How LayoffTracker Collects and Verifies Layoff Data

LayoffTracker.us compiles a public dataset of U.S. layoff events from official state Worker Adjustment and Retraining Notification (WARN) filings. The site's About page sets out what the project is and the principles behind it; this page goes a level deeper — into how the data is actually collected, verified, computed, and bounded. It is written for journalists, researchers, analysts, businesses, and any reader who needs to understand the dataset's provenance and limits before citing or relying on it.

What follows describes that pipeline end to end, and is candid about where the public record — rather than our handling of it — sets the limit on what the data can show. The dataset reports what has been filed; it does not forecast, estimate, or editorialize. Every figure on the site traces back to a government filing, and the sections below explain how.

Data Sources

Every layoff record on LayoffTracker is sourced from a state WARN filing. As of this writing, we do not publish records based on company press releases, news reports, or other secondary sources. This is a deliberate methodological choice, not a limitation we have yet to overcome. WARN filings are the official, legally mandated disclosure of qualifying workforce reductions, submitted by employers to state authorities. Grounding every record in those filings keeps each one traceable to an authoritative, verifiable source rather than to reporting that may be partial, contested, or unconfirmed. If an event is in the dataset, it is backed by a state WARN notice that any reader can locate and check.

This standard may evolve as additional verifiable sources prove reliable. Any such expansion, however, would be a deliberate and disclosed change to our methodology — not an incidental loosening of what counts as a source. For now, the rule is exactly as stated: WARN filings only.

Two channels feed the dataset. The primary channel is the publicly available WARN data that U.S. state departments of labor publish — the live record of new filings as states release them. The second is direct outreach to state agencies, used chiefly to recover historical filings that predate our collection (described under Historical Depth). The complete state-by-state source list, including direct URLs to each state's official WARN database, is published in our FAQ; rather than duplicate it here, we maintain it there as the canonical reference. For the federal definition of the notice requirement itself, see the U.S. Department of Labor's WARN Act page.

Collection Process

Collection runs on a daily cadence across all covered states, but the mechanism depends on what each state publishes and how.

For states with machine-readable WARN systems, collection is automated: our systems scan the official state databases daily and ingest new notices as they appear. A subset of states proactively notify us when new filings are available; those notices are ingested as soon as the notification arrives.

For states without machine-readable WARN data — where notices are PDF documents, arrive in inconsistent layouts, or are published through systems that do not support automation — collection is manual. This work also runs daily.

Source formats vary widely, and our process accommodates each. Some states publish structured, machine-readable data. Some publish PDFs that require parsing. Some publish only basic event details — company, count, and date. Because the dataset reflects what each state makes available, the depth of detail on a given record is partly a function of its source state's publishing practices rather than of our handling.

Ingestion Lag

How quickly an event reaches the dataset depends on the same automated-versus-manual split, layered on top of each state's own publication schedule.

Records from automated states typically appear within a day. Manually collected states lag by a couple of days, depending on volume and data-entry nuances. On top of our ingestion sits the state's publication cadence, which ranges from daily to quarterly — and quarterly-reporting states can lag the actual events by months.

The total lag for any given event therefore compounds two separate delays: the state's publication cadence and our ingestion lag. The practical consequence is that the most recent weeks of data should be read as provisional. They fill in over time as states publish and as our collection catches up, and analyses of very recent periods should be framed accordingly.

Historical Depth

LayoffTracker began systematic collection in June 2023. Collection was initially manual; automated collection has expanded over time, while manual collection continues for states whose WARN systems do not support automation.

The dataset's historical reach extends well before that operational start. When systematic collection began, several states had limited public archives. We contacted state department-of-labor representatives directly, explained our research, and requested historical filings. Several states provided extensive archives — in some cases extending back to 1997. As a result, the dataset's historical depth exceeds our own operational age, though completeness varies by state and by era depending on what each archive was able to provide. We continue to maintain communication with state agencies when historical or clarifying information is needed.

This produces a completeness gradient that matters when working with older records. Records from June 2023 forward — collected in real time — are the most consistently complete. Earlier records reflect what state archives were able to provide, which differs from state to state and from one period to another. Analysis spanning the full time range should account for this unevenness rather than assume uniform coverage.

Verification and Enrichment

Once a record is ingested, it passes through review before it contributes to anything published. Two distinct things happen at this stage, and they rest on different sources of authority.

First, verification. Reviewers confirm that the data was extracted correctly, and the layoff record itself is verified against the source state WARN notice. The WARN filing is the authority for the layoff event — the company, the affected count, the date, and the location all trace back to it.

Second, enrichment. Reviewers clean up company details and add company-level context drawn from outside the WARN notice. For public companies, we cross-reference SEC EDGAR. For private companies, we draw from company websites and public search. Enrichment includes a website, social-media handles, a brief company description (taken from website metadata), and a logo where available.

The distinction matters for anyone citing the data. The layoff event is grounded in the official WARN notice; the surrounding company context is supplementary information assembled from third-party public sources. The two do not carry the same evidentiary weight, and the dataset keeps the filed event — not the enrichment — as the record of record.

Analytics Computation

The site's analytics are computed at end of day, once all incoming records have been ingested, verified, and processed. Because computation waits for the day's processing to finish, the published figures reflect a consistent daily snapshot rather than a continuously shifting number.

Three rules govern how individual records roll up into those analytics:

• Date attribution. We attribute an event to its layoff date when that date is available. When the layoff date is not reported, we substitute the announcement date. A share of events are therefore dated by announcement rather than by the layoff itself.
• Geographic attribution. We attribute an event to the layoff location reported in the WARN notice, not to corporate headquarters. A layoff at a facility in one state by a company headquartered in another is counted where the affected workers are.
• Deduplication. The same event is sometimes filed more than once — most often through amended notices. We deduplicate these at the event level, so a single layoff event is counted once even when it appears in the source data as multiple filings.

Coverage Limitations

No layoff dataset built on WARN filings is complete, and several specific gaps are worth stating plainly. Most reflect the structure of the public WARN system rather than our collection process; we disclose them so the data can be used with appropriate caution.

State disclosure gaps. Arkansas, New Hampshire, and Wyoming do not publish WARN data publicly. Our coverage in those states is limited to records we have obtained through direct outreach to state agencies, and should not be read as comprehensive.

WARN reporting thresholds. Federal WARN applies only to layoffs that meet the size thresholds defined in the WARN Act (29 U.S.C. § 2101) — typically larger employers conducting layoffs of 50 or more workers at a single site, under specified conditions. Layoffs below those thresholds are not required to file WARN notices. Smaller workforce reductions are therefore systematically undercaptured, and aggregate counts should be understood as a floor rather than a full accounting of layoff activity.

Compliance and enforcement. The dataset can only reflect notices that were actually filed. Employers who fail to file required WARN notices escape disclosure entirely. Penalties for noncompliance exist, but enforcement is uneven, so an unknown number of qualifying events never enter the public record.

Industry classification in progress. We are actively enriching the dataset with industry classifications for affected companies, drawn from public sources, and coverage continues to expand. Until that work is complete across the dataset, industry-level analytics should be understood as reflecting only the subset of records for which classification has been completed — not the full set of layoff events.

Historical completeness gradient. As described above, records from June 2023 forward are the most consistently complete, while earlier records reflect the varying contents of state archives. Comparisons that span the operational boundary should account for this difference in collection method.

Correction Policy

We do not independently correct records after publication. When a state amends or corrects a WARN notice in its database, our record follows suit; we mirror the source rather than override it.

This policy follows directly from our role. We publish the public WARN record — not an independent assessment of layoff events. Substituting our own judgment for the state filing would change what the dataset represents. By mirroring the source, the dataset stays anchored to the official record, and any reader who wants to verify a specific record can do so by checking the source state's database directly.

Using This Data

The dataset is most reliable when used with its boundaries in view. Counts are a floor, not a ceiling: sub-threshold layoffs, non-compliant employers, and the three non-publishing states all sit outside it. Recent periods are provisional and fill in as state publication cadences and our ingestion catch up. Industry-level figures reflect the classified subset of records, and pre-June-2023 history varies in completeness by state and era. Every record can be traced to its source state WARN notice, and the full list of those sources is in our FAQ. For the broader editorial position behind these practices, see the About page. Used with those caveats in view, the data offers a consistent, source-anchored record of public WARN filings.