Source documents

Every figure on this site comes from the public files below. Nothing is hand-entered, inferred from a private dataset, or supplied by a third party. Each entry gives the publisher's original, the exact copy this deployment scored, and what that file cannot establish.

Snapshot 2026-10-27T06:24:53.986685+00:00 · 2 of 5 sources loaded.

Read a score as a question, not an answer.

These files support exactly one claim: an employer's filing pattern is unusual relative to its peers. They cannot establish intent, wrongdoing, or that anything unlawful occurred.

On this deployment the only signal based on an adjudicated finding — a prior Wage and Hour Division violation — carries weight in the config but can never fire, because that source is not loaded (3 optional sources absent). Every score you see is therefore built entirely from filing structure and registry shape. A high score is a prompt to go read the underlying filings; a low score is not a clearance.

LCA Disclosure Data (H-1B / E-3 / H-1B1)

Loaded

U.S. Department of Labor — Office of Foreign Labor Certification (OFLC)

The spine of the dataset. Every filing-level fact — employer name, FEIN, job title, wage offered, prevailing wage, wage level, worksite, secondary entity, attorney, case status — is read from this workbook.

Publisher page: https://www.dol.gov/agencies/eta/foreign-labor/performance
Direct download (publisher): https://www.dol.gov/media/LCA_Disclosure_Data_FY2026_Q3.xlsx
Copy scored here: LCA_Disclosure_Data_FY2026_Q3.xlsx (240.2 MB)
data/raw/lca/LCA_Disclosure_Data_FY2026_Q3.xlsx
SHA-256: f8ca8448a528671f784d3692923775a13741c99caec2816b2fe30c8885bd28b3
Offline archive: Preserved this revision is retained even after the publisher replaces it
Retrieved: 2026-08-26T22:54:27.893995+00:00
Publisher last-modified: Fri, 07 Aug 2026 17:33:30 GMT
Update cadence: Quarterly cumulative workbook for the federal fiscal year.

Used for: Filing volume and peer percentiles · Wage-level and prevailing-wage comparisons · Job-title vs. wage-level consistency · Secondary (end-client) placement rates · Shared-address clustering

Checking an individual record.

DOL does not publish a stable per-case permalink. To verify one filing, open the workbook below and search the CASE_NUMBER shown on the record. Each cluster page also offers an lcas.csv containing the exact rows used.

What this file does not establish

An LCA is a pre-petition wage attestation. It is not an approved petition, not a visa, and not evidence that anyone was hired.
Counts here are filings, not people. Amendments, renewals, and multi-worksite filings each add rows, and one filing can cover several workers. Filing volume is not headcount.
Certified filings only. DOL will not certify a wage below the prevailing wage, so the wage-underpayment signal is near-dead by construction — a clean score on it means nothing was detectable, not that nothing happened.
Employer names are free text and undeduplicated. One company can appear under several spellings, splitting its filings and depressing its apparent volume.

Active Franchise Tax Permit Holders

Loaded

Texas Comptroller of Public Accounts (data.texas.gov)

State registry corroboration. Supplies entity existence, registration (responsibility begin) date, registered address, organization type, and right-to-transact status for employers matched to a Texas record.

Publisher page: https://data.texas.gov/dataset/Active-Franchise-Tax-Permit-Holders/9cir-efmm
Direct download (publisher): https://data.texas.gov/api/views/9cir-efmm/rows.csv?accessType=DOWNLOAD
Copy scored here: active_franchise_taxpayers.csv (440.0 MB)
data/raw/tx/active_franchise_taxpayers.csv
SHA-256: 321ad2d529e082820eceb3c137d631236cfefe051b324440c4e9614da89944bb
Offline archive: Preserved this revision is retained even after the publisher replaces it
Retrieved: 2026-08-28T17:46:44.929176+00:00
Publisher last-modified: Sat, 22 Aug 2026 10:26:26 GMT
Update cadence: Open-data refresh, roughly weekly.

Used for: Thin-footprint signal · New-entity / high-volume signal · Registered-address corroboration

Checking an individual record.

Individual records are directly checkable. Given a taxpayer number, the publisher's API returns that exact row — see the per-record link shown beside any matched employer.

What this file does not establish

No FEIN in this extract. Employers are matched to registry records by normalized name, so a generic or common name can attach the wrong entity. Treat any single registry match as a lead to check, not an established fact — follow the per-record link and confirm the address and registration date.
A registered address is a mailing or registered-agent address. Several sponsors sharing one suite very often means one shared registered agent or accountant, not shared operations. This is the most common way to misread the address-clustering signal.
Active permit holders only. Absence means 'not in this list' — not defunct. Out-of-state, exempt, and differently-named entities are all missing.

Franchise Tax Officer and Director Information

Not loaded

Texas Comptroller of Public Accounts

Officer/director names used to detect the same people behind several sponsors.

Publisher page: https://data.texas.gov/browse?q=franchise%20officer

Used for: Officer-reuse signal · Officer-based cluster edges

Checking an individual record.

Not loaded.

What this file does not establish

Not present in this deployment. The officer_reuse weight in config.yaml is therefore inert — no employer can score on it.

USCIS H-1B Employer Data Hub

Not loaded

U.S. Citizenship and Immigration Services

Approved-petition counts, which can be compared against LCA filing volume. A large gap between attestations and approvals is meaningful context.

Publisher page: https://www.uscis.gov/tools/reports-and-studies/h-1b-employer-data-hub

Used for: LCA-vs-USCIS mismatch signal

Checking an individual record.

Not loaded.

What this file does not establish

Not present in this deployment. The lca_vs_uscis_mismatch weight is inert.

Wage and Hour Division enforcement actions

Not loaded

U.S. Department of Labor — Wage and Hour Division

Adjudicated violations — the only genuinely adverse finding in the model.

Publisher page: https://enforcedata.dol.gov/views/data_summary.php

Used for: Prior WHD violation signal

Checking an individual record.

Not loaded.

What this file does not establish

Not present in this deployment. The prior_whd_violation weight is inert.

Offline archive

Publishers overwrite these files in place — DOL replaces the quarterly workbook and the Texas extract is regenerated continuously. Once that happens, a ranking produced last month can no longer be checked against the file it came from. Every distinct revision the pipeline has scored is therefore copied to external storage, content-addressed by SHA-256 so an unchanged file is kept only once.

Archive root: /Volumes/YOTUO/h1btracker-archive
Status: Online 6 snapshots recorded

Layout: objects/<aa>/<sha256> for the files themselves, snapshots/<utc>.json for what each run saw.

Re-deriving a number yourself

Open the group page for the employer you care about.
Download lcas.csv from that page — those are the exact filing rows behind every count and wage figure shown, including each CASE_NUMBER.
Search those case numbers in the DOL workbook above to confirm them against the publisher's own file.
For any Texas registry claim, follow the per-record link beside the employer. It queries the Comptroller's API for that taxpayer number and returns the row itself.

If a figure here disagrees with the publisher's file, the publisher's file is correct and this is a bug worth reporting.