I agree with you to a large extent, Ajinkya. I think it is more constructive to think of 538 as an interpreter of polls rather than a predictor of elections.
538 this spring reminded me of the meteorologist who tells you all week that it is going to rain on Saturday. Then Saturday morning comes, he looks out his window and sees dark clouds on the horizon, and he issues a storm warning. Is he wrong to issue the storm warning? Of course not. But when it starts raining it would be disingenuous for him to pound his chest and bellow, “I was right!”
There is nothing wrong with being an interpreter of polls. That is a valuable service. But given it’s bounciness from day-to-day and poll-to-poll it stretches the definition of “prediction” to call 538’s model a prediction model.