Pollsters must make assumptions about turnout.
If their assumptions are wrong their polls will be wrong.
Easy peasy!
They polled D+8 to D+10, actual was D+4.
"Sexual Orientation?" That's a bit like holding a jellyfish on your fork. Judges will want to see Men and Women, only.
In other words, there is much that is useful in this open-sourced article, particularly presentation of largely yet-to-be certified election returns against political subdivision, states and counties, for example. But the voter category cross-tabs are based on educated, perhaps informed, conjecture.
But these data can't possibly be truly accurate, and won't possibly be for several months. Even then, some of the racial categories (ignoring "Sexual Orientation") are not conventional. And I say that as someone who knows and dislikes the conventions defined by the Census Bureau.
So the Electoral College will report its vote to Congress next month with disputed results and on-going recounts in many places, which is another point in its favor.
"The Roads Must Roll..."