All groups included
No wage floor applied
Net-new = positions flagged NEW_EMPLOYMENT, i.e. genuinely new hires
rather than renewals of people already on staff.
Decompressing dataset…
Source. U.S. Department of Labor, Office of Foreign Labor Certification — LCA Disclosure Data. This is the public federal dataset that commercial H-1B sites repackage; nothing here is scraped or licensed.
Period coverage is not continuous. OFLC publishes a cumulative file through Q3 of the current fiscal year, and separate Q4 files for closed years. This build therefore holds three isolated Jul–Sep quarters (FY2023 Q4, FY2024 Q4, FY2025 Q4) plus a continuous FY2026 Q1–Q3 (Oct 2025 – Jun 2026). Those combine into a continuous twelve months from Jul 2025 to Jun 2026, and give clean like-for-like Jul–Sep comparisons across 2023, 2024 and 2025. The FY2026 Q3 file is cumulative — it contains every case decided from 1 October 2025 onward, which is where the FY2026 Q1 and Q2 figures come from. Periods are chips rather than a slider precisely because the gaps are real — do not read two adjacent chips as consecutive time unless their labels say so.
The October 2025 federal shutdown distorts FY2026 Q1. DOL's FLAG system went offline when the shutdown began on 1 October 2025 and OFLC stopped issuing decisions. The data shows it plainly: 1,388 decisions on 1 October, then between one and four per day for the rest of the month, resuming on 3 November with visible catch-up spikes on 7 and 14 November. FY2026 Q1 therefore holds roughly two months of decisions rather than three, and November and December are inflated by backlog. The quarter is flagged with ⚠ in the period strip. Nothing is missing from the file — the gap is real-world, not a data defect — but do not read FY2026 Q1 as a normal quarter or compare it against another Q1 at face value.
What an LCA is — and isn't. An LCA is the wage-and-worksite filing an employer makes before petitioning USCIS. It proves intent to sponsor, not an approved visa or a filled job. Employers routinely file more LCAs than they use, so treat counts as sponsorship appetite rather than headcount.
No petition is counted twice. The cumulative Q3 file overlaps the Q4 files: 5,896 cases appear in two source files, in every instance because the LCA was certified once and then withdrawn later (a median of 409 days later). Each case is kept exactly once, at its earliest decision, so a petition certified in July 2023 and withdrawn in May 2026 counts in FY2023 Q4 where the sponsorship actually happened, rather than being carried forward into FY2026. Selecting several quarters therefore adds them cleanly — FY2026 Q1 + Q2 + Q3 sums to the FY2026 total with nothing repeated, and the six quarters partition all 799,016 petitions.
Petitions vs. worker positions. One LCA can cover many positions. Across the file the median is 1 position per filing, but a few employers file blanket LCAs covering dozens at once. Rankings sort by petitions by default, and any employer averaging 10+ positions per filing is marked with a ×N badge in the Workers column. Read those as sponsorship capacity, not hires.
Wages are annualized (hourly × 2080, monthly × 12, weekly × 52, bi-weekly × 26). Where a pay range is filed the midpoint is used. Values below $15k or above $2M are dropped as data-entry errors. Percentiles are rounded to $500. A median shown without a marker is exact; one marked ≈ is estimated from the filtered subset by interpolating the weighted distribution of group medians. Selecting a period counts as filtering.
Employer names are canonicalized. The raw files spell the same company many ways. Case, punctuation and legal suffixes are normalized so one company isn't split across several rows, and the most frequent original spelling is shown. This is deliberately conservative — genuinely distinct subsidiaries stay separate.
Net-new counts positions flagged NEW_EMPLOYMENT rather than CONTINUED_EMPLOYMENT (renewing someone already on staff). This is the most useful column for a job search: a firm filing thousands of petitions that are almost all renewals isn't hiring, it's maintaining.
vs prevailing is offered wage ÷ the DOL-determined prevailing wage for that role and area. 100% means paying exactly the legal floor. Very high values usually mean the occupation's prevailing wage comes from a low survey baseline — common in medicine and academia — not that the job pays several times market. Level is the DOL prevailing-wage skill level, I (entry) through IV (expert), averaged and weighted by petitions.
Occupation groups roll the SOC occupations into 11 categories using the official SOC major group (first two digits of the code). Technology & Software is major group 15, Healthcare & Medicine is 29 and 31, Education & Academia is 25, and so on; low-volume groups collect into Other & Support. A few titles are filed under two major groups in the source; each is assigned the group it appears under most often. Group counts reconcile exactly to the dataset total, and the count on each chip reflects every other active filter but not the group selection itself.
Grain. Two tables: one row per employer × occupation × worksite state carrying period-independent attributes, and one row per that key × fiscal quarter carrying everything that varies over time. Individual case numbers are not included, which is what lets the whole dataset live in a single offline file.