Source and coverage
This edition uses actual public HMDA records from the 2024 and 2025 snapshots for Delaware, Rhode Island, Vermont and Wyoming. There are 195,598 source records before eligibility filters: 92,919 for 2024 and 102,679 for 2025. Each downloaded state-year file contains fewer than 50,000 records. Exact download dates, URLs and SHA256 hashes are retained in the receipts. These are selected-state data, not a national estimate.
State pages and lender-state pages include only that state. The main lender profiles and two-lender comparison cover all four included states; their state breakdowns link to individual state profiles. The homepage opens on Vermont and provides a state selector. The loan-type and loan-size reference pages currently use Vermont as their stated reference geography. Every data page displays its own geographic coverage.
A comparable population
Eligible records have action_taken 1 (originated) or 3 (denied); lien_status, occupancy_type and construction_method equal 1; total_units between 1 and 4; open-end_line_of_credit, reverse_mortgage and business_or_commercial_purpose equal 2; loan_purpose in 1, 31 or 32; and loan_type in 1, 2, 3 or 4. Each combination of year, type, purpose, amount band and conforming status is calculated independently.
Rates and spreads
Rate spread is APR minus the average prime offer rate for a comparable loan, measured in percentage points. Pricing uses originated loans with a nonmissing rate spread. Missing spread counts and shares are reported separately. Quantiles use linear interpolation, and the histogram has 20 intervals from -1 to +3 in 0.2-point steps, plus an underflow and overflow bucket. Interval upper bounds are exclusive. Interest-rate quantiles use originated records with both rate spread and interest rate reported.
Denial rate and reasons
Denial rate is denied applications divided by originated plus denied applications. Withdrawals, incomplete files, approvals not accepted and purchased loans are excluded. Denial reasons are counted once per application, even if repeated across reason fields. An application may have several reasons, so shares can exceed 100 percent. Only reasons reported on at least 30 denied applications are displayed.
Costs and missing values
Points and credit shares use originations that report the corresponding field. Median points as a percent of loan amount use positive points and positive loan amounts. Cost ratios use reported costs and positive amounts. Numeric NA, Exempt, 1111 and blank values become null, not zero. Geographic identifiers remain strings; ratio and age buckets are preserved. The Data Browser's loan_to_value_ratio column is normalized to the combined-loan-to-value field used by this pipeline.
Publication thresholds
A data cell must have at least 30 eligible applications. Pricing needs at least 30 reported spreads, and every other quantitative field needs at least 30 observations. An unavailable field appears as not available. A CLTV or DTI bucket distribution is withheld in full if any nonempty bucket has fewer than 30 observations. Cells below the threshold are not exported. This threshold improves reliability; it is not a guarantee against inference from overlapping public aggregates.
Limits of comparison
Public loan and property amounts are interval midpoints, not exact balances. Public HMDA does not contain credit scores, and certain fields are bucketed or suppressed. Lender and geographic comparisons reflect borrower mix, products and timing. They do not establish causality, explain an individual decision, or identify the best lender for a person. No national spread comparison is published until a matching national dataset is loaded.