What they are good for
A WARN notice has one property no statistical series can match: it arrives before the event. Employers must file ahead of qualifying layoffs, so a notice is forward-looking in a way that unemployment data never is. For anyone who needs early warning, that is the whole value.
Our table holds 7,065 filings across six states. In the 2026 window alone there are 1,482 filings covering 95,546 affected workers.
Limit one: they only catch layoffs above a threshold
WARN obligations attach at size thresholds — how many employees a site has, how many are affected, what share of the site that represents. A layoff below the threshold produces no notice. So the absence of a WARN notice is not evidence that nothing happened; it is evidence that nothing happened *that cleared the threshold*.
This matters most in places with many small employers, where real displacement can accumulate entirely underneath the reporting line.
Limit two: states do not file the same way
We published separate figures on this, and the gap is large: in 2026 the median California filing covers 18 workers, while the median New Jersey filing covers 86. Same federal floor, very different filing practice and state-level rules. Any cross-state ranking by notice count is really a ranking of reporting regimes.
Limit three: a notice is a plan, not an outcome
A filing states an intention and a date. Plans change. Layoffs get reduced, delayed, cancelled, or expanded after the notice is filed, and amendments do not always arrive as cleanly as the original. A notice tells you an employer expected to cut a number of roles on a date. It does not tell you what finally happened.
Limit four: the data you want may not be in the data you have
We will be concrete about our own structure here. Our main filing table carries a `days_early` column that is empty for all 7,065 rows — so anyone querying that table directly would conclude we cannot tell how much warning a notice gave. That conclusion would be wrong, and the distinction matters.
The notice-interval computation does not read that column. It is derived from the recorded notice and effective dates at the point the short-notice candidates are built, against the state register record rather than a cached field. The column is simply unused.
We mention it because it is exactly the trap this post is about: **a null column is not the same as a missing capability, and a populated column is not the same as a verified one.** If you need a specific answer from any provider, ask how the number is derived and from which source — not whether they "have the data".
How to use them properly
Use WARN notices for early warning and targeting, not for counting. Pair them with a complete statistical series for level, because WARN is a threshold sample and will never be a census. Read the distribution, not just the total. And check which states a provider actually covers, over which period — coverage that varies by state makes totals uncomparable even within one dataset.