Simplify TT move extension formula

[FauziAkram]

Simplify TT move extension formula

Passed STC:
LLR: 2.94 (-2.94,2.94) <-1.75,0.25>
Total: 110560 W: 28603 L: 28469 D: 53488
Ptnml(0-2): 303, 12146, 30254, 12268, 309
https://tests.stockfishchess.org/tests/view/6958ed4dd844c1ce7cc7e98e

No functional change

[robertnurnberg]

Judging by the comment, this would need a matetrack run as well.

Not sure how to check effect on retrograde analysis. @pb00068

[vondele]

I think this was added with the comment to make sure it stays for this reason?
Edit: but this is non-functional now?

[FauziAkram]

Yes, this is non-functional.
Can someone please run a matetrack run?

[pb00067]

Judging by the comment, this would need a matetrack run as well.

Not sure how to check effect on retrograde analysis. @pb00068

Unfortunately this patch breaks the retrograde analysis ability of SF,
tried on position FEN: 2K1k1br/3p1n1r/2p2pN1/P2p1N2/2P3pP/5B2/P2P4/8 b - - 9 9
and going backwards the moves:

startpos: 6br/1KNp1n1r/2p2p2/P1ppRP2/1kP3pP/3PBB2/PN1P4/8 w - - 0 1

[robertnurnberg]

Thanks! Could you double-check that master is still fine?

So the patch is indeed functional, it just doesn't change bench, correct?

[pb00067]

Thanks! Could you double-check that master is still fine?

Good catch.
Effectively also current master has problems on that particular FEN: 2K1k1br/3p1n1r/2p2pN1/P2p1N2/2P3pP/5B2/P2P4/8 b - - 9 9

P.S.: In order to re-establish the previous backward analysis capability it seems that now we need following condition:

  if (move == ttData.move && ((is_valid(ttData.value) && is_decisive(ttData.value)) || ttData.depth > 1))
           newDepth = std::max(newDepth, 1);

Edit 2: condition above sometimes work sometimes not.

[robertnurnberg]

I think the best strategy would be to find the (simplification?) commit that broke the retrograde analysis fix you introduced and then possibly revert that (while also adding a comment in the source)?

[pb00067]

I think the best strategy would be to find the (simplification?) commit that broke the retrograde analysis fix you introduced

Not an easy task to find it.

[robertnurnberg]

Starting point would probably be a git-blame on the parts of the code you introduced and believe are critically important.

Alternative explanation is that your fix was not that robust in the first place? So maybe it "broke" due to some unrelated patch.

[pb00067]

Alternative explanation is that your fix was not that robust in the first place?

Since TT entries might be overriden by other positions, no fix is really robust in that sense when relying on TT only.
In meantime I test https://tests.stockfishchess.org/tests/view/695ce092912b7ff140de60c6 which restores more or less the retrograde analysis capability we had before.

Finding a robust solution would surely require lot of more code and further research...

[robertnurnberg]

So, similarly to #6531 (comment), the comparison with master for this patch is:

Patch

> python retrocheck.py
Loaded 6554 FENs, with max(|bm|) = 126.

retrocheck started for ./stockfish on matetrackpv.epd with --nodes 1000000 --nodesFill 100000 ...
100%|##########| 97/97 [1:05:58<00:00, 40.81s/it] 


Using ./stockfish on matetrackpv.epd with --nodes 1000000 --nodesFill 100000
Engine ID:     Stockfish dev-20260104-dc58db0e
Total FENs:    6554
Found mates:   4592
Best mates:    4312

Master:

> python retrocheck.py
Loaded 6554 FENs, with max(|bm|) = 126.

retrocheck started for ./stockfish on matetrackpv.epd with --nodes 1000000 --nodesFill 100000 ...
100%|#########################################| 97/97 [1:03:31<00:00, 39.29s/it]


Using ../stockfish/src/master on matetrackpv.epd with --nodes 1000000 --nodesFill 100000
Engine ID:     Stockfish dev-20260106-c27c1747
Total FENs:    6554
Found mates:   5294
Best mates:    5063

So a quite clearly regression for retrograde analysis, from 5294 to 4592 mates found.

Edit: Numbers are now corrected, after an earlier version of retrocheck.py has been fixed.

[Disservin]

based on matetrack results closing