Methodology

Our primary source is the U.S. Department of Labor, Office of Foreign Labor Certification (OFLC) Disclosure Data — the official quarterly releases of H-1B Labor Condition Applications (LCA), PERM permanent labor certifications, and Prevailing Wage Determinations.

• DOL OFLC Disclosure Data (H-1B LCA, PERM, Prevailing Wage) — the source for every wage, premium, and filing count on the site.
• FLC Wage Library (OEWS prevailing-wage levels) — referenced for context but not yet ingested; see Data Processing below for how this affects the lottery wage-level breakpoints.

We cover fiscal years FY2021–FY2026 (through Q2). We begin at FY2021 because it is the first fully FLAG-system year with a stable, consistent schema; FY2020 mixed legacy iCERT and FLAG records and is excluded.

We refresh quarterly, typically within about two weeks of each DOL release. Quarterly files are cumulative within a fiscal year, and our loader de-duplicates on case number, keeping the record from the highest available quarter so re-running an update is idempotent.

• Last refresh: June 1, 2026
• Next expected refresh: September 2026

The H-1B program is in active regulatory flux. The figures and tools on this site reflect the rules as we currently understand them; all are subject to ongoing litigation and may change.

• A $100,000 fee on new H-1B petitions, effective September 21, 2025.
• A wage-weighted lottery, effective February 27, 2026, for the FY2027 cap. Each registration receives pool entries by OEWS wage level: Level IV → 4 entries, Level III → 3, Level II → 2, Level I → 1.
• These rules are subject to ongoing litigation; this page tracks the current state.

The following steps describe how raw filings become the published numbers. They are the data contract behind every statistic on the site.

Prevailing-wage premium

The prevailing-wage premium is the median of (offered wage ÷ prevailing wage) − 1, stored as a fraction and shown ×100 as a percent. It is a median, not a mean, chosen specifically to resist outlier distortion — a single absurd filing can drag a mean badly, but barely moves the median. We do not describe it as an average.

Outlier handling

Roughly 5.8% of rows are flagged as suspicious and excluded from every published statistic (a NOT is_suspicious filter), as well as from search. These are implausible values such as annualized wages produced by unit-of-pay errors — for example an hourly rate mistakenly multiplied to an eight-figure annual salary. Flagged rows are excluded, not retained: they do not appear in company stats, aggregations, the dashboard parquet, or detail search.

SOC occupation-code normalization

SOC codes are normalized to their 6-digit base, stripping O*NET detail suffixes (including doubled suffixes such as 11-1021.00.00). Deprecated 2010 SOC codes are crosswalked toward 2018 codes. A small number of same-occupation rows remain split across legacy and new codes — a known limitation we are actively consolidating.

Company name normalization

Company names are consolidated using fuzzy matching against a manually curated alias table, so variants like “Google LLC”, “GOOGLE INC.”, and “Alphabet Inc.” collapse to a single canonical entity. A single-token guard prevents unrelated short names (e.g. “Tesla” vs “Tesla Dentistry”) from being merged; borderline cases are reviewed quarterly.

Lottery wage-level breakpoints

The Level I–IV wage thresholds shown in the lottery tools are derived — the median observed annual prevailing wage per disclosed wage level — and are not the official OEWS / FLC Wage Library figures, which we have not yet ingested. They are labeled “approximate” wherever they appear. The lottery-odds selection rate r defaults to 0.30, an explicitly labeled placeholder pending a real registrations-vs-cap figure; the resulting odds are a historical-distribution estimate, not a guarantee.

Dashboard “weighted wage”

The dashboard’s “weighted wage” is a filing-count–weighted mean of per-group medians (SUM(median × count) ÷ SUM(count)), not a true pooled median. We label it honestly as a weighted wage and do not call it a median.

Wage annualization

Wages reported by the hour, week, bi-week, or month are converted to an annual figure (hourly annualized at 2,080 hours/year). A reverse guard handles records mis-tagged as hourly: a value above $500/hour (≈ $1.04M/year if annualized) is treated as already-annual rather than multiplied again, which prevents inflated eight-figure salaries.

PERM field availability

PERM filings from FY2025 onward do not include the country-of- citizenship or class-of-admission fields, which DOL removed from disclosure data that year. We store these as null for affected rows; they are not used in wage analysis.

• LCA filings precede actual H-1B issuance: a filing does not guarantee the visa was approved or that the worker was hired.
• PERM data covers permanent labor certifications only — not the subsequent I-140 or I-485 stages.
• Small SOC × metro combinations may have insufficient sample size; thin combinations are flagged and de-indexed until the data thickens.
• A small number of same-occupation rows remain split across legacy and new SOC codes, as noted under SOC normalization above.
• Lottery wage-level breakpoints are derived approximations, not official OEWS figures.
• Quarterly DOL releases occasionally include retroactive corrections, so figures for a given period can change slightly between refreshes.

• June 2, 2026: Published the methodology, about, privacy, and terms pages.
• June 1, 2026: Consolidated SOC occupation codes to their 6-digit base (stripping O*NET detail suffixes) so a single occupation in a metro is no longer split across suffixed codes.
• May 2026: Switched the published prevailing-wage premium from a mean to a median, and began excluding outlier filings (≈5.8% of rows) from every published statistic and from search.
• May 2026: Added the in-browser salary dashboard and the FY2027 lottery decision tools (wage-level lookup, odds estimator, wage-threshold calculator).
• April 2026: Initial publication: company, job, and wage-by-city pages from DOL OFLC disclosure data, FY2021 onward.