No, I understand what the probabilistic prediction means. The problem is that there is no way to tell if 538's prediction was any more accurate than if they'd said 50-50 (or any numbers whatsoever) because the presidential election is only every 4 years. We can't measure the predictive ability of his model at all. If Hillary won, we'd say "Yup, the model gave her 70%". When Trump won, we said "Yup, the model gave him 30%".
I get where you're coming from, but it's not strictly true. For instance if their model had given Hillary a 99.5% chance and Trump only 0.5%, then we could be pretty sure it wasn't a good model when Trump won.
You might say even then, how do we know the result didn't just fall into that 0.5% remaining chance? But polls can and do still give some useful idea of what the result might be.
