poll internals are important in understanding these beasts. The registered vs probable voters is one key - just like you said many will not vote whereas probable voters more likely giving a closer result.
Also is the amount of ‘special sauce’ - often in the form of oversampling. I’ve noted some polls that oversample by double digits only to result in single digit harris leads.
I don’t hand wring about polls at this stage. Pollsters will sharpen their pencils so that they are not outrageously wrong. They lose business if they are too far off.
You’re right on every point...