# Scope and limitations

These 136 constructed evaluation pairs isolate a mathematical sensitivity. No engine was run and no games, players, skill levels, time controls or cheating labels were sampled. The transformation’s Win% is not a measured chance that a particular person will win, nor Stockfish’s separate win/draw/loss output. Values use ChessAnalysis’s pinned coefficients, +1 adjustment, 0–100 result clipping and ±1000-centipawn input clipping. All input scores are White-centric and converted once to the mover’s view. Mate scores and whole-game aggregation are excluded. A loss flattened by clipping does not mean a move is objectively harmless. The tables cannot reproduce Chess.com’s proprietary scores or guarantee current Lichess results.