What is an "evaluation library" (审局库)?
An evaluation library (审局库) is a kind of endgame tablebase, but without moves.
An endgame tablebase tells you both the result of a position and the winning moves in won positions, while an evaluation library only tells you the result (win, loss or draw). As a consequence, an evaluation library is much smaller than an endgame tablebase.
A common misconception is that an evaluation library is about the middlegame. This comes from being misled by the name ("审局" means "evaluating the game"); in fact it serves the same purpose as an endgame tablebase.
Because an evaluation library contains only results and no move information, some complex winning endgames that follow a fixed pattern are hard to win even when the engine searches together with an evaluation library. For example, in a complex winning endgame, the vast majority of positions the engine searches will be "wins" in the evaluation library, which is not much different from having no library at all, except that the engine does not waste search effort on the "draw" and "loss" positions.