I am sorry, but I think you incorrectly understand my question.
"Forced error correction " cannot be implemented in all rounds because what would be the point of randomization then? The idea of the training round before the main ones is to teach participants how to work with this survey element so that in the main rounds their responses will not be skewed by not knowing what they are supposed to do in this survey.
This is a super-critical component of IAT, as well it is discussed in the major studies on IAT as an important preparatory round before the major ones. Because one needs to make sure that he can isolate the effect from “pushing a button not knowing what to do in the survey” from “implicit attitudes”. Otherwise, responses will produce a mess that cannot be interpretable.
In the link I posted earlier in the IAT part of the survey you will have a training session that explains you what to do (I am not talking about the instruction, because written instructions proved to be not enough to explain people what to do in the IAT).