Datasets:
pretty_name: US WARN Act Layoff Notices — rebuilt daily from state portals
license: cc-by-4.0
language:
- en
task_categories:
- tabular-classification
- time-series-forecasting
tags:
- layoffs
- warn-act
- employment
- economics
- labor
- labor-market
- united-states
- time-series
- public-records
- government-data
- alternative-data
- daily-updated
- warn-notices
- plant-closings
- job-cuts
- workforce
- hr
- recruiting
- tabular
size_categories:
- 10K<n<100K
configs:
- config_name: default
data_files:
- split: train
path: data/warn_notices.csv
US WARN Act Layoff Notices — normalized, 48 states, rebuilt every day
Last rebuilt: 2026-09-15. An automated pipeline re-scrapes 48 state labor-department portals every day, re-normalizes, re-deduplicates and re-uploads this file. Compare that date with the "last modified" date on any other US WARN dataset on the Hub before you choose one — WARN data is a moving target and a one-shot upload starts rotting the week it is posted (states amend headcounts, re-issue notices, and silently drop rows).
Every US layoff notice filed under the WARN Act that we can reach, scraped
daily from 48 state portals, normalized to one schema, deduplicated,
with canonicalized employer names. 61,320 notices from 48
states in this free tier (notices dated 1988-01-01 onward,
published with no delay). States in this CSV: AK, AL, AZ, CA, CO, CT, DC, DE, FL, GA, HI, IA, ID, IL, IN, KS, KY, LA, MA, MD, ME, MI, MN, MO, MS, MT, NC, ND, NE, NJ, NM, NV, NY, OH, OK, OR, PA, RI, SC, SD, TN, TX, UT, VA, VT, WA, WI, WV. Not
covered: AR, NH, WY — those states publish no public WARN list
(surveyed 2026-09-13; the per-state survey with archive depth is its own
dataset: which-us-states-publish-warn-act-layoff-notices-coverage),
so 48 is every US jurisdiction that publishes one. coverage.json
(included) lists what the pipeline currently scrapes, and flags the
6,799 notices that the issuing agencies' live pages no longer
list (recovered from archived copies of retired agency pages; published on
their own as us-historical-layoffs-archive-warn-act-notices-removed-from-state-websites).
The data above is free forever — CC BY 4.0, no login, no API key, no delay, rebuilt daily.
If you only care about particular employers or states, you do not have to re-download the whole file and diff it yourself. WARN Watch runs your list against every daily refresh and gives you a private alert page, a calendar (.ics), an RSS feed and an optional Slack/Discord/Teams webhook — no login.
Try it free for 30 days — no card (up to 3 employers or one state) · then $49/year, one-off, no auto-renew, 14-day refund · see a real alert page first
Quickstart
from datasets import load_dataset
ds = load_dataset("APProjects/us-warn-act-layoffs-notices-daily", split="train")
print(ds[0])
import pandas as pd
df = pd.read_csv("https://huggingface.co/datasets/APProjects/"
"us-warn-act-layoffs-notices-daily/resolve/main/data/warn_notices.csv")
# layoffs by month
df.groupby(df.notice_date.str[:7]).employees_affected.sum().tail(12)
How much notice did workers actually get?
The WARN Act is a law about a number of days, and no state publishes that number. Both dates are in this dataset, so we compute it on every refresh:
15,685 of 33,691 notices (47%) that publish both a notice date and a later layoff date gave workers fewer than 60 days — 1,550,726 workers in total. The median notice period is exactly 60 days: employers file at the line, and half land under it. 2,016 notices were filed the same day the layoff took effect.
This is arithmetic on public records, not a finding that anyone broke the law: WARN excuses short notice for a faltering company, for business circumstances that were not reasonably foreseeable, and for natural disasters, and several states run their own notice laws. We also exclude rather than reinterpret the 12,200 records whose layoff date precedes their own notice date, and the 12,169 with no layoff date at all.
- Free CSV, 33,691 rows with a computed
notice_dayscolumn: notice_period.csv - Charts, per-state ranking and the full method: notice-period.html
notice_daysis computed for all 61,320 rows back to 1988, free
Rebuilt every day
17 consecutive days of data commits to the public repo, 2026-08-30 through 2026-09-15, none missed. Layoff trackers are usually one person's side project and stop when that person gets busy; this one is a pipeline, and the commit history is the proof. Counted from git log -- data on every build.
This year against last, same months
2,359 WARN notices Jan-Jul 2026 vs 2,415 Jan-Jul 2025 (-2.3%); 210,630 workers this year vs 221,870. Counted by notice_date from data/warn_notices.csv on every build. The two most recent months are left out because state portals post notices dated in the prior month for weeks afterwards; every month since 1988 is in the monthly time-series companion dataset.
How far back does this go?
This file is the whole archive: 61,320 notices back to 1988, with no paywalled years. The free page US layoff notices by year, 1988–2026 charts every one of those years — notices, workers affected, and how many states were filing (the coverage caveat, stated rather than hidden). The starkest thing it shows: 2020 alone holds 16,295 notices covering 2,045,768 workers, more workers than 2021–2025 combined (1,428,928) — and that year, like every other, is inside this file.
For scale on what a paid comparable costs: WARNTracker's public pricing page (checked 2026-09-05) lists $2,000 one-time for its historical CSV and $250/month for a weekly email of new notices, covering 46 states and D.C. This file is free and rebuilt daily across 48 states, and the whole 1988-present archive is in it at no charge. What we sell is the daily watch on YOUR employers: WARN Watch, $49/year.
If this saved you a scrape
Two free things, five seconds, no account changes hands:
- Like this dataset (the heart, top right) — likes are how the next person searching the Hub for "WARN" or "layoffs" finds a file that was rebuilt today instead of one uploaded once in 2023.
- Watch the source repo's releases — a notification each day the snapshot is republished, with the row counts and the per-state freshness table in the notes.
Something wrong in a row, or a state you need that is not here? Open an issue — the last reader's fix shipped the same day.
What a fork of the scraper cannot get
Collection uses the Apache-2.0 biglocalnews/warn-scraper, and you should fork it — it is good software. A fork returns whatever each state portal serves today. Five states have taken their history down:
| State | On the state portal today | In this dataset | Recovered from deleted pages | Period recovered |
|---|---|---|---|---|
| Michigan | 116 | 2,202 | 2,086 | 2000–2025 |
| Pennsylvania | 310 | 1,895 | 1,585 | 2011–2023 |
| New York | 5,010 | 6,507 | 1,497 | 2001–2014 |
| Connecticut | 27 | 871 | 844 | 2010–2025 |
| North Carolina | 295 | 1,082 | 787 | 2014–2021 |
6,799 notices in this file exist only because they were rebuilt one Internet Archive capture at a time from pages the agencies removed (Michigan's retired milmi.org listings, Pennsylvania L&I's month-by-month pages, Connecticut's year index, New York's retired labor.ny.gov/app/warn application, North Carolina's archived yearly PDFs). Those five portals list 5,758 notices between them right now. Nothing here is estimated or interpolated; fields a source never printed are left empty. The archive-only rows are also published on their own: us-historical-layoffs-archive-warn-act-notices-removed-from-state-websites.
Is there a state feed you could subscribe to instead?
We checked the official source behind all 48 states on 2026-09-11: 0 publish a feed you can subscribe to, 3 post a spreadsheet, and the other 45 publish a web page, a pile of PDFs, a POST-only search form — or no WARN page at all. Two that advertise a feed (Georgia, Idaho) serve the agency's WordPress comments feed with no notices in it. That is why this dataset is rebuilt daily and why each state gets a normalized RSS feed here: the per-state census with every portal linked (re-run weekly; evidence in data/source_formats.json in the GitHub repo).
What a single state's feed cannot tell you
Layoffs cross state lines; WARN portals do not. Because this dataset normalizes 48 portals into one schema and canonicalizes employer names across them, you can ask a question no single state's feed can answer:
2,756 employers filed WARN notices in two or more states since 1988 — 605 of them in five or more, 2,915,033 workers in total, spanning 48 states (build of 2026-09-15).
| Employer | States | Which | Workers |
|---|---|---|---|
| Aramark | 37 | AL, AZ, CA, CO, CT, DC, DE, FL, IA, ID, IL, IN, KS, KY, MA, MD, MI, MN, MO, MS, NC, NJ, NV, NY, OH, OK, OR, PA, RI, SC, SD, TN, TX, UT, VA, WA, WI | 49,229 |
| Sodexo | 37 | AL, AZ, CA, CO, CT, DC, DE, FL, HI, ID, IL, IN, KS, KY, LA, MA, MD, MI, MN, MO, NC, NE, NJ, NV, NY, OH, OK, OR, PA, RI, SC, TN, TX, VA, WA, WI, WV | 28,690 |
| Walmart | 30 | AK, AL, AZ, CA, CO, CT, DC, FL, GA, IL, IN, KS, MA, MD, MI, NC, NJ, NM, NY, OH, OK, OR, PA, SC, TN, TX, UT, VA, WA, WI | 23,084 |
| Conduent | 30 | AZ, CA, CO, CT, DE, FL, GA, ID, IN, KY, LA, MD, MI, MO, NC, NE, NM, NV, NY, OH, OK, OR, PA, RI, SC, SD, TN, VA, WA, WI | 3,605 |
| ABM | 27 | AL, AZ, CA, CO, DC, FL, GA, IL, IN, MA, MD, MN, MO, NC, NJ, NV, NY, OH, OK, PA, SC, TN, TX, UT, VA, WA, WI | 12,448 |
Full ranked list, filterable, free and rebuilt daily: companies laying off in multiple states.
Don't want to re-download this every day? (free, no card)
If you only care about a few employers, you do not need the whole CSV.
Check them right now, in your browser →
— paste any list, see every match from the last 180 days; no account, no email, nothing
sent anywhere (?q=Boeing|Kaiser or ?s=CA pre-fills it).
To be told when one of them files next, name up to three employers (or one whole state) and every daily rebuild is matched against them for you — matches land on your own alert page with a calendar (.ics) and an RSS feed, and optionally a Slack/Discord/Teams webhook.
Start a free 30-day watch →
— $0, no card, nothing renews. Matching runs on the employer name as filed, on
our canonical name and on the trade name, so general motors also catches
General Motors, LLC -- Lansing Region.
Want the one column no state publishes? The companion dataset
how many days of notice each filing actually gave
computes effective_date - notice_date per notice — half of all measurable US
WARN filings since 1988 gave workers under 60 days, and the median sits exactly
on 60. Same daily rebuild, same ids, so the two files join on id.
Which employers are cutting in more than one state? No single state portal
can show that; the companion
multi-state employer dataset
merges the free window across every state, one row per employer (states,
notices, workers as filed) plus every underlying notice keyed by the same id.
Need the aggregate rather than individual notices? BLS killed its Mass Layoff Statistics program on 2013-09-30 and nothing replaced it. The companion US Mass Layoff & Plant Closing Statistics rebuilds a monthly national and per-state series from this same file — events and workers affected, WARN's own 50-worker threshold, back to 1988, refreshed on the same daily run.
Want to know whose job ends NEXT rather than what was announced? The companion upcoming layoffs dataset is only the notices whose separation date falls in the next 90 days: 414 notices across 27 states, of which 230 take effect within 30 days (21,500 workers named). Those people are still employed as of this rebuild. A copy of that file is wrong within a day, which is the point of it.
What did this filing say before it said this? No state portal keeps a history - the page
simply changes. The companion
observed change log
is a diff of this dataset against itself, every day since 2026-08-30: 593 observed
changes over 16 daily builds, touching 404 distinct notices, joinable on id.
It labels only what is mechanically decidable, because a difference between two builds can
come from the agency amending the notice or from our own parser improving, and a row that
stops appearing is not proof the employer withdrew anything. Nobody can reconstruct what those
portals said last Tuesday, including us - the file only grows forward.
Only want this year? The companion US layoffs 2026 dataset is the 2,690 notices filed in 2026 (243,008 workers named, 45 states), with by-state, by-month and top-employer rollups already computed. Same rebuild, same ids.
Only want the tech layoffs? The companion [US tech company layoffs dataset] (https://huggingface.co/datasets/APProjects/us-tech-company-layoffs-warn-act-notices-daily) is the 1,732 filings from software, internet, semiconductor and electronics employers (191,402 workers, 367 companies, 1989-2026), with the per-company history already rolled up. Unlike the trackers built from news stories, every row there is a filing with a state, a site and an effective date. No state publishes an industry field, so sector is assigned from the employer NAME and the file is a floor, not a census: 47.4% of all notices carry a name that reveals no sector. Same daily rebuild, same ids.
Want it one row per COMPANY instead of one row per notice? The companion employer layoff-history dataset rolls all 61,320 notices up into 29,900 canonicalized employers with their states, first and latest filing and total workers noticed — and shows what the notice-level file hides: the 6,410 employers that have filed more than once account for 62% of every notice on record. Same daily rebuild, same canonical names.
Want it as a TIME SERIES? The companion
monthly layoffs time-series dataset
is one row per month, 1988-11 to 2026-09 (455 months, zeros
included), notice-date and effective-date bases side by side, with a states_reporting
column so nobody mistakes growing coverage for growing layoffs, plus a per-state table.
Busiest month on record: April 2020, 4,769 notices.
Prefer clicking to coding? The live explorer Space searches every employer in this file, lists the newest notices, and shows per-state freshness — it reads the same files, so it is never staler than the dataset.
This is a live mirror of github.com/APVentureEngine/warn-act-notices — stats site, per-state CSVs, weekly/monthly pages, RSS, and API docs live there.
Schema
Why this repo exists in one paragraph: of the 48 state sources in this refresh, 47 arrive in 37 different column layouts using 193 distinct column names, and the employer's name alone is published under 8 different headings — Doing Business As Name (IL) · Name of Company (MT) · Organization Name (DC). Every one is mapped to the 11 fields below on every refresh, dates parsed to ISO and employers resolved to a canonical name, so one company can be followed across state lines. Counted from the raw files on each build, never hand-written.
| column | meaning |
|---|---|
id |
stable row id (source-issued key where the state provides one) |
state |
two-letter state code |
company |
employer name as filed |
company_canonical |
canonicalized employer name (dedup/rollup-friendly) |
company_dba |
"doing business as" alias, when filed |
location |
site location as filed |
employees_affected |
headcount on the notice (blank when the state omits it) |
notice_date |
date the notice was filed |
effective_date |
planned layoff/closure date |
notice_type |
layoff / closure / amendment etc., as filed |
first_seen |
date our pipeline first observed this row |
Freshness & coverage
Regenerated from state sources every day; broken scrapers are repaired
within a cycle, and per-state coverage is tracked in coverage.json
(included here). A one-off copy of this dataset rots in weeks — several
states silently amend, re-issue, or remove rows; the pipeline reconciles
that daily with stable-id retention logic.
Hugging Face lists several other US WARN datasets. Before choosing one, compare their last-modified dates: the site's data-sources comparison reads them from the Hugging Face API on every rebuild and prints them next to this dataset's rebuild date, so the freshness claim above is checkable rather than asserted.
Browse by state
Every state below has a page with its notices, a monthly trend, RSS and JSON,
rebuilt from this dataset each day. Latest notice is the most recent notice
date that state has published — not the date we checked — so a stale column is
the STATE being quiet, not this pipeline being broken.
The whole archive is in this file
Since 2026-09-07 there is no paywalled history: all 61,320 notices back to
1988 are in data/warn_notices.csv here and in the GitHub repo, CC BY 4.0.
The one paid tier
- WARN Watch — up to 500 employer terms plus whole states, matched on every daily refresh for a year; private alert page + calendar (.ics) + RSS + optional Slack/Discord/Teams webhook, no login. Every alert carries that employer's whole filing record ("Boeing — 416th WARN notice from this employer since 1998 · 14 states · 43,948 workers to date · last filed 2025-04-04 · 1 of 5 locations this employer filed in CA that day") — history a keyword rule on an RSS feed cannot see. Buy — $49/year · see a real alert page · what it looks like · try it free for 30 days, no card
License & citation
Free tier: CC BY 4.0. Cite as "WARN Feed, github.com/APVentureEngine/warn-act-notices" (see CITATION.cff in the GitHub repo). Underlying scrapers include Apache-2.0 code from biglocalnews/warn-scraper (attributed in-repo).
A layoff record you can audit, not just download
This dataset is one cut of a single daily rebuild: 61,320 US WARN Act layoff notices from 48 state agencies, 1988 to today, one schema, no login, no delay, CC BY 4.0. Snapshot as of 2026-09-15; the files above are rebuilt every day, so the live count is the truth.
Several projects publish a current WARN scrape and two of them carry more rows than we do. None of them publish what the records used to say:
- 593 observed changes to already-published notices, logged daily since 2026-08-31.
data/revisions.csvrecords every field that differed between two consecutive daily builds — employee counts, effective dates, notice types, company names — with the old value, the new value and the date we saw it. We publish the observation and not the cause: a change is equally explained by the agency amending the notice or by our own parser improving, and we do not guess which (seedata/revisions.README.txt). A scrape that starts tomorrow cannot backfill any of it; it only exists if someone was watching. - 6,799 notices whose state agency page no longer lists them. Agencies take notices down. We keep them, flagged as archive-only, so a count you ran last year still reconciles.
- Point-in-time employer identity. The ticker crosswalk resolves a filer to the company as it existed at the time of the notice — Kmart, Sears Holdings, Symantec — not to whatever is on today's ticker file.
If you have to defend a number to an editor, a referee or a compliance reviewer, that provenance layer is the part you cannot rebuild yourself. How to cite this dataset →
Look something up right now — free, no signup, nothing to install. Check any employer or state against the last 180 days → It runs in your browser against these same files.
Building something with it? The same files are a free HTTP API — JSON and
CSV, no key, no signup, access-control-allow-origin: * so fetch() works from
a browser: endpoints, schema and curl examples →
Or have it watch a list for you. Coming back to look is the part a CSV cannot do. WARN Watch — $49 for a year, one payment, nothing auto-renews, 14-day refund, no login: up to 500 employer names plus whole states, matched on every daily refresh, delivered to a private alert page + calendar (.ics) + RSS + an optional Slack / Discord / Teams webhook. Every alert carries that employer's whole filing history from the archive, which a keyword rule on an RSS feed cannot see. There is no built-in email — we do not claim one.
- See a real alert page before paying · what you get
- Try it free for 30 days, no card · Buy — $49/year
Not deciding today? Join the update list → — used only for a state added, a column change or a tier change; nothing promotional. The list is shared across APProjects datasets, holds an email address only, is run by Gumroad, and any message unsubscribes you. Rather give no address at all? Watch the repo's releases — GitHub notifies you on every daily republish, and a new state or changed field is in those notes the day it lands.
Reaching a human. WARN Feed is published by APProjects, an automated data publisher — that is stated plainly rather than dressed up. Corrections, coverage gaps, schema questions and refund requests all go here and are read: open an issue. Payments are handled by Gumroad as merchant of record, so an invoice can carry your company name.
Source, scrapers and methodology · the 48-state site
Card views are counted anonymously: one 1×1 image on a public CDN, no cookies, no script, no personal data. The count is public: jsDelivr stats.