Acquire
Download or query the authoritative federal publisher. Raw source files are cached without modifying the synced project materials.
Methodology
Evidence before
interpretation.
The product is designed to make the data’s limits as visible as its values. This is the standard for every release.
Constrain
Apply an explicit inclusion rule: population threshold, official HBCU flags, published business population, or named HMDA geography.
Preserve
Keep identifiers, vintages, units, margins of error, relative standard errors, and suppression flags beside the values.
Derive
Calculate only simple, disclosed presentation measures. Every derived share shows its numerator and denominator definition.
Validate
Check unique keys, record counts, plausible domains, cross-file joins, missing-value rules, and fixed source benchmarks.
Publish
Separate sourced facts from editorial framing and withhold composite scores until every transformation is reproducible.
Published vs. derived
Simple math.
Visible labels.
B02001_E003÷B01003_E001
No separately calculated ratio MOE in this release.B25003_E002÷B25003_E001
Owner-occupied units divided by occupied units.H_MEDIAN×2,080
Presentation calculation—not a distinct BLS estimate.Action 3÷Action 1 + 3
Narrow two-outcome measure, not overall application denial.—/100
Publication gate
A score is a model.
It must be inspectable.
BlackMaxxing Scores will remain unpublished until all seven conditions are versioned and testable.
- Variable dictionary
- Reference population
- Transformations and normalization
- Weights and sensitivity analysis
- Missing-data treatment
- Uncertainty and suppression policy
- Geography crosswalk and update cadence