I suspect Harris got her "convention bounce" (as defined by the model) right when she became the nominee, this made the model think she was overperforming pre-convention and now the bounce is fading "early" when the model thinks she should still have it so it seems like she's underperformed.
If this is the theory, knowing how close the swing states are and thus how swingy it can be, most likely this number goes back to maybe 55/45 Trump.