I’m not a statistician. But it seems to me that repeated revisions of the jobs number *in the same direction* month after month is prima facie evidence of either a) a bad predictive model in the first place, or b) data being deliberately manipulated.
Stastically, shouldn’t honest revisions go half one way and half the other?
This is based on a telephone survey of I think 60,000 households. They can ask questions differently and change the definition of “work”. Rick Santelli even suggested this may be the case.