There should really be a notational distinction between 'correct' play(a la TOP), and the most likely estimate(mle? the 'right' play?) of correct play when there's imperfect information. By taking mle play as much as possible, good play comes close to 'correct' play more often than any other strategy. The donk that always calls will ocasionally beat the mle strategy (like this example) but it doesn't happen often enough to make his overall strategy not terrible.
I like this example of the distinction between correct play and the right play:
http://www.pokerstove.com/analysis/quad-jacks.php