WARN Feed · coverage
Which US states publish WARN layoff notices — and which publish none
The federal WARN Act makes large employers notify the state before a mass
layoff or plant closing. It does not make the state publish what it
receives. So the honest coverage question is not “how many of the 50
states do you have?” — it is how many publish at all, and how far
back each one goes.
We hold {f['total']:,} notices from all {f['n_pub']} US
jurisdictions that publish a public WARN list, rebuilt every morning.
{f['n_none']} states publish none, and they are named below with what we found
when we looked. The whole file is free under CC BY 4.0.
{kpi_fig}
Download the latest notices (CSV)
Every jurisdiction we publish
“Archive starts” is the earliest notice date we
hold for that state. It reflects how far back that agency's portal goes and
how deeply we read it — it is not a claim about when the state started requiring
notices. Rows with no usable date are counted separately and never dropped:
{f['dated']:,} of {f['total']:,} notices carry a usable notice date,
{f['undated']:,} do not. Generated {html.escape(f['generated_at'][:10])}.
{table}
{bars}
The {f['n_none']} states you cannot get, and why
We searched for a public notice list from each of these agencies on
{SEARCHED_ON} and found none. We are reporting what we could and could not find
— not a legal conclusion. If you know of a public source for any of them,
open an issue and it will be in the file the next
morning.
| State | Agency | What we found |
|---|
{none_rows}
{TERRITORIES} are covered by federal WARN but run no public
notice portal we could find, so they appear in no dataset, ours or anyone
else's.
Where our archive is shallow — read this before you buy anything
Several large states publish only a recent window. Our California rows start
in {next((d['first'] for d in sts if d['st'] == 'CA'), '—')} and our Texas rows
in {next((d['first'] for d in sts if d['st'] == 'TX'), '—')} because that is how
far back those portals reach, not because nothing happened before. Any dataset
built from these portals has the same floor, whether or not it tells you. Ours
tells you:
| State | Archive starts |
Notices |
|---|
{sh_rows}
Where a state's portal drops its own history, we keep what we already
scraped: {f['cov'].get('archive_only_rows', 0):,} notices in our file no longer
appear on the agency site at all. That is the part of this job that only gets
done if somebody is reading every portal every day.
How to check any of this
Every number above is computed from the published files at build time, and
you can recompute it yourself:
coverage.json holds the per-state row counts,
scrape times and freshness; this table is one CSV row per jurisdiction at
warn_coverage_by_state.csv (also
on Hugging Face); the free HTTP API serves the
whole archive as JSON, NDJSON and per-state CSV with no key and no login. If a
number here disagrees with the file, the file is right and it is a bug —
tell us.
{cta}
WARN Feed ·
API · privacy & terms ·
contact · data CC BY 4.0
"""
return page, f
# ----------------------------------------------------------------- markdown twin
def markdown():
f = facts()
sts = f["states"]
base = _served_base()
lines = [
"# WARN notice coverage by state",
"",
"Which US jurisdictions publish a public WARN Act layoff-notice list, "
"which publish none, and how far back each archive goes.",
"",
f"**{f['total']:,} notices · {f['n_pub']} publishing jurisdictions · "
f"earliest {f['first_year']} · rebuilt daily · CC BY 4.0**",
"",
f"Full page with charts: <{base}/coverage.html> · "
f"free API: <{base}/api.html>",
"",
f"This table as one CSV row per jurisdiction (51 rows, incl. the three "
f"non-publishers): [warn_coverage_by_state.csv]({base}/data/warn_coverage_by_state.csv) · "
f"[Hugging Face dataset]({HF_COVERAGE_DS})",
"",
"## States that publish no WARN list",
"",
f"Searched {SEARCHED_ON}; no public notice list found from these "
"agencies. This reports what we found, not a legal conclusion. Know a "
f"public source? [Open an issue]({ISSUES}).",
"",
"| State | Agency | What we found |",
"| --- | --- | --- |",
]
for st, agency, url, note in NO_PUBLIC_LIST:
lines.append(f"| {sname(st)} | [{agency}]({url}) | "
+ re.sub(r"\s+", " ", note) + " |")
lines += [
"",
f"{TERRITORIES} are covered by federal WARN but run no public notice "
"portal we could find.",
"",
"## Every jurisdiction we publish",
"",
"`Archive starts` is the earliest notice date **we hold** — a fact "
"about the portal's depth and our reading of it, not about when the "
f"state began requiring notices. {f['dated']:,} of {f['total']:,} "
f"notices carry a usable notice date; {f['undated']:,} do not.",
"",
"| Jurisdiction | Notices | Archive starts | Latest notice | Source |",
"| --- | ---: | ---: | ---: | --- |",
]
for d in sts:
src = f"[{d['agency']}]({d['url']})" if d["url"] and d["agency"] else (d["agency"] or "—")
lines.append(f"| {d['name']} ({d['st']}) | {d['rows']:,} | "
f"{d['first'] or '—'} | {d['latest'] or '—'} | {src} |")
lines += [
"",
"## Where the archive is shallow",
"",
"Large states that publish only a recent window. Any dataset built from "
"these portals has the same floor, whether or not it says so.",
"",
"| State | Archive starts | Notices |",
"| --- | ---: | ---: |",
]
for d in shallow(sts):
lines.append(f"| {d['name']} | {d['first']} | {d['rows']:,} |")
lines += [
"",
f"Where a portal drops its own history we keep what we already scraped: "
f"{f['cov'].get('archive_only_rows', 0):,} notices in our file no longer "
"appear on the agency site at all.",
"",
f"Generated {f['generated_at'][:10]} from the published files. "
f"Recompute from [coverage.json]({base}/data/coverage.json).",
"",
]
return "\n".join(lines)
def _served_base():
import repo_host_fix
return repo_host_fix.NEW.rstrip("/")
def inject_readme(path):
line = (f"- [Coverage by state]({_served_base()}/coverage.html) — which "
f"states publish a WARN list, which publish none, and how far back "
f"each archive goes. {README_ANCHOR}")
try:
txt = open(path, encoding="utf-8").read()
except FileNotFoundError:
return False
out = [ln for ln in txt.split("\n") if README_ANCHOR not in ln]
for i, ln in enumerate(out):
if "" in ln:
out.insert(i + 1, line)
break
else:
out.append(line)
new = "\n".join(out)
if new != txt:
open(path, "w", encoding="utf-8").write(new)
return True
def link_from_api_page():
"""api.html is the developer front door and has no byte budget problem
(index.html does — see gen_api_page.write). Idempotent."""
p = os.path.join(OUT_ROOT, "api.html")
if not os.path.exists(p):
return False
txt = open(p, encoding="utf-8").read()
if 'href="coverage.html"' in txt:
return False
needle = '