The nature of the argument, as I understood it, is that the actual unemployment rate being much higher than the predicted rate (even higher than the predicted rate without a recovery plan), is evidence that the stimulus did not work. However, the strength of this evidence is related to the strength of the model, and because the model’s predictive power is so poor, the strength of the evidence is also poor. What this also could suggest is that even if the actual employment rate tracked the prediction, it might not be much evidence that the prediction worked anyway. I didn’t look into the original model, but knowing a bit about statistics and that there are numerous factors that affect unemployment rate, the argument sounds plausible.
This is really frustrating as a proponent and sometimes-practitioner of mathematical modeling, who is starting to dabble in economics. If these macro real-world problems are so difficult to model that the best modelers are not able to say anything useful about their work, what hope is there for the rest of us? Why don’t we all just pack it in and stock up on gold, guns, and supplies? Depressing.