What is Discovery with Models?

What are the core benefits of Discovery with Models, both according to Hershkovitz, and in your opinion?

Do you buy Hershkovitz et al.'s operationalization of carelessness? Are there other reasons why a student could make an error when he/she "knows the skill"? Could careless behavior occur on a correct answer?

What are the dangers to conducting discovery with models if your model is imperfect?

What are the safeguards for drawing conclusions, if your model is imperfect?

Even if a model is itself an imperfect operationalization, is substantive validity -- as shown in Pardos -- sufficient?