Methodology — how every number is computed

Every figure on EmployerFilings is computed from public government disclosure files by deterministic code — no estimates, no adjustments, no manual edits. This page states each formula, its data source, the minimum-volume floors that suppress noisy values, and the known limitations. When a benchmark is missing on an employer page, it is because the data does not clear these floors, never because it was removed.

Data sources

All sources are public and free: the USCIS H-1B Employer Data Hub (approvals and denials per employer per fiscal year), and the Department of Labor OFLC disclosure files for LCA (wages, worksites, prevailing-wage levels) and PERM (green-card labor certifications). Each employer page's sources section states the exact fiscal periods used, and every page carries a "data through" stamp. For a field-by-field walkthrough of the source files, see how to read an LCA and reading the USCIS Employer Data Hub.

G1 — Approval rate

Initial approval rate = initial approvals ÷ (initial approvals + initial denials) for one employer and one fiscal year, from the USCIS Data Hub. Continuing-petition approval rate is computed separately with the continuing columns. A rate is shown as not published — never as zero — when the denominator is zero. Rounding rule (applies to every percentage on the site): values display to the nearest whole percent, except that a rate between 99.5% and 100% or between 0% and 0.5% keeps one decimal — a record with any denials is never shown as 100%, and a nonzero rate is never shown as 0%. Trend tables show the five most recent fiscal years; totals labeled "on record" always cover every year in the files.

G2 — PERM follow-through rate

Distinct PERM labor certification filings over the trailing three fiscal years ÷ initial H-1B approvals over the same window. It approximates how often an employer that wins H-1B petitions goes on to start the green-card process. Floor: at least 10 initial approvals in the window, otherwise not published. The ratio can exceed 100%: PERM filings also cover workers hired in earlier years or on other visa types, so it is a filing-intensity ratio, not a probability. A PERM filing is not a green card; certification is an early step — the PERM stages guide walks the full path.

G3 — Advanced-degree mix (proxy)

Share of the employer's PERM filings whose minimum education requirement is a master's degree or higher. This is a proxy for EB-2 versus EB-3 mix — the disclosure files do not state the visa category directly. Vintage limitation: DOL's PERM disclosure files stopped publishing worker education after the FY2024 file, so this share is computed only over filings from vintages that carry the field, and the denominator counts only those filings. Floor: at least 10 such filings in the window.

G4 — Backlog exposure

Share of the employer's PERM filings for beneficiaries born in India or China, the two countries with the longest employment-based green-card queues. For a reader born in one of those countries, this approximates the queue they would join at that employer. Same floor and same vintage limitation as G3: citizenship left the DOL disclosure files after the FY2024 vintage, so the share is computed only over filings that carry it.

G5 — Wage percentile

The employer's median certified wage in its highest-volume occupation (SOC code) and worksite state is placed against the distribution of all certified H-1B wages for that same occupation, state and fiscal year. A 40th percentile means 40% of the market's certified wages for that occupation and state fall below the employer's median. Floors: at least 10 certified LCA rows for the employer and at least 50 wage records in the market cell. Current limitation: the comparison cell is occupation × state; a metro-level cell will replace it when the worksite-to-metro mapping is complete.

G6 — Prevailing-wage Level I share

Share of the employer's certified LCAs rated at prevailing-wage Level I, the lowest of the four DOL wage levels. Stated as a fact; readers can compare it against the market and the employer's job titles. Floor (shared with G7): at least 10 certified LCA records in the latest fiscal year on file, otherwise not published.

G7 — Third-party placement share

Share of certified LCAs that name a secondary worksite entity — positions performed at a client site rather than the employer's own location. Common in consulting and staffing business models; the number is stated without judgment. Same 10-record floor as G6.

G8 — WARN layoff notices

Count of WARN (Worker Adjustment and Retraining Notification) layoff notices naming the employer, filed in the trailing 24 months, in the five states EmployerFilings tracks: California, Illinois, New Jersey, New York and Texas. Alongside the count, the sum of the affected-worker figures the notices state. Each state is windowed against the date EmployerFilings pulled that state's file, not against the newest notice in it — otherwise a state with a quiet quarter would silently shift its own window.

What counts as one notice. State agencies publish one row per impacted worksite, and the same employer often appears under several spellings on the same filing day — a hospital system listed once per campus, a retailer once per store. EmployerFilings counts a notice once per employer per notice date: not once per site row, and not once per name spelling. The affected-worker figure for that date is the sum of the site rows behind it. Two separate notices an employer filed in one state on one day therefore count as one, which understates rather than overstates. The table on an employer page uses the same grain as the count, one row per state and notice date.

Where a notice in the window does not state a headcount, the notice is still counted and the worker total is written as "at least" that many — the site never presents a total that implies completeness it does not have. A notice is a filing by an employer, not a completed layoff: notices are sometimes withdrawn, amended or re-filed, and the worker figure is the employer's own estimate at filing time. Employers are matched to notices by the same identity rules described below; a notice that matches no employer under those rules is counted for nobody rather than attributed to a similarly-named company.

Absence from a state's WARN listing is not evidence that an employer avoided layoffs. WARN filing duties depend on employer size, site size, and the size and timing of the job losses, so many layoffs are never reportable; EmployerFilings does not track the other states and territories, so no notice filed in them appears here at all; and agency listings lag filings. For that reason a EmployerFilings page never characterises an employer's absence from these listings, and this section appears on an employer page only when at least one in-window notice exists.

Where a state agency publishes only the month a notice was posted rather than the day, the employer page shows that month ("January 2026"). EmployerFilings does not fill in a day the agency did not state.

Rescinded notices. One state agency marks a row that withdraws an earlier notice. EmployerFilings records those rows and excludes them from the count, the worker total and the table. The earlier notice remains counted: the listing does not state which filing a rescission cancels, so removing one would be an attribution EmployerFilings cannot source. Where an agency publishes a site identifier, notices are also reconciled on it, so a site an agency lists twice — or that two overlapping exports of the same listing both carry — is counted once.

Sources are the five state agencies' own public WARN listings, linked in the sources block of any employer page that shows a notice.

G9 — Department of Labor compliance record

Two federal sources, both published by the U.S. Department of Labor:

The window is 5 years of findings periods, ending on the date EmployerFilings pulled the enforcement file. It is longer than the WARN window on purpose. A compliance action is recorded against the period in which the division determined its findings occurred, and the dataset states that those dates are not the dates the case was opened or closed; cases conclude long after the conduct they concern. A two-year window would leave most published H-1B cases off the site while a page implied a recency the file cannot support.

What a concluded compliance action is, and what it is not. It is the Wage and Hour Division's own administrative record of a case it closed. It is not a court judgment, and nothing on this site says a court decided anything. The dataset's own definition of a finding covers both outcomes — the division records cases where it found nothing as well as cases where it did — which is why EmployerFilings publishes only rows that carry the H-1B violation indicator, and why the figure shown is the amount the record says was agreed to, which may already have been paid. Counts are of cases, never of violations: the division changed how it counts violations in October 2025 and states that numbers either side of that change are not directly comparable.

How the figures are shown. Dollar amounts are shown in whole dollars, rounded down, so a displayed amount never exceeds the amount in the file. Where the file records no figure or 0.00 for a column, the table shows "not stated". A case the file prints without findings dates cannot be placed in the window and is not shown.

Debarment is reported as a period, not a status. A debarment appears on an employer page only while the period the list itself published covers the day the page was built, and it disappears the day that period ends. Where a list prints its own reason text, that text is quoted exactly as written and is never reworded; where a list prints none, none is shown. Each list states the date it is effective as of, and that date is shown alongside the period. Where an agency publishes both a current list and an older file of the same kind, the older one is kept as the record of periods that have since expired and is never used to say who is debarred today.

Absence from these files is not evidence about any employer. This section appears on an employer page only when there is at least one in-window case or a current debarment period. No page states that an employer has no enforcement record, is in good standing, or is missing from a debarment list — the files record the cases the division closed and the entities the lists name, and say nothing at all about anyone else. Employers are matched to these records by the same identity rules described below; a record that matches no employer under those rules, or that review finds names a different company from the one its name would match, is shown for nobody rather than attributed to a similarly-named company.

G10 — Score

The composite EmployerFilings Score publishes with a later data refresh. The score's weights will be stated on this page in full before the first score appears.

Employer identity resolution

The same company appears under many spellings across files and years. Names are canonicalized (case, punctuation, legal suffixes, conjunctions, spaced initialisms), then matched against a hand-curated alias table; ambiguous near-matches are reviewed by a person before any merge. The standing rule: an unmerged duplicate understates a company, which is safe; a wrong merge attributes one company's record to another, which is not. Because government files also truncate and vary names in ways no rule fully unifies, an employer's filings can be split across related name entries despite this pipeline — which is why zero-filing statements on employer pages are always scoped to "under this employer name", why an automated check blocks publishing when a zero-filing page has a likely name-sibling with real filings, and why every employer page states how many raw name spellings roll up into its figures. If a split misrepresents an employer, the corrections channel merges it in the next build.

Known limitations

LCA certifications are not hires — one certification can cover multiple workers, and certified positions may never be filled. USCIS approvals count petitions, not people. PERM education requirements are a proxy, not the visa category. Fiscal-year files are cumulative and revised by the agencies; EmployerFilings reloads each quarter and stamps every page with the period used. If you find an error, use the corrections channel — reports are reviewed within 5 business days. Recent rankings are on the top sponsors page.

This page and every benchmark are informational only, not legal advice; consult a licensed immigration attorney for decisions about your case.