Data status

Failures are entities a source couldn't return (for example, places incorporated after 2020 that the 2020 ZIP file doesn't know). Expected counts use the 2026 Census gazetteer.

Validation

29 of 29 checks against hand-checked values pass. Details on Sources.

Notes by source

• Census ACS 2020-2024 5-year housing: Path: api (https://api.census.gov/data/2024/acs/acs5, keyed). Tables B25001 B25003 B25024 B25034 B25035 B25077; line labels checked against Census metadata. Counties 3144/3144 in release; Connecticut = 9 planning regions 09110-09190 in both the release and the 2026 gazetteer (all matched). Places 2148/2149 in release; not in release (all fields null): 4870070 Starbase city TX; 3908859 Brinkhaven village OH read from ACS GEOID 3929288 (re-coded in the 2026 gazetteer, same ANSI code). Open-ended medians kept at the bound with a flag: median_year_built 1939-or-earlier (Census code 1938) 4 counties/149 places, 2020-or-later (code 2021) 0/0; median_value $10,000- (code 9999) 0/1, $2,000,000+ (code 2000001) 0/6. Nulls, counties: median_value 5 insufficient_sample; moe_median_value 5 insufficient_sample; moe_median_year_built 4 open_ended_median. Nulls, places: median_value 48 insufficient_sample; median_year_built 2 insufficient_sample; moe_median_value 48 insufficient_sample, 7 open_ended_median; moe_median_year_built 2 insufficient_sample, 149 open_ended_median; owner_share 1 zero_denominator. Housing-unit totals B25001/B25034/B25024 disagree in 0 geos.
• Census 2026 Gazetteer (counties, places): 3144 counties and county-equivalents in 50 states + DC; 1227 Texas and 922 Ohio active incorporated places (FUNCSTAT A, LSAD not 57).
• Census TIGERweb boundaries (map geometry): Counties: Census cartographic boundaries (500K, January 1, 2025 vintage, clipped to the shoreline), fetched per state and generalized by the server at 0.01 degrees for counties.json (3144 shapes, 76,915 vertices, 1260 KB; 3109 conus, 30 ak, 5 hi; 0 small counties refetched finer; Honolulu County without the Northwestern Hawaiian Islands) and 0.03 degrees for the contiguous-US overview us.json (3109 counties, 0 borrowed from the 0.01 pass, 49 state outlines, 37,492 vertices, 518 KB). Places: Census cartographic boundary Incorporated Places 500K (January 1, 2025 vintage, clipped to the shoreline; 2 newer places use their legal boundary), Douglas-Peucker 0.1 km after projecting (places.json, 2149 shapes, 103,629 vertices, 1412 KB). Albers equal-area conic on GRS80, km with 1 decimal, y = -northing (north up in SVG); Alaska and Hawaii in their own spaces. Paths are absolute M/L with holes: draw with fill-rule evenodd. Rings dropped because generalization broke their nesting: 61 (counties 5, overview 26, places 7, states 23).
• OpenFEMA NFIP Redacted Claims v3: Window: dateOfLoss 2016-01-01 through 2026-09-07 (newest loss in the 2026-09-09 refresh). County key = the record's countyCode as reported; records_all_time counts every record since 1970. paid = amountPaidOnBuildingClaim + amountPaidOnContentsClaim + amountPaidOnIncreasedCostOfComplianceClaim (gross, as reported; a few are negative for reissued checks). median_water_depth_in = median exteriorWaterDepth > 0 (inches above the lowest adjacent grade, outside the building); median_interior_water_depth_in = median interiorWaterDepth > 0 (interior water marks, same reference); the *_reported fields are their denominators. The older waterDepth field is not used: labelled inches, it is mostly feet (values 1-4 plus a 99 code). basement/crawlspace/slab/elevated_share = foundationType 2 / 3 / 1 / 4-6 over records with foundationType 1-6 (foundation_type_reported; the undocumented code 0, common before 2023, counts as not reported). basementEnclosureCrawlspaceType is not used: its crawlspace code 3 appears in no record and its basement codes include enclosures under elevated buildings. latest_loss per county is the newest loss on record for that county, all time. Connecticut: NFIP v3 countyCode (and censusGeoid) use Connecticut's 8 legacy counties (09001-09015), not the planning regions; no crosswalk applied. Legacy-county figures are under the top-level key ct_legacy_counties. 1,606 window records have no countyCode and 740 are in territories (outside counties.json). ZIPs: reportedZipCode with 5 digits; ZIP+4 values keep their first five (226); shorter or empty values are left out (zip_excluded). The v3 counts run above the plan's v2 numbers almost entirely through 2026 losses filed after v2 froze on 2026-06-01 (see validation).
• NOAA nClimDiv county precipitation + 1991-2020 Climate Normals: (a) nClimDiv county precipitation through 2026-08: 3099 counties in the contiguous US + DC (NCEI code 18511) from 1895, Alaska (30 boroughs/census areas) from 1925, Hawaii (5 counties) from 1991; record_from = each county's first complete year for wettest/driest (through 2025). Normal = mean of 1991-2020 calendar-year totals, rounded half up; anomalies = value minus that rounded normal (matches NOAA Climate at a Glance for calendar years; CAG's 12-month departures use a window-specific base, so last12_anom_in differs from CAG by a few tenths). Connecticut: nClimDiv still uses the 8 legacy counties 09001-09015 -> ct_legacy_counties; the 9 planning regions 09110-09190 are not in counties (station fields only, in ct_planning_regions). No nClimDiv series: Lexington city, VA (51678). (b) 1991-2020 station normals from 8424 US stations publishing ANN-PRCP-AVGNDS-GE100HI (774 SNOTEL mountain snow sites excluded): the nearest standard station (completeness S, 24+ years) within 50 km of the Census internal point, else the nearest provisional one (P, 10-23 years). 3117 of 3135 counties have one (83 provisional); median distance 14.7 km. station_normal_annual_in = ANN-PRCP-NORMAL at the same station.
• USDA soil series (SSURGO components + NRCS series descriptions): 160 series pages; 141 with an OSD summary paragraph, 9 with labelled OSD fields only.
• USDA NRCS SSURGO via Soil Data Access: 3379 soil survey areas, versions dated up to 2026-08-05. 33 counties without SSURGO county-overlap rows (mostly Connecticut planning regions, which SSURGO still maps as the old counties, and parts of Alaska). 75 counties with under 50% of land area mapped. Unknown overlap symbols: {'CT005': 222, 'CT009': 218, 'CT003': 197, 'CT007': 196, 'CT001': 173, 'AK261': 171, 'CT011': 150, 'PR-127': 142, 'CT013': 138, 'AK280': 133, 'PR013': 125, 'PR023': 107} 33 of them are filled by clipping the Census county boundary (ssurgo_places.py); 0 remain without soil data.
• SSURGO on Census place boundaries (SDA spatial): 2149 TX/OH places and 33 fallback counties (CT planning regions and counties the SSURGO overlap table can't place). Boundaries generalized (0.00025-0.001 degrees) before clipping. 0 places without a TIGERweb boundary.
• US Drought Monitor county statistics: Percent of county area. 'now', 'prev_week_d1', 'd1_series', 'd2_series' and the _52 fields use statisticsType=1 (traditional = cumulative: d0 = D0-D4, d1 = D1-D4 'in drought', d2 = D2-D4 ...; none = 100 - d0). 'now_cat' uses statisticsType=2 (categorical: each category alone; none + d0..d4 = 100). Values rounded half-up to 1 decimal from the source's 2 decimals (so a rounded block can sum to 99.9 or 100.1). Series are the 53 maps in 'weeks', newest first; the _52 fields cover the newest 52 of them (weeks[0..51]). weeks_d*_any_52, max_d1_52 and max_category_52 are computed on the published 1-decimal values, so each record agrees with its own series: a share under 0.05% of the county (0.0 in the series) does not count. That leaves out 350 county-weeks of 0.01-0.04% D1-D4 in 127 counties (D2-D4: 234 in 93). max_category_52 is the worst category above 0.0% of the county in any of those weeks, D0 (abnormally dry, not drought) included; 'none' = no D0-D4 at all. CT: the API returns the 9 planning regions (09110-09190) for all 53 maps, so they sit in 'counties' like any county; no old-county (09001-09015) rows. released = 'Map released: September 24, 2026' on the USDM homepage (its 'Data valid' date matches as_of); the API itself returns only mapDate/validStart/validEnd. Every county has all maps. Source integrity checks passed (cumulative monotonic and none + d0 = 100, categorical sums to 100, cumulative = sum of categorical, all within rounding).
• Census 2020 ZCTA-to-place relationship file: ZIP listed when the land overlap is >=10% of the place or >=25% of the ZIP.