Eval parameters tuning

Book: 3mvs 251-350_51933
TC: 100+1
Total/Win/Draw/Lose: 6242 / 1630 / 3100 / 1512
PTNML: 8 / 626 / 1733 / 748 / 6
WinRate: 50.95%
ELO: 6.07[1.80, 10.29]
LOS: 99.72
LLR: 2.93[-2.94, 2.94]

src/evaluate.cpp

@@ -104,12 +104,12 @@ Value Eval::evaluate(const Position& pos) {
   Value v = NNUE::evaluate(pos, true);
 
   // Scale nnue score according to material and optimism
-  Value optimism = pos.this_thread()->optimism[pos.side_to_move()] * abs(pos.material_diff() - v) / 512;
-  Value material = pos.material_sum() / 64;
-  v = (v * (601 + material) + optimism * (184 + material)) / 1024;
+  Value optimism = pos.this_thread()->optimism[pos.side_to_move()] * abs(pos.material_diff() - v) / 568;
+  Value material = pos.material_sum() / 40;
+  v = (v * (622 + material) + optimism * (171 + material)) / 1041;
 
   // Damp down the evaluation linearly when shuffling
-  v = v * (240 - pos.rule60_count()) / 136;
+  v = v * (258 - pos.rule60_count()) / 134;
 
   // Guarantee evaluation does not hit the mate range
   v = std::clamp(v, VALUE_MATED_IN_MAX_PLY + 1, VALUE_MATE_IN_MAX_PLY - 1);

src/nnue/evaluate_nnue.cpp

@@ -148,7 +148,7 @@ namespace Stockfish::Eval::NNUE {
     // overaligning stack variables with alignas() doesn't work correctly.
 
     constexpr uint64_t alignment = CacheLineSize;
-    constexpr int delta = 24;
+    constexpr int delta = 26;
 
 #if defined(ALIGNAS_ON_STACK_VARIABLES_BROKEN)
     TransformedFeatureType transformedFeaturesUnaligned[