some ideas so far,
N/A before the midterm if no such models are created with someone sharing the sha256 of a zip of the model

The rationale for making the model cutoff early is that what you don't want is for someone to overfit on midterm results and 2028 pre-election polling data

Original phrasing idea that I modified:

Will a model that predicts the swing state results in 2016, 2020, and 2024 based only on final vote counts of all non-swing states successfully predict the swing state results in 2028 based only on the final vote counts of all non-swing states in that year?

seems like that constrains people a bit more than one might want ig? but idrk