Agreed. See a note below re: Bradley effect. In looking at the “lock” states, it’s those where

— All polls used are rated “B” and above by RCP. They’re excellent

— The candidate has to be in the lead more than two points above the margin of error. In most cases, that’s a minimum of 6 points ahead

— The consensus (read: all) major horse racers (NYT, Kos, 538, etc) have the candidate above 80% likely to win. Even one says less than 80%, they move into “in play”

Early on, I actually favored Internet polls (real ones, not open cattle calls) due to exactly that effect. During the primaries, they were a better indicator of Trump support, in my opinion.

But even the Bradley effect, if memory serves, was a six percent swing. Statistically, in analyzing the numbers (and running through a couple of scenarios, though I fall back on Monte Carlo… simplest is best sometimes), that’s allowing for a lot of fudge.

All that said, this is one person’s opinion, nothing more. But it’s opinion backed by a few hundred thousand scenarios run through multiple models.