DATA DICTIONARY

raw/gao-table1.csv: manual transcription of GAO Table 1, not microdata. Count fields are integer properties. Risk rows low/moderate/high are mutually exclusive; total is their sum. first_street_score is a label, not numeric interval arithmetic. SFHA means special flood hazard area. Unmapped means digital FEMA map unavailable, including paper-only maps. Geography 50 states and D.C., residential and nonresidential. table_label and vintage fields preserve the source's different dates.

derived/high-risk-location.csv: one record per high-risk map category. share_pct is count / high-risk total * 100, stored to eight decimal places and shown to one decimal place in the article. denominator is properties, not households. No confidence bounds claimed.

raw/hypothetical-inputs.csv: six invented scenarios; monetary columns are nominal educational USD, not real observations. replacement_loss is the cost assumed by the model; eligible is the adjusted insured amount before deductible; limit and deductible apply separately to building and contents. Zero limit means no modeled coverage, not missing data. Living costs receive no modeled payment. No blank numerical inputs. Formula outputs are in derived/household-scenarios.csv.

coverage-audit.csv: blank reader worksheet. Empty answer/date/document fields mean not checked, not zero or no coverage. Do not place personal addresses or policy details in a public download.

All data files are UTF-8 comma-separated CSV with a single header row. Keep caveats when reusing. See methodology.txt and sources.csv. No private property records included.

VERSION 1.1
Run python reproduce.py from the clean extracted kit. Zero coverage_limit means no modeled coverage, not missing data. Blank coverage-audit answer fields mean not checked. All scenario amounts are invented nominal USD; no price year or typical-loss claim is assigned. The script verifies each output column and records results under verification/.
