Stockfish Testing Queue

Finished - 977 tests

18-12-25 31m tropism_passer diff
LLR: -2.96 (-2.94,2.94) [0.50,4.50]
Total: 20197 W: 4409 L: 4474 D: 11314
sprt @ 10+0.1 th 1 King tropism penalties for enemy passers in our kingFlank and Camp. These have great long-term potential in attacks, but as far as I can tell, we don't currently assign much weight to them beyond their immediate attacks (which, as pawns, are quite limited).
18-12-25 31m kingsafety_Viz diff
LLR: 0.97 (-2.94,2.94) [0.00,5.00]
Total: 2332 W: 545 L: 490 D: 1297
sprt @ 10+0.1 th 1 A tweak of @Vizvezdenec's promising idea: if the opponent has sufficient material, certain special cases may merit king safety evaluation regardless of kingAttackersCount. In addition to the tropism case by @Vizvezdenec, evaluate king safety if our king flank has no friendly pawns. Note that this differs from PawnlessFlank, which considers pawns of either color and is concentrated on the endgame rather than middlegame.
18-12-24 31m combo_O_SS_KA diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 51778 W: 11272 L: 11233 D: 29273
sprt @ 10+0.1 th 1 Merge @Chess13234's KingAttackWeightE40 (STC: 77K yellow) into this combo (previously, STC 82K yellow; LTC: 98K yellow).
18-12-24 31m tropism_DPA2 diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 13262 W: 2870 L: 2936 D: 7456
sprt @ 10+0.1 th 1 Two highly-related changes (both using double pawn attacks and king danger), so I think it's permissible to test them as one idea. Use double pawn attacks to refine both kingAttackersCount and tropism, where they are both currently ignored.
18-12-23 31m tropism_DPA^ diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 20046 W: 4333 L: 4366 D: 11347
sprt @ 10+0.1 th 1 Track these squares independently of b2, in a new Bitboard b3. Since b2 and b3 may overlap, the result is a greater-or-equal effect size compared to take 1.
18-12-23 31m tropism_DPA diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 18844 W: 4059 L: 4098 D: 10687
sprt @ 10+0.1 th 1 Double effect.
18-12-23 31m tropism_DPA diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 54757 W: 12060 L: 11922 D: 30775
sprt @ 10+0.1 th 1 Use double_pawn_attacks_bb in the calculation of b2 for king tropism.
18-12-22 31m tweak_kingRing diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 49744 W: 10884 L: 10853 D: 28007
sprt @ 10+0.1 th 1 I know many different ideas were tried with the introduction of double_pawn_attacks_bb...I hope I'm not duplicating previous work. Reorder this code so that kingRing is made smaller before kingAttackersCount[Them] is calculated, not after. Searching for a strong [0, 4] in its own right or a possibility to add to my developing combo.
18-12-22 31m kingAttackers_DPA diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 31536 W: 6825 L: 6802 D: 17909
sprt @ 10+0.1 th 1 Intuitively, if two pawns attack a square in the kingRing, then that square has two kingAttackers, not just the one.
18-12-22 31m kingAttackers_DPA diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 9415 W: 2044 L: 2129 D: 5242
sprt @ 10+0.1 th 1 Same idea, with the reordering included.
18-12-22 31m RestrictedPiece_passer diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 9235 W: 2024 L: 2110 D: 5101
sprt @ 10+0.1 th 1 Elo estimate is still neutral. Double again to quadruple effect.
18-12-22 31m RestrictedPiece_passer diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 20444 W: 4462 L: 4493 D: 11489
sprt @ 10+0.1 th 1 Elo estimates so far appear essentially neutral. Try double effect. (All ranks)
18-12-22 31m RestrictedPiece_passer^ diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 8942 W: 1930 L: 2018 D: 4994
sprt @ 10+0.1 th 1 Elo estimates so far appear essentially neutral. Try double effect. (Just the three furthest ranks)
18-12-21 31m RestrictedPiece_passer diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 16832 W: 3677 L: 3726 D: 9429
sprt @ 10+0.1 th 1 Same, but only the two most advanced ranks.
18-12-21 31m RestrictedPiece_passer diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 26948 W: 5922 L: 5921 D: 15105
sprt @ 10+0.1 th 1 Same, but only on the three most advanced ranks.
18-12-21 31m RestrictedPiece_passer^ diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 18360 W: 4013 L: 4054 D: 10293
sprt @ 10+0.1 th 1 Based on Bryan's analysis of Cscuile Game 111. For RestrictedPiece, consider not only the not-strongly-protected, enemy-attacked squares that we attack, but consider also the ones that are immediately in front of our passed pawns. The enemy is also obligated to defend these.
18-12-21 31m combo_O_SS_CP_t1 diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 15617 W: 3443 L: 3546 D: 8628
sprt @ 10+0.1 th 1 Combine everything with a positive score so far, and hope for a green.
18-12-21 31m combo_O_SS_ss diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 11156 W: 2374 L: 2494 D: 6288
sprt @ 10+0.1 th 1 @xoroshiro's ssTweak looked promising at LTC (July 25; 73K yellow) but one of its changes was never merged. Try it again.
18-12-20 31m combo_O_SS_ROP diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 15105 W: 3363 L: 3468 D: 8274
sprt @ 10+0.1 th 1 Add in another tweak of mine from late July, changing the logic of RookOnPawn.
18-12-20 31m combo_O_SS_t1 diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 43409 W: 9462 L: 9456 D: 24491
sprt @ 10+0.1 th 1 Add tweaks from @FauziAkram's test1 (Dec. 14; STC 86K yellow). Don't include CP. (Fixed bench)
18-12-19 31m combo_O_SS_CP diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 48616 W: 10761 L: 10733 D: 27122
sprt @ 10+0.1 th 1 Add @locutus2's connected_pawns (Aug. 15; STC 33K green; LTC 102K yellow). Does anyone know of any other recent promising [0, 4]s that weren't committed? Please let me know!
18-12-19 31m CandidatePF diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 12416 W: 2737 L: 2807 D: 6872
sprt @ 10+0.1 th 1 Bugfix. (The last version accidentally disabled passed pawn evaluation altogether.)
18-12-19 31m CandidatePF diff
LLR: -0.81 (-2.94,2.94) [0.00,4.00]
Total: 77 W: 5 L: 54 D: 18
sprt @ 10+0.1 th 1 Bryan's idea from David Grovesnor: Rnd 33. Only apply PassedFile to candidate passers. [0, 4] because this only indents existing code into an existing if; no complexity is added.
18-12-19 31m combo_O_SS diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 98421 W: 16495 L: 16334 D: 65592
sprt @ 60+0.6 th 1 Framework is completely empty, so speculative LTC for this 82K yellow. Low throughput (166).
18-12-19 31m combo_O_SRAP diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 6840 W: 1430 L: 1567 D: 3843
sprt @ 10+0.1 th 1 Combo: my tweak_outpost (STC 190K yellow; LTC 66K yellow) and @Rocky640's SpaceRankAndPawn (Aug. 29; STC 29K green; LTC 64K yellow).
18-12-19 31m combo_O_SS diff
LLR: -2.96 (-2.94,2.94) [0.00,4.00]
Total: 82471 W: 18020 L: 17861 D: 46590
sprt @ 10+0.1 th 1 Beginning to search for other nearly-successful [0, 4]s to push my own (STC 190K yellow; LTC 66K yellow) through. Try @protonspring's ps_shelter31 (Nov. 14; STC 125K yellow; LTC 77K yellow).
18-12-18 31m tweak_outpost diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 66615 W: 11194 L: 11134 D: 44287
sprt @ 60+0.6 th 1 Speculative LTC for 190K yellow. Low throughput (166).
18-12-18 31m tweak_outpost diff
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 190243 W: 41786 L: 41203 D: 107254
sprt @ 10+0.1 th 1 For bishops and knights that can reach an outpost, require that the outpost square be unoccupied--not only by friendly pieces, but also enemy ones. Ideally we would like to make sure the outpost isn't defended, but we can't easily in this part of the code. However, I assume that we don't typically search positions where enemies occupy outposts and are clearly hanging--one side would be clearly winning--so perhaps this can serve as an indirect way of assessing outposts' defense.
18-12-18 31m simplify_outpost diff
LLR: -2.94 (-2.94,2.94) [-3.00,1.00]
Total: 36565 W: 7912 L: 8148 D: 20505
sprt @ 10+0.1 th 1 Do we need to check that we don't occupy the outpost square? If we do, we can still reach that outpost--it just takes two moves rather than one.
18-12-18 31m noMobPawnThreat diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 12962 W: 2764 L: 2832 D: 7366
sprt @ 10+0.1 th 1 ThreatByPawnPush, double effect.
18-12-18 31m noMobPawnThreat^ diff
LLR: -2.97 (-2.94,2.94) [0.00,5.00]
Total: 23630 W: 5122 L: 5138 D: 13370
sprt @ 10+0.1 th 1 Derived from Bryan's ideas and analysis on Cscuile Game 27. Give extra ThreatByPawnPush if the enemy target has no mobility. Here, double.
18-12-18 31m noMobPawnThreat diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 12881 W: 2797 L: 2865 D: 7219
sprt @ 10+0.1 th 1 The analogous approach, applied to ThreatBySafePawn instead.
18-12-17 31m simplify_candidatePasse diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 93489 W: 20433 L: 20453 D: 52603
sprt @ 10+0.1 th 1 Experiment with the effect size of bonus in the simplified version. 3/4.
18-12-17 31m simplify_candidatePasse diff
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 75116 W: 16227 L: 16539 D: 42350
sprt @ 10+0.1 th 1 Experiment with the effect size of bonus in the simplified version. 1/4.
18-12-17 31m simplify_candidatePasse diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 22215 W: 4765 L: 4973 D: 12477
sprt @ 10+0.1 th 1 Much more radical version: eliminate candidate passers altogether. (Performance of / 2 and / 4 seems similar, so there's a small chance it makes no difference at all.)
18-12-17 31m simplify_candidatePasse diff
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 42369 W: 6903 L: 7124 D: 28342
sprt @ 60+0.6 th 1 LTC. If we have two doubled pawns, which would otherwise be considered passed, do we need to not consider the back pawn passed?
18-12-17 31m simplify_candidatePasse diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 16750 W: 3707 L: 3576 D: 9467
sprt @ 10+0.1 th 1 If we have two doubled pawns, which would otherwise be considered passed, do we need to not consider the back pawn passed?
18-12-17 31m BlockSqAttack diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 20364 W: 4424 L: 4455 D: 11485
sprt @ 10+0.1 th 1 More complex logic: if the blocking square is enemy-occupied, we attack it, and it is either attacked by our pawn or not doubly defended.
18-12-16 31m BlockSqAttack diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 20633 W: 4514 L: 4544 D: 11575
sprt @ 10+0.1 th 1 My last attempt, handling a sub-case (~attackedBy[Them][PAWN]) only, was -2.2 Elo while the original was +0.85. Sanity check: try the opposite case.
18-12-16 31m BlockSqAttack diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 10160 W: 2176 L: 2257 D: 5727
sprt @ 10+0.1 th 1 @Vizvezdenec's suggestion. Make sure the blockSq isn't pawn-defended. (If we capture, then they can recapture with a pawn, and our pawn cannot become a passer.)
18-12-16 31m BlockSqAttack diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 32712 W: 7110 L: 7081 D: 18521
sprt @ 10+0.1 th 1 A passed pawn is safe to advance not only if the blockSq is empty, but if it is enemy-occupied and attacked by one of our pawns. (Since the pawn is passed, the enemy is not a pawn. Thus, this anticipates unblocked passed pawns one move early--the blocker is forced to move away.)
18-12-16 31m PinnedPasser diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 24126 W: 5271 L: 5284 D: 13571
sprt @ 10+0.1 th 1 A passed pawn is not "free to advance" if it is pinned to our king. Consider this to be the same as the case where the block square is occupied.
18-12-16 31m PinOpacity diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 6776 W: 1444 L: 1542 D: 3790
sprt @ 10+0.1 th 1 No attacks through pieces pinned to our king.
18-12-15 31m simplify_RestrictedPiec diff
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 37078 W: 6125 L: 6030 D: 24923
sprt @ 60+0.6 th 1 LTC. Use ~stronglyProtected to calculate RestrictedPiece.
18-12-15 31m RookOnQueen diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 42330 W: 9279 L: 9202 D: 23849
sprt @ 10+0.1 th 1 Double effect. Though this applies to a very narrow case, it is a substantial bonus.
18-12-15 31m RookOnQueen^ diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 25220 W: 5447 L: 5455 D: 14318
sprt @ 10+0.1 th 1 So far, it seems that increased effect size is an improvement. Try triple effect.
18-12-15 31m RookOnQueen diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 17899 W: 3978 L: 4021 D: 9900
sprt @ 10+0.1 th 1 Quadruple effect (2 * ThreatByRook[QUEEN]).
18-12-15 31m simplify_RestrictedPiec diff
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 35924 W: 7978 L: 7885 D: 20061
sprt @ 10+0.1 th 1 A fairly intuitive simplification: ~stronglyProtected is quite similar to (and simpler than) ~attackedBy[Them][PAWN] & ~attackedBy2[Them]. The only difference is that this includes squares attacked twice by both sides.
18-12-15 31m RookOnQueen diff
LLR: -2.96 (-2.94,2.94) [0.00,5.00]
Total: 39752 W: 8742 L: 8678 D: 22332
sprt @ 10+0.1 th 1 Apply the bonus if the rook is aligned with the enemy king and the enemy queen.
18-12-15 31m RookOnQueen^ diff
LLR: -2.95 (-2.94,2.94) [0.00,5.00]
Total: 8046 W: 1279 L: 1370 D: 5397
sprt @ 60+0.6 th 1 LTC. Attempt to revive my threat_strongqueen idea with a RookOnPawn-style approach (i.e., based on the rook's rank). Use ThreatByRook[QUEEN] / 2 for now, since it worked well in my previous attempts.