Skip to the document
Madhuopen lab

Public data, kept honest · part 3 of 4

adopted13 September 2026cost: local CPU; runtime ≈ 2.5 minutes a run

History and coverage: every file since 2010

5 runs

The question

Did the change make the dataset more accurate without moving the headline numbers for a reason nobody can explain?

adopted

History: DOL LCA files FY2010-FY2022 (4 legacy schemas mapped in scripts/convert_lca_legacy.py) + USCIS hub FY2010-FY2021 via Chrome agent; fiscal_years now 2010-2026; ZIP zero-padding — 9.64M rows -> 7.17M cases; KC last-3 headline…

Runs5
Kept5

Every run

Every run
#what changedverdictKC filings (3 y)geo resolved %notes
6Spec (user): 'last 3 years' = 36-month window counted monthly; per-year buckets + KC by_month serieskept597799.98window first Sep2023-Aug2026 then user aligned to Jul2023-Jun2026
8Spec (user): data from start of 2023 -> calendar-year views (CY2022 partial..CY2026 YTD) + full-scope monthly series; USCIS Employer Data Hub FY2023-FY2026 integrated…kept597799.98USCIS KC: 3959 approvals FY2023-26YTD, 551 petitioners; FY2023 archived CSV looks under-complete
11USCIS hub: FY2023 (and FY2022) re-exported from Tableau via Chrome agent; FY2023 now 57,415 rows vs 33,332 in archived CSV; uscis_hub.py uses Tableau files, min FY 2023kept597799.98README caveat about FY2023 under-coverage removed
15Year attribution: case counted in FY/month of its FIRST decision (status = latest). Matches h1bdatahub per-year counts exactly (Children's Mercy 28/44/31 FY2023-25 nationally)kept594999.98KC 3-yr 5977->5949 (cases first decided before Jul 2023 leave the window); totals unchanged
16History: DOL LCA files FY2010-FY2022 (4 legacy schemas mapped in scripts/convert_lca_legacy.py) + USCIS hub FY2010-FY2021 via Chrome agent; fiscal_years now 2010-2026; ZIP…kept590599.6289.64M rows -> 7.17M cases; KC last-3 headline unchanged at 5949; runtime 854s; FY2010-14 have no worksite ZIP (city-based geo)
public dataH-1Bhistory