Use stockfish piece values

src/starfish/internal/board.gleam

···
   King
 }
 
-pub const pawn_value = 100
+// Values taken from https://hxim.github.io/Stockfish-Evaluation-Guide/
 
-pub const knight_value = 300
+pub const pawn_value = 124
 
-pub const bishop_value = 300
+pub const knight_value = 781
 
-pub const rook_value = 500
+pub const bishop_value = 825
 
-pub const queen_value = 900
+pub const rook_value = 1276
+
+pub const queen_value = 2538
 
-pub const king_value = 1000
+pub const king_value = 10_000
 
 pub fn piece_value(piece: Piece) -> Int {
   case piece {

src/starfish/internal/game.gleam

···
 pub fn is_insufficient_material(game: Game) -> Bool {
   game.black_pieces.pawn_material == 0
   && game.white_pieces.pawn_material == 0
-  && {
-    game.black_pieces.non_pawn_material == board.bishop_value
-    || game.black_pieces.non_pawn_material == board.knight_value
-    || game.black_pieces.non_pawn_material == 0
-  }
-  && {
-    game.white_pieces.non_pawn_material == board.bishop_value
-    || game.white_pieces.non_pawn_material == board.knight_value
-    || game.white_pieces.non_pawn_material == 0
-  }
+  && game.black_pieces.non_pawn_material <= board.bishop_value
+  && game.white_pieces.non_pawn_material <= board.bishop_value
 }
 
 pub fn is_threefold_repetition(game: Game) -> Bool {
···
 
 const phase_multiplier = 128
 
-/// About queen + rook, so one major piece per side
-const endgame_material = 1400
+// Values taken from https://hxim.github.io/Stockfish-Evaluation-Guide/
+
+/// About queen + rook, so one major piece per side. If total material is less
+/// than this, then we are completely in the endgame. 
+const endgame_material = 3915
 
-/// Below this material limit, the endgame weight is zero. this is about enough
-/// for three minor pieces to be captured.
-const middlegame_material = 3000
+/// Above this material limit, the endgame weight is zero. this is about enough
+/// for three minor pieces to be captured .
+const middlegame_material = 15_258
 
 pub fn phase(game: Game) -> Int {
   let non_pawn_material =
···
     piece_square_score_midgame:,
     piece_square_score_endgame:,
   ) = pieces
+
+  // Pawns become much more valuable in the endgame, about 1.5x
+  let pawn_material_endgame = pawn_material * 15 / 10
+
   non_pawn_material
-  + pawn_material
+  + interpolate_phase(pawn_material, pawn_material_endgame, phase)
   + interpolate_phase(
     piece_square_score_midgame,
     piece_square_score_endgame,

src/starfish/internal/piece_table.gleam

···
 
 import starfish/internal/board
 
-// Values taken from https://www.chessprogramming.org/Simplified_Evaluation_Function
+// Values taken from https://hxim.github.io/Stockfish-Evaluation-Guide/
 
 // We use tuples for constant time accessing. The values are constant so we don't
 // need to worry about the cost of updating.