TalkChess.com

[Ralph Stoesser]

A="small delta => accurate evaluation"

For the perfect evaluation delta is 0. Let's look at the worst possible evaluation. Let's say the worst possible evaluation would always report a loss when the position is a win and vice versa. For a draw position the worst possible evaluation would always report a win. What about delta now? Delta would be obviously 0 again, because the worst possible evaluation is as reliable as the best possible evaluation. Now lets make the best possible evaluation a little weaker and the worst one a little stronger. We could say there is one position p for which the best evaluation report randomly the wrong and the worst evaluation the right outcome. In both cases delta would be the same small value. Therefore I think A is not true.