If you want to keep posting numbers that show that some metric is getting better and better, you are mathematically forced to go back and reset the baselines every so often, because otherwise you're going to claim that the United States has a billion and a half jobs before long.

Plus, revisions are expected and normal. These stats are hard, and over time more data comes in about the past. What gets suspicious is how consistently the revisions are always in the direction of worse, for whatever "worse" is for the given metric. "Random" errors ought to be distributed more-or-less around zero. It indicates, if not proves, that there are more politics in these numbers than we'd like for supposedly objective metrics.

But... if you think about it hard enough, and perhaps have a dash of cynicism, it's what you'd expect. It just requires something that pretty much everyone reading this has personal experience of doing with their own managers. You know which direction to shade the results, to indicate how far the bug is towards being fixed or whatever. Nobody has to tell you. You figure it out pretty quickly.

Well, it's not just you and your manager.