I wonder how compactly you could represent most chess games. On a per-piece basis, each non-pawn would require 6 bits of location, plus 1 bit for whether it exi

I wonder how compactly you could represent most chess games. On a per-piece basis, each non-pawn would require 6 bits of location, plus 1 bit for whether it exists, for 16 * 7 = 112 bits. Each pawn can only access half of the squares at most (never row 1 or 8, capturing diagonally creates a fan of at most 32 squares, at minimum 21 for the flank pawns) so an additional 16 * 6 = 96. You would need to add an extension for promoted pawns, likely average out to never relevant. It could be stored as an illegal pawn position, so 208 bits in the normal case, but a variable-length format. With bitboards: There are 12 types of pieces, so could use 4 bitboards[0] to capture all of the information. 64 * 4 = 256 bits. I think the most compact would be a combination. One bitboard (64 bits) marking where pieces are, and then a list of 4 bit integers[0] noting which piece is which, ordered by how they show up on the bitboard. With 32 pieces that's an additional 128 bits, for 192 total. And the size could only go down from there, as pieces are removed. Writing this out, I see how 4 bitboards would be much simpler than 1 bitboard and a weirdly-indexed list that needs to be reordered every move. [0] Boards could be 1=white/0=black, 1=more/0=less than 4 points, 1=royal-or-minor/0=rook-or-pawn, 1=queen-bishop-rook-or-pawn/0=king-knight-or-nothing, so that all 0s means no piece, and 1000, 0100, and 1100 as unused values. The full list would be: 0000 - empty 0001 - black pawn 0010 - black knight 0011 - black bishop 0100 - reserved 0101 - black rook 0110 - black king 0111 - black queen 1000 - reserved 1001 - white pawn 1010 - white knight 1011 - white bishop 1100 - reserved 1101 - white rook 1110 - white king 1111 - white queen