H-1B LCA disclosure data, FY2010-FY2025, without cleaning 34 Excel files.

8,918,577 case rows, 8,238,223 unique applications, 9,375,219 worksite rows and a 1,916,599-row employer x occupation x fiscal-year panel, each row traceable to its DOL source file and row.

H-1B LCA Employer x Occupation Dataset, FY2010-FY2025

Who is this dataset for?

• Immigration-policy, labor-economics and workforce researchers who need sixteen years of LCA filings in one schema
• Compensation and talent-intelligence analysts benchmarking offered wages by occupation, state and employer
• Data engineers who would otherwise reconcile 34 DOL spreadsheets with changing layouts and cumulative quarterly files

What can you do with it?

How to use the files

1. Download the free sample and open README.md and the data dictionary to learn the columns, row roles and identity flags.
2. Read LIMITATIONS and the ZIP corrections before you rely on a number; check employer_id quality for the employers you care about.
3. Query the Parquet files directly with DuckDB or pandas; start from the applications and panel tables for counts.

Try real records before buying

The table below uses actual rows from this release’s free sample, not simulated data. Blank cells are left blank as supplied. The preview is shortened for readability; download the sample for all columns and complete cell values.

Download sample CSV — no account or email required → · Full sample package (ZIP) →

Which edition and license should you buy?

One fixed edition (2026-10-09) covering DOL fiscal years 2010 to 2025, delivered as Parquet only. The Individual / Academic tier is for one named person’s academic or professional work; the Business / Commercial tier is for internal use within one legal entity. Neither allows redistribution.

Why buy this instead of using free source data?

The free DOL files must otherwise be found, converted from Excel, mapped across layout changes and SOC vintages, de-duplicated across cumulative quarterly files and reconciled to DOL totals. This edition does that once and documents where it still falls short.

What happens after payment?

After secure checkout you receive a private download link for one ZIP (49 files) with the Parquet tables, data dictionary, methodology, limitations, QA files and checksums. Read the ZIP corrections: some documents inside the ZIP are older than the ones on this site.

What You Get

• Case files, one per fiscal year (8,918,577 rows, 71 columns)
• Worksites (9,375,219 rows)
• Applications (8,238,223 rows, one per case number)
• Employer x occupation x fiscal-year panel (1,916,599 rows, 45 columns)
• SOC 2000-to-2010 and 2010-to-2018 crosswalk tables
• Data dictionary, methodology, limitations, release notes, patch notes, quality report, certification report, changelog, release metadata, SHA-256 checksums and manifest, and the 34-file source inventory
• QA files: reconciliation to DOL published totals, tie-out and verification results, the identity audit, the independent review report, and the privacy redaction ledger (the 4,608 changed cells, without their values)

What This Does Not Contain

• FEIN, contact-person, attorney/agent and preparer columns, employer street addresses and phones, and law-firm fields.
• Any claim that an employer id is a legal entity or corporate family.
• CSV files (this edition is Parquet only).
• Updates after FY2025, or an API.

What the file does that the raw DOL files do not

• Unique applications and status records are separate columns, so counts are not double-counted across fiscal years or cumulative quarterly files.
• Every row carries its source file and row; filed values stay next to the normalized ones.
• SOC vintage and mapping class are explicit; split mappings list candidates instead of inventing one code.
• Known limitations and reconciliation exceptions are published, not hidden.

Methodology & QA

H-1B rows of 34 DOL disclosure files are mapped to one schema with source file and row kept on every row. Unique applications are keyed on case number; repeated year-to-date snapshots and later status events are separate row roles. The certified/withdrawn classification is a provisional rule (and an estimate for FY2010-FY2015). Employer ids follow the provisional R5-v1 policy and can be undone through employer_id_l1. The material limitations known to us are listed on this page and in LIMITATIONS.md in the package; the redaction, identity and reconciliation exceptions are explained there in full.

Read the full methodology →

Certification status: Engineering: certified with documented exceptions, based on builder QA and one AI reviewer; no human audit has been done. No independent human audit and no legal opinion on rights or license have been obtained; the license is the Persistfolio Data License v1.0. — read the full certification report →

Data Dictionary

Full data dictionary (JSON) → · Complete dictionary with enumerated values and date semantics (Markdown) →

Known Limitations

• FY2010 has no visa-class or offered-wage-unit column; H-1B status and the wage unit are assumed and labeled.
• The certified/withdrawn classification is a provisional rule. For FY2010-FY2015 it rests on a submit-date proxy that agrees with DOL totals in aggregate only; no error bound is claimed.
• Eight source files differ from DOL’s published totals (six by 0.9% to 2.4% of certified; cause not proven). Worker-position totals are not reproduced.
• Employer ids are provisional. One company can carry several ids and some ids span several offices; a name-variant link is never an assertion of one company.
• SOC codes are kept as filed. DOL’s FY2011 and FY2012 files hold Excel-corrupted codes, some FY2022 codes are not standardized, and split SOC 2010-to-2018 mappings are not resolved to one code.
• Prevailing wage level is absent in FY2010-FY2014 and FY2016 files.
• Free-text fields are kept as filed except 4,608 cells in which pattern matching found e-mail addresses, tax-id-format numbers, phone numbers or long digit runs; those are replaced by tokens and the original values are not in the files. The redaction is pattern-based: it is not a guarantee that no personal data remains, it does not detect personal names, and it has not been reviewed by a human or by a lawyer.
• 84,759 ids (employer, evidence-component and candidate-group ids) are opaque replacements for ids that had been derived from excluded fields. The replacement is one-to-one, so joins and counts are unchanged, but those ids cannot be recomputed from the published columns.
• An application whose last loaded status is CERTIFIED may be withdrawn later; the data ends with the FY2025 files, and 22,593 case numbers have no original decision in the loaded files.
• willful-violator and H-1B-dependent flags are as filed on each filing and are not attributes of a company.
• Source-file defects (blank padding rows, truncated names, odd status values) are kept, not repaired. DOL may have restated records after publication; this is a snapshot of the 34 listed files.
• FY2025 Q1 differs from DOL’s published certified count by +1,733 and the cause is unresolved.
• Not independently reviewed: SOC vintage assignment, employer-id quality and wage annualization correctness.
• Verification is the builder’s QA plus one independent AI reviewer (same vendor). No human has audited the data, the redaction or the identity audit, and no lawyer has reviewed the rights position or the license.
• Employer names are reproduced from DOL’s public files, including some that are or contain an individual’s name (sole proprietors, professional practices). A heuristic screen found about 8,500 such-looking distinct names with many false positives; the redaction does not detect personal names. Public availability does not remove every legal obligation that may apply to your use. Correction and removal requests: info@persistfolio.com (see the license, section 7).
• The files are Parquet only; there are no CSV files in this edition.

FAQ