I had Claude go through each occupation, compare it to O*Net as they did, and recode it, then had a separate agent go through and argue about which ones they had wrongly coded as not doable remotely.

So, you generated some random data that has absolutely no basis in fact, and proceeded to base a large portion of your argument on it. Wonderful.