They're just the results of surveys. They get revised twice after the first report, and there's also a yearly adjustment based on unemployment benefits numbers. Revisions of over 100,000 are not that uncommon. (500,000 is pretty rare though!) https://projects.fivethirtyeight.com/jobs-report-growth-unem...

The top level of US federal agencies has been a skeleton crew since 2017. Much of the reported data was supposed to have been provided by agencies which have been hamstrung due to a lack of leadership in appointed positions. Whether intentionally or not, many of those appointed positions are those that carry the responsibility of ensuring that the congressional mandates of the agency are carried out. Without appointed officials in those positions, that duty falls to "acting" personnel who essentially end up working two full-time jobs. In some agencies this problem is magnified because acting personnel are not authorized to hire administrative support staff. Imagine a company like Facebook eliminating their entire upper management and allowing randomly selected employees to fill in for those positions, whether they are qualified or not.

You'd be surprised how infrequently someone does a basic sanity check on the voluminous economic numbers published by the government, all of which are derived from models (probably in a spreadsheet somewhere). Every so often a fresh pair of eyes in the government will notice an obvious flaw in the output of the model, which typically causes a one-time correction to be issued but they don't necessarily retroactively fix previous official numbers. Most of these errors happen with numbers that are sufficiently obscure and specialized that it isn't newsworthy.

The underlying methodology for the BLS model is prone to serious bias and brittle. You can see this in the level of disagreement between the BLS numbers and the widely cited ADP numbers. The ADP numbers also have significant fundamental bias, but it is an entirely different kind of bias so both models capture effects that are largely invisible to the other. For Americans in most parts of the US, I think the ADP numbers are a bit closer to ground truth but take either with a large grain of salt.

The agency was recently moved and lost a huge part of their staff, they're not going to be as competent as they used to be. The same things is happening to the USDA.

Initial jobs reports are rough estimates that come out fairly quickly. Over time they get revised either up or down as more data becomes available.

Well, they always revise the numbers. I assume it's because they send out surveys, and more answers are always trickling back in. Maybe the late responses were far different than the first ones this time for some reason?