I think in general engine evaluation algorithms can't be trusted in endgames. That's why we have Nalimov tablebases. Coding an evaluation algorithm that works for endgames, if possible, would likely weaken it's middlegame performance.
Perhaps the solution is for the chess.com analyzer / computer workout to use tablebases.


The chess.com engine isnt very accurate in analysis. for example, it says that white has a decisive advantage when its clearly a draw and it even points out that i should play kf7 rather than kh8 so white will have a decisive advantage. besides this example, it lists openings such as the sicillian and caro kann as a mistake or inaccuracy and lists the french as the most equalizing move. i think this should be fixed