According to the attached reference, does it mean that the adversary computes $PPT(initial~input)$, $PPT(z)$ and $PPT(f_i(x',y'))$?
For example, suppose the corrupted party, in the malicious model, has inputs $x,z$ and it wants to compute $x \land y$, then the adversary's output will be:
$PPT(x)$, $PPT(z)$ and $PPT(x \land y))$. Am I right?
Note: PPT is a probabilistic polynomial-time algorithm