hmda-analysislisted

# HMDA Analysis (descriptive) Pulls HMDA LAR data from the CFPB API and produces **descriptive** cuts — lending by county/state/tract, top lenders, and a **CRA-proxy income distribution** — via the published, audited **hmda-analyzer** package. ## FIREWALL — read before anything else (non-negotiable) **This skill is DESCRIPTIVE ONLY.** It does not perform, and the AI must not produce using it: - disparate-impact or disparity-ratio analysis - protected-class (race/ethnicity/sex) stratified denial or approval analysis - any "fair lending" screening, inference, statistical significance, or disparity claim - any interpretation of the CRA-proxy output as CRA performance, a CRA rating, or an assessment-area result **If the user asks for inferential fair-lending analysis, decline and explain the descriptive/inferential distinction:** this skill counts and distributes what was lent (descriptive); disparity/disparate-impact analysis draws inferential conclusions about *why* and *whether lending is discriminatory*, which requires court-defensible methodology, protected-class stratification, and significance testing that this skill deliberately does not do. Do **not** point the user to any fair-lending tool as the "v1 alternative" — just hold the line on the distinction. The installed `hmda-analyzer` package *does* expose disparity functions (`disparity_ratio`, `denial_rate_by_race`, `denial_reasons_by_race`, `generate_disparity_report`, and `summary_table`, whose output is a d