It's a small sample, like 50-100k IIRC, through interviews and I'm not sure the method. It's been years since I read up on it beyond a cursory interest and old government/political classes, but again IIRC they extrapolate that out for the analysis. Since the sample is so small, restating the seasonal adjustments allows for huge swings month to month, which has a big impact when you are say... leading up into an election, and the restated numbers will be after the election. I'll be the first to tell you that I frankly think that's a bit of a stretch and conspiracy theory sounding.

It doesn't affect my opinion that the number published monthly is bullshit though.