Methodology

• An adversarial audit spot-checked 22 country-year cells against the source PDFs across six fiscal years. All 22 matched exactly, including deliberately odd mixed-precision values.
• The FY2025 nationality table was separately verified line-by-line against an independent published copy and matched on all 191 values.
• Class-level tables were verified by reconciling every column against each source document's own printed grand total. All 12 years reconcile exactly — the check that catches a dropped or misread row.

• Country label drift. Only 173 of 245 raw spellings appear in all 18 years; casing flips from Title Case to UPPER at FY2015, and CHINA - MAINLAND/CHINA, MACAU/MACAU S. A. R., REPUBLIC OF PALAU/PALAU are the same countries. Untreated, any cross-year comparison silently drops rows.
• Aggregate rows excluded. NON-NATIONALITY BASED ISSUANCES and UNKNOWN NATIONALITY OR STATELESS are not countries and would corrupt every ranking.
• 45 community outcome labels downgraded to "unclear" because the quoted text did not support them. This moved reported-approved from 21% to 13%.

• A printed 0.00% is genuinely ambiguous in the source — every application issued, or no applications at all. Flagged, never rendered as 100% approval.
• Two negative adjusted rates are real (Luxembourg FY2006, Kuwait FY2009): overcomes exceeded refusals.
• Microstates swing between 0% and 100% on tiny denominators. Excluded from rankings and from country pages.
• Three FY2024 class rows don't sum (B1/B2 +2, BBBCC −1, BBBCV −1). The deltas net to zero, the signature of a reclassification. Kept as printed rather than silently fixed.
• Official tables were extracted from PDFs via a language model rather than a deterministic parser. The verification above is what stands behind them.

Rank, year-over-year change, change versus FY2019, volatility, percentile and peer sets are computed by us and are not published by any government. They are derived arithmetic on the published rates, reproducible from the CSVs on the data page.