Speed up eval/move-gen/SEE; remove EvalLookForDanger, overhaul _EvalKing

SPRT-confirmed improvement (1814 games, st 4.0): candidate nps=1498118
vs head=1378448, avg depth 14.30 vs 13.87, ELO=+24.2+/-16.0,
LOWER95=0.5118, LLR=+3.126 crossing the H1 bound.

Most of the diff is non-semantic speed work across eval.c, generate.c,
movesup.c, see.c, probe.c and friends. Two real semantic changes:
removed EvalLookForDanger and overhauled _EvalKing's king-safety scoring.

Co-Authored-By: Claude Sonnet 5 <[email protected]>

1 files changed, 14 insertions, 21 deletions

diff --git a/src/probe.c b/src/probe.c
index 269a1b9..d08ef75 100644
--- a/src/probe.c
+++ b/src/probe.c
@@ -515,28 +515,21 @@ Return value:
             (mv.cTo == cTo) &&
             (mv.pPromoted == pPromotes))
         {
-            // Root-only: use the DTZ ply count tb_probe_root already
-            // packed into uResult (no extra probe) to report a real,
-            // ply-distance-aware mate score instead of the flat
-            // EGTB_WIN_SCORE/LOSS_SCORE used by in-search probes
-            // (ProbeEGTB/RecognLookup) -- those stay WDL-only and
-            // untouched. util.c's SCORE-to-string formatter already
-            // treats any |score| >= NMATE as a mate score and derives
-            // the move count as (INFINITY_SCORE - iScore) / 2, so
-            // INFINITY_SCORE -/+ uDtz round-trips through that
-            // unmodified as "+MATE<n>"/"-MATE<n>". Clamp defensively:
-            // a uDtz large enough to push the score below NMATE would
-            // just silently fall back to plain-decimal display, not
-            // corrupt anything, but keep it a real mate score at every
-            // legal DTZ value we could plausibly see (TB_MAX_PLY=256
-            // exceeds NMATE's 200-point margin in principle, however
-            // unlikely in practice for <=7-man tables).
-            ULONG uDtz = TB_GET_DTZ(uResult);
-            if (uDtz >= (+INFINITY_SCORE - NMATE)) uDtz = (+INFINITY_SCORE - NMATE - 1);
+            // DTZ (distance to the next fifty-move-rule-zeroing move)
+            // is not the same thing as distance to mate -- a previous
+            // version of this code reported (INFINITY_SCORE -/+ uDtz)
+            // as a mate score, which is wrong in general: DTZ can be
+            // small in a dead-won position that's still many plies
+            // from actual mate (e.g. right as pos->uFifty climbs
+            // towards 100, a reset becomes forced soon even though
+            // mate itself is nowhere near), producing a bogus
+            // "+MATE<n>"/"-MATE<n>" display untethered from reality.
+            // The draw/cursed-win/blessed-loss cases already returned
+            // FALSE above, so uWdl here is only ever TB_WIN or
+            // TB_LOSS -- report the same flat mate-magnitude score
+            // ProbeEGTB/RecognLookup use at interior nodes instead.
             *pmv = mv;
-            *piScore = (TB_WIN == uWdl) ?
-                (SCORE)(+INFINITY_SCORE - uDtz) :
-                (SCORE)(-INFINITY_SCORE + uDtz);
+            *piScore = (TB_WIN == uWdl) ? +NMATE : -NMATE;
             return(TRUE);
         }
     }