A typical example of this interface problem just occurred in my game: shown on move #30 here:
lichess.org/study/wmor48QNOn move 30: sideways capture of a rook failed to work here loosing the game by poor interface construction. Over a real chess set a loss would never occur like this by anyone, it is only caused by the inappropriate difficulty of landing the web chesspiece icon at a sideways angle, and thus having to avoid touching the tiny corners of all the neighboring incorrect squares. The horizontal-vertical oriented chess squares are not drawn optimally for that motion. Losses from this occur approx. every 20 games in speed chess, in my games, wrecking rating accuracy.
@WindowMaker 's diagram
gyazo.com/34222946ef25e710c5715efecafee045.pgn shows one way to fix that disaster from ever happening, but his diagonal borders could extend to all the squares on the board. Thus when a bishop is picked up by the mouse, why have horizontal-vertical boundaries and illegal destination squares in the way of its path?; instead have the board redrawn with a diagonally oriented matrix of squares. That way only legal squares will ever be landed on (unless there is extremely bad (dozens of pixels off) mouse aim). In a high speed chess game with the current Lichess board, landing a bishop on a non-diagonal square, because of slightly sloppy mouse aim, causes the move to get undone, and so the time loss from that can loose the game for that non-chess skill reason, and it seems fixable.