4-d5 e6xd5 h2-h3 Bc8-a6 h3xg4 Ba6xe2 Ke1xe2

[RubiChess]

What is the fun in giving NNUE engines positions they never were trained on and wondering about evaluation?

Again. And again. And again...

[Uri Blass]

What is the fun in giving NNUE engines positions they never were trained on and wondering about evaluation?

Again. And again. And again...

I do not see a reason not to train NNUE engines in these type of positions.
i suspect that the main problem is the fact that number of moves is totally unimportant in testing.

I would like to have a decision that the score of a win in testing is not 1 but something like 0.9+0.1*0.99^number of moves so a faster win is better.

Maybe I am wrong but my intuition is that this type of decision is going to lead to NNUE engines that are practically not weaker but mate faster(if people are afraid of reduction of playing strength then they can check at least for patches that pass that they do not get less than 50% when the score is as today with no additional games).