VVLTC search tune

This set of parameters was derived from 3 tuning attempts:

    https://tests.stockfishchess.org/tests/view/65d19ab61d8e83c78bfd8436 (80+0.8 x8, ~40k games)
    Then tuned with one of linrock's early L1-3072 nets:
    https://tests.stockfishchess.org/tests/view/65def7b04b19edc854ebdec8 (VVLTC, ~36k games)
    Starting from the result of this tuning, the parameters were then tuned with the current master net:
    https://tests.stockfishchess.org/tests/view/65f11c420ec64f0526c46fc4 (VVLTC, ~45k games)

Additionally, at the start of the third tuning phase, 2 parameters were manually changed:

    Notably, the triple extension margin was decreased from 78 to 22. This idea was given by Vizvezdenec:
    https://tests.stockfishchess.org/tests/view/65f0a2360ec64f0526c46752.
    The PvNode extension margin was also adjusted from 50 to 40.

This tune also differs from previous tuning attempts by tuning the evaluation thresholds for smallnet and psqt-only.
The former was increased through the tuning, and this is hypothesized to scale better at VVLTC,
although there is not much evidence of it.

Passed VVLTC 1st sprt: https://tests.stockfishchess.org/tests/view/65f6761d0ec64f0526c4be88
LLR: 2.94 (-2.94,2.94) <0.00,2.00>
Total: 44688 W: 11421 L: 11140 D: 22127
Ptnml(0-2): 1, 4170, 13722, 4449, 2

Passed VVLTC 2nd sprt: https://tests.stockfishchess.org/tests/view/65fa31a30ec64f0526c4f611
LLR: 2.94 (-2.94,2.94) <0.50,2.50>
Total: 27450 W: 7057 L: 6778 D: 13615
Ptnml(0-2): 4, 2545, 8346, 2828, 2

STC Elo estimate: https://tests.stockfishchess.org/tests/view/65fd3e540ec64f0526c521ae
Elo: -7.84 ± 1.8 (95%) LOS: 0.0%
Total: 40000 W: 9899 L: 10802 D: 19299
Ptnml(0-2): 203, 5221, 10025, 4378, 173
nElo: -14.91 ± 3.4 (95%) PairsRatio: 0.84

closes #5130

Bench: 1876107

src/evaluate.h

@@ -29,7 +29,7 @@ class Position;
 
 namespace Eval {
 
-constexpr inline int SmallNetThreshold = 1050, PsqtOnlyThreshold = 2500;
+constexpr inline int SmallNetThreshold = 1165, PsqtOnlyThreshold = 2500;
 
 // The default net name MUST follow the format nn-[SHA256 first 12 digits].nnue
 // for the build process (profile-build and fishtest) to work. Do not change the

src/search.cpp

@@ -55,9 +55,9 @@ namespace {
 
 // Futility margin
 Value futility_margin(Depth d, bool noTtCutNode, bool improving, bool oppWorsening) {
-    Value futilityMult       = 122 - 46 * noTtCutNode;
-    Value improvingDeduction = 57 * improving * futilityMult / 32;
-    Value worseningDeduction = (331 + 45 * improving) * oppWorsening * futilityMult / 1024;
+    Value futilityMult       = 118 - 44 * noTtCutNode;
+    Value improvingDeduction = 53 * improving * futilityMult / 32;
+    Value worseningDeduction = (309 + 47 * improving) * oppWorsening * futilityMult / 1024;
 
     return futilityMult * d - improvingDeduction - worseningDeduction;
 }
@@ -69,15 +69,15 @@ constexpr int futility_move_count(bool improving, Depth depth) {
 // Add correctionHistory value to raw staticEval and guarantee evaluation does not hit the tablebase range
 Value to_corrected_static_eval(Value v, const Worker& w, const Position& pos) {
     auto cv = w.correctionHistory[pos.side_to_move()][pawn_structure_index<Correction>(pos)];
-    v += cv * std::abs(cv) / 11450;
+    v += cv * std::abs(cv) / 11175;
     return std::clamp(v, VALUE_TB_LOSS_IN_MAX_PLY + 1, VALUE_TB_WIN_IN_MAX_PLY - 1);
 }
 
 // History and stats update bonus, based on depth
-int stat_bonus(Depth d) { return std::min(249 * d - 327, 1192); }
+int stat_bonus(Depth d) { return std::min(223 * d - 332, 1258); }
 
 // History and stats update malus, based on depth
-int stat_malus(Depth d) { return std::min(516 * d - 299, 1254); }
+int stat_malus(Depth d) { return std::min(536 * d - 299, 1353); }
 
 // Add a small random component to draw evaluations to avoid 3-fold blindness
 Value value_draw(size_t nodes) { return VALUE_DRAW - 1 + Value(nodes & 0x2); }
@@ -302,12 +302,12 @@ void Search::Worker::iterative_deepening() {
 
             // Reset aspiration window starting size
             Value avg = rootMoves[pvIdx].averageScore;
-            delta     = 9 + avg * avg / 12800;
+            delta     = 10 + avg * avg / 12493;
             alpha     = std::max(avg - delta, -VALUE_INFINITE);
             beta      = std::min(avg + delta, VALUE_INFINITE);
 
             // Adjust optimism based on root move's averageScore (~4 Elo)
-            optimism[us]  = 130 * avg / (std::abs(avg) + 90);
+            optimism[us]  = 132 * avg / (std::abs(avg) + 89);
             optimism[~us] = -optimism[us];
 
             // Start with a small aspiration window and, in the case of a fail
@@ -494,7 +494,7 @@ void Search::Worker::clear() {
         for (StatsType c : {NoCaptures, Captures})
             for (auto& to : continuationHistory[inCheck][c])
                 for (auto& h : to)
-                    h->fill(-71);
+                    h->fill(-67);
 
     for (size_t i = 1; i < reductions.size(); ++i)
         reductions[i] = int((19.80 + std::log(size_t(options["Threads"])) / 2) * std::log(i));
@@ -729,7 +729,7 @@ Value Search::Worker::search(
     // Use static evaluation difference to improve quiet move ordering (~9 Elo)
     if (((ss - 1)->currentMove).is_ok() && !(ss - 1)->inCheck && !priorCapture)
     {
-        int bonus = std::clamp(-14 * int((ss - 1)->staticEval + ss->staticEval), -1621, 1238);
+        int bonus = std::clamp(-13 * int((ss - 1)->staticEval + ss->staticEval), -1578, 1291);
         bonus     = bonus > 0 ? 2 * bonus : bonus / 2;
         thisThread->mainHistory[~us][((ss - 1)->currentMove).from_to()] << bonus;
         if (type_of(pos.piece_on(prevSq)) != PAWN && ((ss - 1)->currentMove).type_of() != PROMOTION)
@@ -752,7 +752,7 @@ Value Search::Worker::search(
     // If eval is really low check with qsearch if it can exceed alpha, if it can't,
     // return a fail low.
     // Adjust razor margin according to cutoffCnt. (~1 Elo)
-    if (eval < alpha - 462 - (296 - 145 * ((ss + 1)->cutoffCnt > 3)) * depth * depth)
+    if (eval < alpha - 488 - (289 - 142 * ((ss + 1)->cutoffCnt > 3)) * depth * depth)
     {
         value = qsearch<NonPV>(pos, ss, alpha - 1, alpha);
         if (value < alpha)
@@ -763,21 +763,21 @@ Value Search::Worker::search(
     // The depth condition is important for mate finding.
     if (!ss->ttPv && depth < 12
         && eval - futility_margin(depth, cutNode && !ss->ttHit, improving, opponentWorsening)
-               - (ss - 1)->statScore / 287
+               - (ss - 1)->statScore / 267
              >= beta
         && eval >= beta && eval < VALUE_TB_WIN_IN_MAX_PLY && (!ttMove || ttCapture))
         return beta > VALUE_TB_LOSS_IN_MAX_PLY ? (eval + beta) / 2 : eval;
 
     // Step 9. Null move search with verification search (~35 Elo)
-    if (!PvNode && (ss - 1)->currentMove != Move::null() && (ss - 1)->statScore < 16211
+    if (!PvNode && (ss - 1)->currentMove != Move::null() && (ss - 1)->statScore < 16878
         && eval >= beta && eval >= ss->staticEval && ss->staticEval >= beta - 20 * depth + 314
         && !excludedMove && pos.non_pawn_material(us) && ss->ply >= thisThread->nmpMinPly
         && beta > VALUE_TB_LOSS_IN_MAX_PLY)
     {
         assert(eval - beta >= 0);
 
         // Null move dynamic reduction based on depth and eval
-        Depth R = std::min(int(eval - beta) / 151, 6) + depth / 3 + 4;
+        Depth R = std::min(int(eval - beta) / 144, 6) + depth / 3 + 4;
 
         ss->currentMove         = Move::null();
         ss->continuationHistory = &thisThread->continuationHistory[0][0][NO_PIECE][0];
@@ -825,7 +825,7 @@ Value Search::Worker::search(
     // Step 11. ProbCut (~10 Elo)
     // If we have a good enough capture (or queen promotion) and a reduced search returns a value
     // much above beta, we can (almost) safely prune the previous move.
-    probCutBeta = beta + 168 - 64 * improving;
+    probCutBeta = beta + 170 - 64 * improving;
     if (
       !PvNode && depth > 3
       && std::abs(beta) < VALUE_TB_WIN_IN_MAX_PLY
@@ -881,7 +881,7 @@ Value Search::Worker::search(
 moves_loop:  // When in check, search starts here
 
     // Step 12. A small Probcut idea, when we are in check (~4 Elo)
-    probCutBeta = beta + 410;
+    probCutBeta = beta + 409;
     if (ss->inCheck && !PvNode && ttCapture && (tte->bound() & BOUND_LOWER)
         && tte->depth() >= depth - 4 && ttValue >= probCutBeta
         && std::abs(ttValue) < VALUE_TB_WIN_IN_MAX_PLY && std::abs(beta) < VALUE_TB_WIN_IN_MAX_PLY)
@@ -964,15 +964,15 @@ Value Search::Worker::search(
                 {
                     Piece capturedPiece = pos.piece_on(move.to_sq());
                     int   futilityEval =
-                      ss->staticEval + 298 + 288 * lmrDepth + PieceValue[capturedPiece]
+                      ss->staticEval + 297 + 284 * lmrDepth + PieceValue[capturedPiece]
                       + thisThread->captureHistory[movedPiece][move.to_sq()][type_of(capturedPiece)]
                           / 7;
                     if (futilityEval < alpha)
                         continue;
                 }
 
                 // SEE based pruning for captures and checks (~11 Elo)
-                if (!pos.see_ge(move, -202 * depth))
+                if (!pos.see_ge(move, -203 * depth))
                     continue;
             }
             else
@@ -984,24 +984,24 @@ Value Search::Worker::search(
                   + thisThread->pawnHistory[pawn_structure_index(pos)][movedPiece][move.to_sq()];
 
                 // Continuation history based pruning (~2 Elo)
-                if (lmrDepth < 6 && history < -4125 * depth)
+                if (lmrDepth < 6 && history < -4040 * depth)
                     continue;
 
                 history += 2 * thisThread->mainHistory[us][move.from_to()];
 
-                lmrDepth += history / 5686;
+                lmrDepth += history / 5637;
 
                 // Futility pruning: parent node (~13 Elo)
                 if (!ss->inCheck && lmrDepth < 15
-                    && ss->staticEval + (bestValue < ss->staticEval - 55 ? 153 : 58)
-                           + 118 * lmrDepth
+                    && ss->staticEval + (bestValue < ss->staticEval - 59 ? 141 : 58)
+                           + 125 * lmrDepth
                          <= alpha)
                     continue;
 
                 lmrDepth = std::max(lmrDepth, 0);
 
                 // Prune moves with negative SEE (~4 Elo)
-                if (!pos.see_ge(move, -26 * lmrDepth * lmrDepth))
+                if (!pos.see_ge(move, -27 * lmrDepth * lmrDepth))
                     continue;
             }
         }
@@ -1021,11 +1021,11 @@ Value Search::Worker::search(
             // so changing them requires tests at these types of time controls.
             // Recursive singular search is avoided.
             if (!rootNode && move == ttMove && !excludedMove
-                && depth >= 4 - (thisThread->completedDepth > 29) + ss->ttPv
+                && depth >= 4 - (thisThread->completedDepth > 30) + ss->ttPv
                 && std::abs(ttValue) < VALUE_TB_WIN_IN_MAX_PLY && (tte->bound() & BOUND_LOWER)
                 && tte->depth() >= depth - 3)
             {
-                Value singularBeta  = ttValue - (58 + 55 * (ss->ttPv && !PvNode)) * depth / 64;
+                Value singularBeta  = ttValue - (58 + 58 * (ss->ttPv && !PvNode)) * depth / 64;
                 Depth singularDepth = newDepth / 2;
 
                 ss->excludedMove = move;
@@ -1040,11 +1040,11 @@ Value Search::Worker::search(
                     // We make sure to limit the extensions in some way to avoid a search explosion
                     if (!PvNode && ss->multipleExtensions <= 16)
                     {
-                        extension = 2 + (value < singularBeta - 78 && !ttCapture);
+                        extension = 2 + (value < singularBeta - 22 && !ttCapture);
                         depth += depth < 14;
                     }
                     if (PvNode && !ttCapture && ss->multipleExtensions <= 5
-                        && value < singularBeta - 50)
+                        && value < singularBeta - 37)
                         extension = 2;
                 }
 
@@ -1079,7 +1079,7 @@ Value Search::Worker::search(
             else if (PvNode && move == ttMove && move.to_sq() == prevSq
                      && thisThread->captureHistory[movedPiece][move.to_sq()]
                                                   [type_of(pos.piece_on(move.to_sq()))]
-                          > 4315)
+                          > 4026)
                 extension = 1;
         }
 
@@ -1129,10 +1129,10 @@ Value Search::Worker::search(
         ss->statScore = 2 * thisThread->mainHistory[us][move.from_to()]
                       + (*contHist[0])[movedPiece][move.to_sq()]
                       + (*contHist[1])[movedPiece][move.to_sq()]
-                      + (*contHist[3])[movedPiece][move.to_sq()] - 4587;
+                      + (*contHist[3])[movedPiece][move.to_sq()] - 4723;
 
         // Decrease/increase reduction for moves with a good/bad history (~8 Elo)
-        r -= ss->statScore / 14956;
+        r -= ss->statScore / 13659;
 
         // Step 17. Late moves reduction / extension (LMR, ~117 Elo)
         if (depth >= 2 && moveCount > 1 + rootNode)
@@ -1151,7 +1151,7 @@ Value Search::Worker::search(
             {
                 // Adjust full-depth search based on LMR results - if the result
                 // was good enough search deeper, if it was bad enough search shallower.
-                const bool doDeeperSearch    = value > (bestValue + 48 + 2 * newDepth);  // (~1 Elo)
+                const bool doDeeperSearch    = value > (bestValue + 47 + 2 * newDepth);  // (~1 Elo)
                 const bool doShallowerSearch = value < bestValue + newDepth;             // (~2 Elo)
 
                 newDepth += doDeeperSearch - doShallowerSearch;
@@ -1269,7 +1269,7 @@ Value Search::Worker::search(
                 else
                 {
                     // Reduce other moves if we have found at least one score improvement (~2 Elo)
-                    if (depth > 2 && depth < 12 && beta < 13665 && value > -12276)
+                    if (depth > 2 && depth < 12 && beta < 14206 && value > -12077)
                         depth -= 2;
 
                     assert(depth > 0);
@@ -1312,7 +1312,7 @@ Value Search::Worker::search(
     // Bonus for prior countermove that caused the fail low
     else if (!priorCapture && prevSq != SQ_NONE)
     {
-        int bonus = (depth > 5) + (PvNode || cutNode) + ((ss - 1)->statScore < -14446)
+        int bonus = (depth > 5) + (PvNode || cutNode) + ((ss - 1)->statScore < -14963)
                   + ((ss - 1)->moveCount > 11)
                   + (!ss->inCheck && bestValue <= ss->staticEval - 150);
         update_continuation_histories(ss - 1, pos.piece_on(prevSq), prevSq,
@@ -1472,7 +1472,7 @@ Value Search::Worker::qsearch(Position& pos, Stack* ss, Value alpha, Value beta,
         if (bestValue > alpha)
             alpha = bestValue;
 
-        futilityBase = ss->staticEval + 221;
+        futilityBase = ss->staticEval + 226;
     }
 
     const PieceToHistory* contHist[] = {(ss - 1)->continuationHistory,
@@ -1552,7 +1552,7 @@ Value Search::Worker::qsearch(Position& pos, Stack* ss, Value alpha, Value beta,
                 continue;
 
             // Do not search moves with bad enough SEE values (~5 Elo)
-            if (!pos.see_ge(move, -79))
+            if (!pos.see_ge(move, -78))
                 continue;
         }
 
@@ -1620,7 +1620,7 @@ Value Search::Worker::qsearch(Position& pos, Stack* ss, Value alpha, Value beta,
 
 Depth Search::Worker::reduction(bool i, Depth d, int mn, int delta) {
     int reductionScale = reductions[d] * reductions[mn];
-    return (reductionScale + 1091 - delta * 759 / rootDelta) / 1024 + (!i && reductionScale > 950);
+    return (reductionScale + 1107 - delta * 725 / rootDelta) / 1024 + (!i && reductionScale > 956);
 }
 
 namespace {
@@ -1709,7 +1709,7 @@ void update_all_stats(const Position& pos,
 
     if (!pos.capture_stage(bestMove))
     {
-        int bestMoveBonus = bestValue > beta + 167 ? quietMoveBonus      // larger bonus
+        int bestMoveBonus = bestValue > beta + 173 ? quietMoveBonus      // larger bonus
                                                    : stat_bonus(depth);  // smaller bonus
 
         // Increase stats for the best move in case it was a quiet move