Supposedly, there is no way to know, ahead of time, that the sample taken reflects the expected turnout. So they "assume" a turnout model, and weight the sample according to what they claim is the expected turnout.
Weighting also introduces another assumption, that the DEM vote will be 100% Crooked, and the R vote will be 100% Trump.
No weighting doesn’t assume all d go hill and all r go trump. They weight it based on the results they have. So if results are d are 90% Hillary and 10% trump they reweight using those numbers