Once a month, the Bureau of Labor Statistics provides updated information about the unemployment rate and the number of Americans with jobs. This 12-times-a-year "Jobs Day" is the most important frequently-updated series of economic data around. But the jobs report that comes out in early February of every year — Friday, in this case — is the most important jobs report out of all 12. That's because it's the report that includes the benchmark revisions.
This is a process through which the BLS goes back in time and adjusts the past estimates of how many people were employed in the United States. These revisions are sometimes small, but on average they're pretty big — on average the total number of jobs is adjusted by 0.3 percent, which given the size of the labor force amounts to 357,000 jobs.
Why does the benchmarking update happen?
The headline number in the jobs report is based on a statistical sample of American workplaces (it's called the establishment survey) and while statistical sampling is a very reliable method, it only works if you know what the underlying population that you're sampling looks like. The problem for the BLS is that new workplaces spring up and old ones die all the time.
The main thing that happens during the benchmark update is that the BLS uses data from the Quarterly Census of Employment and Wages — which is mostly derived from state unemployment insurance filings — to update its model of how many workplaces there are, and what sectors they are in. With this improved model in place, the BLS can re-apply the results of its sampling, and get a more accurate look at the labor market.