From what I recall the main narrative was "it's neck and neck but we could be a single polling misstep from a blowout in either direction" which is about as safe a stance as you can take (understandable considering the blowback they got previously).
What you have to do is simulate all the states and DC and run a monte carlo simulation and you will always get an equivocal answer if you do that.
Which is why any model worth its salt only uses national polls to predict state results. (State polls are more meaningful. But the headlines they generate aren't as nationally clickable.)
There aren’t that may states. You don’t need partisan polling in Vermont and Wyoming, for example.
> you have to infer state-level results based on something other than state-level polls
Yes. As I said. That is what you use national polls. “But the electoral college” isn’t a valid argument against election models. The reality is paucity of granular data and complexity of predicting swing states.