Known Issues

NOAA Climate — single-station coverage

NOAA’s Climate Data Online API assigns each county a single representative weather station for annual temperature, precipitation, and snowfall summaries. For small or topographically uniform counties this is a fair simplification. For large counties or counties with significant elevation variation (mountains, coastal/inland gradient) the assigned station can be unrepresentative of where most residents live.

Status: Always-on caveat under the climate section on every county page. Multi-station population-weighted averaging is on the roadmap but is a pipeline-level refactor we’re not shipping for launch.

Estimated impact: ~200–400 counties have meaningfully unrepresentative readings.

NOAA snowfall — specific known-bad counties

Sixteen counties show annual snowfall figures that don’t match local reality. We’ve suppressed the value on these pages and added an explanatory note in the climate section.

Order-of-magnitude errors:

• Teton County, WY — Reads ~82″; the assigned station is below the Jackson Hole population centers, which are near ski-resort terrain where snowfall is materially higher.
• San Diego County, CA — Reads ~13″; the assigned station appears to be at a higher-elevation backcountry location. The coastal urban areas where the population lives effectively never see snow.

Gulf-Coast / deep-South false readings — counties where the station returns several inches of annual snow in a climate that averages above 67°F (snow is effectively never observed at population centers):

• Alabama: Dale, Houston, Mobile
• Florida: Bay, Escambia, Jackson
• Georgia: Bacon, Coffee
• Louisiana: Acadia, Calcasieu, Cameron, Jefferson Davis, St. Tammany

Status: Fixed at display layer.

CFPB HMDA — low-volume mortgage denial rate

CFPB’s Home Mortgage Disclosure Act data reports origination counts and denial rates per county. In counties with very few originated loans the denial rate becomes statistically unstable — a single denial in a county with one origination reads as “100%,” and even 50–75 originations leaves enough variance that a single underwriter’s pattern can swing the number 20 points. We hide the denial rate when a county had fewer than 75 originated loans in the reporting year.

Status: Fixed at display layer. Affects roughly 130 of the smallest US counties (Loving County, TX; King County, TX; Willacy County, TX; etc.).

USDA Food Access — dense-urban borough values

The USDA Food Environment Atlas reports the share of each county’s population with low supermarket access. Typical counties range 5–40%. Manhattan comes back at 0.0001%, which is implausibly low even for a dense-grocery borough. The likely cause is a tract-aggregation issue between USDA’s underlying tracts and the borough FIPS codes. We suppress the field when it falls below 0.5%.

Status: Fixed at display layer. We’re investigating the underlying join.

FDIC Deposits — bank-headquarters outliers

FDIC’s Summary of Deposits attributes a bank’s deposits to the county where the bank is headquartered rather than where its branches and customers sit. For most counties this distinction doesn’t matter. For a small set of counties that host the headquarters of a large bank — a rural ag-lender (Ziebach County, SD with $786M in deposits against ~2,400 residents), an independent city hosting regional bank HQs (Fairfax City, VA), or Manhattan (Wall Street institutions) — the per-capita figure runs far above local wealth.

Status: An inline caveat is shown next to FDIC data when the deposit-to-population ratio crosses $75,000 per resident. Affects about 60 counties.

Census ACS — suppressed small populations

The Census Bureau’s American Community Survey suppresses estimates for very small subpopulations to protect respondent privacy. Suppressed estimates come back as a literal 0 rather than “not reported,” so a county with three Asian residents reads as “0% Asian.” We render these as “<0.5%” instead, matching Census’s own disclosure semantics.

Status: Fixed at display layer.

Education Data — multi-county district rollup

The NCES Education Data Portal aggregates per-pupil spending and enrollment at the school-district level. Multi-county districts get attributed to a single county in our pipeline, which inflates that county’s enrollment figure. New York County (Manhattan) is the most visible case — the New York City public school system is rolled up into Manhattan.

For two counties the rollup also produced a per-pupil-spending figure far from state-reported actuals: Orleans Parish, LA ($1,673/pupil, implausibly low) and Kent County, MD ($53,216/pupil, implausibly high). We suppress the per-pupil-spending field on those two pages.

Status: Open for enrollment. Per-pupil-spending suppressed for the two known outliers. A full fix requires per-district enrollment shares NCES doesn’t directly publish.

HUD Fair Market Rent — missing for roughly 400 counties

HUD publishes Fair Market Rent at the metro-area level. For about 400 counties our pipeline currently fails to map the county FIPS to the relevant HUD area, leaving the housing section partially empty. Affected counties include Los Angeles County, Harris County (Houston), and Miami-Dade.

Status: Open. The housing section on each affected page shows an explanatory note in place of the missing rent ladder. The fix involves a county-to-metro lookup against HUD’s own crosswalk file.

USDA Agriculture Census — not yet loaded

USDA NASS publishes farm counts, farmland acres, principal-crop output, and livestock numbers at the county level through its QuickStats product. The fetcher that loads this data into our pipeline is currently in a degraded state and returns no data for any county. Pages show metadata for the source but no agriculture census figures. Food-security (SNAP, food insecurity) and food-access data come from USDA ERS and are unaffected.

Status: Open. Tracked.

Connecticut — planning regions vs. legacy counties

Connecticut replaced its eight legacy county governments with nine state-defined planning regions in 2022. The Census Bureau formally adopted the new geography for the 2024 release cycle, but most other federal agencies still publish data against the old county FIPS codes. CountyPulse uses planning-region FIPS where available and falls back to a max-by-region merge from the legacy counties when the source only publishes legacy data.

Status: Display banner is shown on all CT planning-region pages. Per-source legacy-to-planning-region routing is partially complete; some sources still show “not available.”

FEMA Disaster Counts — classification edge cases

FEMA’s public declarations dataset includes Emergency Management declarations for events that affected the county only as a destination for evacuees (the most-cited example: Katrina-era Emergency Management declarations issued in landlocked states that received Gulf Coast evacuees). We classify each declaration with a small AI gate to separate “event happened locally” from “jurisdiction received an EM declaration in support of an event elsewhere,” but the classification is imperfect.

Status: Best-effort. If a county’s disaster count looks much higher than its geographic exposure suggests, the AI classifier likely included regional-support EMs.

General data freshness

Most federal datasets have a 1–2 year publication lag. The data year for each metric is shown alongside the value. Storm-event records and FEMA declarations refresh more frequently; ACS and Education Data Portal release annually with a longer lag.