The purpose of the model in question, as I understand it, is to say "players who scored like this turned into NHL players 50% of the time," and I have to assume that the facts bear that out.
The difficulty is figuring out what else is missing that helps you figure out whether a bust is likely or not. You can plug more and more data into the model, but you need data to go into the model and more may not exist. You can scout (which is also data that can go into the model, and I have zero doubt teams do exactly that), but that's not practical for fans, so you publicize your model and what it says. Then you tweak it and improve the data set and recognize its limitations.
One failed data point does not invalidate a model, especially when that failed data point had an expected failure rate. You can say "see, the model was wrong here" and trash the model, but more than anything else that demonstrates your own innumeracy, failure to understand the purpose of practice of modeling, and resort to an appeal to authority (particularly a directed appeal to authority) and confirmation bias.
I suppose the best appeal to authority now is that Grier gave Cagnoni an ELC. He wants to see what Cagnoni can do in the pros, and wants to have control of Cagnoni if he does work out (rather than let him go to some other team to fail or prosper).