Accuracy

The Accuracy Ledger

Anyone can build a land report that looks confident. The hard part is being able to say how often it is right — and being willing to publish the answer when it is we have never checked. That is what this page is. It is generated by a script that queries live services; nothing on it is typed in by hand.

Signals
57
Measured
10
Unmeasured
47
Comparisons made
43

43 of 43 comparisons agreed. Generated 2026-07-10.

Class A — transcription

An authority already published the value; we read it and repeat it. The only error we can introduce is a reading error — the wrong layer, the wrong field, a mis-set spatial reference, a legend we hard-coded, an outage we swallowed. So “accuracy” here means fidelity, and it is testable: ask the same authority through a different endpoint and compare.

Class B — derivation

We computed the number ourselves from raw data — a slope from an elevation model, a wet-frequency from a stack of satellite scenes. No authority publishes the answer for your parcel, so a second endpoint proves nothing. Bounding this error needs independent ground truth, and until we have it the honest entry is UNMEASURED.

What a Class A check does not prove. Agreement means we transcribed the authority correctly. It does not mean the authority is correct. FEMA notes that roughly a third of flood claims fall outside the mapped flood zone; a faithful read of a flawed map is still a faithful read of a flawed map. We publish the fidelity because it is the part we control.

Class A — transcription (20)

Fidelity of our read, measured against the same authority reached a different way. The sample size is small and it is stated on every row — “6 of 6 agree” is a fact; “100% accurate” would be a claim we have not earned.

SignalAuthorityWhat we checkedResultChecked
Land cover / land use
land-cover
ESA WorldCover 10 m v200 (2021 epoch), Zanaga et al. 2022The dominant land-cover class over the parcel, read through Microsoft's Planetary Computer, compared against the same 2021 v200 raster read directly from ESA's own S3 bucket. The class is compared, not the percentage: the two paths resample the window differently, and a percentage disagreement would be an artefact of our test rather than an error of ours.
6 of 6 agree
2026-07-10
Regulatory flood zone
flood-zone
FEMA National Flood Hazard Layer (NFHL), layer 28 S_Fld_Haz_ArOur /28/query read of the flood zone at the point, compared against the same NFHL service's `identify` operation — a different REST operation and code path.
8 of 8 agree
2026-07-10
Soil organic carbon
soil-organic-carbon
ISRIC SoilGrids 2.0 (Poggio et al. 2021, SOIL 7:217) — global 250 m modeled soil organic carbonOur 0-30 cm depth-weighted organic carbon from the SoilGrids REST API, compared against the same three depth layers read as rasters from ISRIC's WCS on a different host, and re-weighted identically. Agreement is within 0.1 g/kg, the rounding the collector applies.
6 of 6 agree
2026-07-10
Soil pH (H₂O)
soil-ph
ISRIC SoilGrids 2.0 (Poggio et al. 2021, SOIL 7:217) — global 250 m modeled soil pH in H2OOur 0-30 cm depth-weighted pH from the SoilGrids REST API, compared against the same three depth layers read as rasters from ISRIC's WCS on a different host, and re-weighted 5/10/15 cm identically. Agreement is within 0.1 pH, the rounding the collector applies.
6 of 6 agree
2026-07-10
Soil productivity (NCCPI)
soil-productivity
USDA-NRCS SSURGO / NCCPI v3.0 (National Commodity Crop Productivity Index), via Soil Data Access (SDA)Our single aggregate SQL result, compared against the index recomputed in Python from the raw per-component rows (component-percent weighted, ruledepth = 0).
4 of 4 agree
2026-07-10
Solar PV yield
solar
EC JRC PVGIS PV-performance model (Huld et al. 2012)Annual PV yield read from PVGIS's JSON document (outputs.totals.fixed.E_y), compared against the `Year` row of the same service's plain-text table. This is the same request to the same service: it does not test the transport, it tests our read — the JSON path, the units and the row we picked, which is precisely the error a transcription can make.
4 of 4 agree
2026-07-10
Wetlands presence
wetlands
USFWS National Wetlands Inventory (NWI), layer 0 WetlandsOur /0/query read of the NWI wetland type at the point, compared against the same NWI service's `identify` operation at the smallest tolerance.
8 of 8 agree
2026-07-10
Wildfire — hazard potential
wildfire-hazard
USFS Wildfire Hazard Potential 2023, classified 5-class (270 m), ImageServer identifyOur hardcoded pixel→class legend (1..7), compared name-by-name against the ImageServer's own published legend. The service's rasterAttributeTable is empty, which is why the legend was hardcoded in the first place.
1 of 1 agree
2026-07-10
Baseline water stress (basin)
water-stress
WRI Aqueduct 4.0 baseline water stress (HydroSHEDS sub-basins), served on the Esri Living Atlas FeatureServerThe candidate check re-queries the same service by the same basin id — not independent.UNMEASURED
Coastal & storm-surge exposure
storm-surge
NOAA Office for Coastal Management SLR viewer, slr_3ft MapServer layer 0 (3ft SLR Low-lying Areas)Live-tested: the candidate second layer answers a different question (inundation depth vs. low-lying extent), so it cannot verify our read.UNMEASURED
Depth to bedrock
bedrock-depth
USDA-NRCS SSURGO corestrictions.resdept_r (cm), min over restrictions of dominant componentSecond representation via SoilWeb verified reachable. Not yet wired.UNMEASURED
Expansive / shrink-swell soil
expansive-soil
USDA-NRCS SSURGO chorizon.lep_r (% linear extensibility), max over horizons of dominant componentSecond representation via SoilWeb verified reachable. Not yet wired.UNMEASURED
Karst / sinkhole
karst
USGS Open-File Report 2014-1156 'Karst in the United States' (Weary & Doctor), served from a THIRD-PARTY Esri-hosted copy (arcgis org hoKRg7d6zCP8hwp2), not a usgs.gov hostNo second free, keyless, nationwide point-queryable endpoint publishes the USGS karst polygons.UNMEASURED
Prime farmland / capability
farmland-capability
USDA-NRCS SSURGO mapunit.farmlndcl + nonirrigated land-capability class component.nirrcapclSecond representation via SoilWeb verified reachable. Not yet wired.UNMEASURED
Radon potential
radon
EPA Map of Radon Zones (ORD/ROE_Radon layer 0), field RadonZoneA second free, keyless EPA endpoint (gispub.epa.gov) does publish the zone, and it agrees with ours — but it re-serves the same EPA Map of Radon Zones layer, so it is a mirror, not an independent authority. Verified live: both hosts return identical attributes for the same county. Bounding this needs an actual radon measurement in a building on the parcel, which is what the EPA itself recommends.UNMEASURED
Seismic hazard
seismic
USGS Seismic Design Web Services (building-codes/asce7-16), ASCE 7-16 S_DS / SDC, Risk Cat II Site Class DNot a fidelity check: the only candidate second endpoint answers for ASCE 7-22, while we publish the ASCE 7-16 design value. Those are different quantities from different code editions, and comparing them would manufacture a disagreement. The USGS Seismic Design service is in any case the sole authority; a second query to it would be our own request wearing a hat.UNMEASURED
Septic / perc suitability
septic-perc
USDA-NRCS SSURGO chorizon.ksat_r (µm/s), averaged over horizons of dominant componentSecond representation via SoilWeb verified reachable. Not yet wired.UNMEASURED
Soil drainage class
soil-drainage
USDA-NRCS SSURGO (Soil Data Access), component.drainagecl of the dominant map-unit componentAn independent second representation (SoilWeb, UC Davis) was verified reachable and keyless. Not yet wired.UNMEASURED
Soil texture (USDA)
soil-texture
ISRIC SoilGrids 2.0 (Poggio et al. 2021, SOIL 7:217) — global 250 m modeled clay/sand/silt fractionsThe published value is a USDA class we compute from three separately-read SoilGrids values, so an independent read of one of them cannot confirm it. The three inputs are checked individually where they are published (see soil-ph and soil-organic-carbon).UNMEASURED
Surface-water seasonality
water-seasonality
JRC Global Surface Water v1.3 `seasonality` layer (Pekel et al. 2016, Nature 540:418), calendar year 2020The independent JRC tile matches our epoch, but our published value is a share of the parcel, not a pixel. Recomputing that share from a second raster differs by a percentage point or two purely through resampling — an artefact of the test, not an error of ours. It needs a justified tolerance before it can be published.UNMEASURED

Class B — derivation (37)

These are ours to be wrong about. Each row names the independent ground truth that would bound its error, and whether we have obtained it. Where a row reports a measured error, read it precisely: bounding a shared input — the elevation model, say — bounds what every value derived from it starts with. It does not measure the derivation itself.

SignalAuthorityWhat we checkedResultChecked
Elevation model (DEM) — the shared input to 9 derived signals
dem-elevation
USGS 3DEP seamless bare-earth DEM (1/3 arc-second), via Microsoft Planetary ComputerThe production DEM loader is sampled at every published NOAA NGS survey mark within 5 km of five golden parcels, and compared with the mark's own orthometric height. Only NAVD 88 marks whose height is ADJUSTED, LEVELING or GPS-observed are used; marks set into bridges, buildings and other structures are excluded, because a bare-earth model is not trying to measure them. This bounds the ELEVATION error of the model. It does not measure the error of slope, HAND or any other value derived from it — only the error of their shared input.
mean |Δ| 0.904 m
RMSE 1.749 m · bias -0.367 m
worst 6.539 m · n = 22
2026-07-10
Reanalysis temperature (ERA5) — the shared input to 8 climate signals
era5-temperature
ECMWF ERA5 reanalysis, served via the Open-Meteo Archive APIThe production ERA5 client is asked for daily minimum and maximum temperature at each NOAA NCEI normals station near a golden parcel, over the station's own 1991-2020 period, and (Tmax+Tmin)/2 is averaged — the same definition NCEI uses for TAVG. The result is compared with the station's published annual normal. Three things are held equal on purpose: the period (our published normals use 2015-2024, and comparing those with a 1991-2020 normal would charge three decades of warming to our error), the definition of a daily mean, and the location. What is left is the honest residual: a ~9-11 km reanalysis grid cell against a point thermometer. This bounds the ERA5 INPUT that eight climate signals share. It does not measure any of those eight.
mean |Δ| 0.859 °C
RMSE 1.011 °C · bias +0.553 °C
worst 1.623 °C · n = 8
2026-07-10
Agricultural legacy (orchard/vineyard residue)
ag-legacy
USDA NASS Cropland Data Layer (CropScape GetCDLValue)Tried and withheld. Planetary Computer republishes the same USDA CDL rasters, so our read can be compared against them at the exact pixel: 9 of 12 point-years agree. But all three disagreements fall in 2008 and every one is between ADJACENT classes (Almonds/Grapes on a field boundary; Developed-Low/Open; Developed-Medium/High), while 2015 agrees 6 of 6. A real read error would not confine itself to one year and to neighbouring classes — a half-pixel grid offset, or a revision of USDA's own 2008 raster, would look exactly like this. Publishing 75% as the fidelity of our read, when the disagreement is most likely an artefact of our test, would be worse than measuring nothing. (Comparing the tree-or-vine decision instead agrees 6/6 and proves nothing at all: both sides would run our own regex.)UNMEASURED
Aridity / water balance
aridity
Open-Meteo Archive ERA5 daily precipitation_sum + et0_fao_evapotranspiration (FAO-56 PM ET0)The reanalysis this is computed from now HAS a measured error, bounded against NOAA NCEI 1991-2020 station normals — see the 'Reanalysis temperature (ERA5)' row. That bounds the input, not this value: the error of the derivation itself (the threshold, the index, the season definition) is still unmeasured.UNMEASURED
Aspect (N/S facing)
aspect
USGS 3DEP or Copernicus GLO-30 DEM via Planetary ComputerThe elevation model this is computed from now HAS a measured vertical error, bounded against NOAA NGS published survey marks — see the 'Elevation model (DEM)' row. That bounds the input, not this value: the error of the derivation itself (the slope operator, the flow routing, the index) is still unmeasured.UNMEASURED
Climate normals (30-yr)
climate-normals
ERA5 reanalysis via Open-Meteo Archive API (Hersbach et al. 2020)The reanalysis this is computed from now HAS a measured error, bounded against NOAA NCEI 1991-2020 station normals — see the 'Reanalysis temperature (ERA5)' row. That bounds the input, not this value: the error of the derivation itself (the threshold, the index, the season definition) is still unmeasured.UNMEASURED
Conservation easements / protected areas
protected-areas
OpenStreetMap (boundary=protected_area/national_park, leisure=nature_reserve) via OverpassUSGS PAD-US is an independent authority for US protected-area boundaries and was verified to confirm Yellowstone at the point. Not yet wired.UNMEASURED
Contamination / hazardous sites
contamination
US EPA Envirofacts EMEF/efpoints layer 0 (final Superfund NPL, ~1,379 sites)No second free/keyless endpoint for the final NPL was confirmed.UNMEASURED
Drainage / pluvial ponding
ponding-twi
USGS 3DEP bare-earth DEM (US) / Copernicus GLO-30 DSM elevation, fetched via Microsoft Planetary Computer STACThe elevation model this is computed from now HAS a measured vertical error, bounded against NOAA NGS published survey marks — see the 'Elevation model (DEM)' row. That bounds the input, not this value: the error of the derivation itself (the slope operator, the flow routing, the index) is still unmeasured.UNMEASURED
Drought exposure
drought
US Drought Monitor (NDMC / USDA / NOAA)Not a fidelity check: we publish the county's dominant USDM class, while the USDM spatial layer gives the class at the point. Those are different quantities and comparing them would manufacture disagreements.UNMEASURED
Earthquake history (observed)
earthquake-history
USGS Comprehensive Catalog (ComCat) via the FDSN event web serviceWe do no arithmetic here: `maxradiuskm`, `minmagnitude` and `starttime` go to USGS's FDSN /count endpoint and USGS does the radial selection and the count. What is unbounded is not our maths but our CHOICE of that 100 km radius, M4.5 floor and 1975 start — a judgement about what counts as seismic history near a parcel, which no second endpoint can settle.UNMEASURED
Elevation & local relief
relief
USGS 3DEP / Copernicus GLO-30 DEM via Planetary Computer (TanDEM-X 2011-2015 epoch)The elevation model this is computed from now HAS a measured vertical error, bounded against NOAA NGS published survey marks — see the 'Elevation model (DEM)' row. That bounds the input, not this value: the error of the derivation itself (the slope operator, the flow routing, the index) is still unmeasured.UNMEASURED
Environmental records nearby (EPA)
env-records
US EPA Facility Registry Service (TRI, RCRAInfo, ACRES Brownfields, NPDES, ICIS-Air) via EMEF/efpointsThe underlying locations are published by the authority; the distance is our arithmetic over them. Bounding it means re-deriving from the published geometry.UNMEASURED
Erosion potential (RUSLE LS)
erosion-rusle
USGS 3DEP bare-earth DEM (US) / Copernicus GLO-30 DSM elevation via Microsoft Planetary Computer STACThe elevation model this is computed from now HAS a measured vertical error, bounded against NOAA NGS published survey marks — see the 'Elevation model (DEM)' row. That bounds the input, not this value: the error of the derivation itself (the slope operator, the flow routing, the index) is still unmeasured.UNMEASURED
Extreme-heat exposure
extreme-heat
Open-Meteo Archive ERA5 daily temperature_2m_max (9 km reanalysis)The reanalysis this is computed from now HAS a measured error, bounded against NOAA NCEI 1991-2020 station normals — see the 'Reanalysis temperature (ERA5)' row. That bounds the input, not this value: the error of the derivation itself (the threshold, the index, the season definition) is still unmeasured.UNMEASURED
Flood / surface-water history
flood
Sentinel-2 L2A via Microsoft Planetary Computer STAC (raw bands B03/B04/B08/B11/SCL)No independent ground truth identified. Bounding this needs field observation or an independent flood/burn record for the same parcel and window.UNMEASURED
Floodplain position (HAND)
hand-floodplain
USGS 3DEP bare-earth DEM (US) / Copernicus GLO-30 DSM elevation via Microsoft Planetary Computer STACThe elevation model this is computed from now HAS a measured vertical error, bounded against NOAA NGS published survey marks — see the 'Elevation model (DEM)' row. That bounds the input, not this value: the error of the derivation itself (the slope operator, the flow routing, the index) is still unmeasured.UNMEASURED
Gas transmission pipeline proximity
gas-pipeline
US EIA Natural Gas Interstate & Intrastate Pipelines (served via HIFLD/ArcGIS FeatureServer)Live-tested: the EIA ArcGIS host is a single-page app that answers HTTP 200 with HTML for any REST path, so no second endpoint could be confirmed.UNMEASURED
Growing season & frost
growing-season
Open-Meteo Archive ERA5 daily temperature_2m_min (9 km reanalysis)The reanalysis this is computed from now HAS a measured error, bounded against NOAA NCEI 1991-2020 station normals — see the 'Reanalysis temperature (ERA5)' row. That bounds the input, not this value: the error of the derivation itself (the threshold, the index, the season definition) is still unmeasured.UNMEASURED
Historical land-use timeline
land-use-history
USGS Landsat Collection-2 Level-2 Surface Reflectance (via Microsoft Planetary Computer STAC)No independent ground truth identified. Bounding this needs field observation or an independent flood/burn record for the same parcel and window.UNMEASURED
Köppen-Geiger class
koppen
Open-Meteo Archive ERA5 monthly T & P normalsThe reanalysis this is computed from now HAS a measured error, bounded against NOAA NCEI 1991-2020 station normals — see the 'Reanalysis temperature (ERA5)' row. That bounds the input, not this value: the error of the derivation itself (the threshold, the index, the season definition) is still unmeasured.UNMEASURED
Landform position (valley / ridge)
landform-tpi
USGS 3DEP bare-earth DEM (US) / Copernicus GLO-30 DSM elevation via Microsoft Planetary Computer STACThe elevation model this is computed from now HAS a measured vertical error, bounded against NOAA NGS published survey marks — see the 'Elevation model (DEM)' row. That bounds the input, not this value: the error of the derivation itself (the slope operator, the flow routing, the index) is still unmeasured.UNMEASURED
Nearby services (grid power & water main)
services
OpenStreetMap via Overpass APIThe underlying locations are published by the authority; the distance is our arithmetic over them. Bounding it means re-deriving from the published geometry.UNMEASURED
Neighboring land use
neighboring-landuse
OpenStreetMap via Overpass API (landuse=industrial|landfill|quarry|farmland, man_made=works)The underlying locations are published by the authority; the distance is our arithmetic over them. Bounding it means re-deriving from the published geometry.UNMEASURED
Outdoor air quality (PM2.5/NO2)
air-quality
CAMS (Copernicus Atmosphere Monitoring Service, ECMWF) via Open-Meteo Air-Quality APIThe CAMS cell is ~11 km; a co-located ground monitor would bound it.UNMEASURED
Precipitation seasonality
precip-seasonality
Open-Meteo Archive ERA5 daily precipitation_sum aggregated to 12 monthly normalsThe reanalysis this is computed from now HAS a measured error, bounded against NOAA NCEI 1991-2020 station normals — see the 'Reanalysis temperature (ERA5)' row. That bounds the input, not this value: the error of the derivation itself (the threshold, the index, the season definition) is still unmeasured.UNMEASURED
Regulatory-review triggers
regulatory-triggers
USFWS Critical Habitat (ESA §4)USFWS IPaC is the authoritative determination path and would bound this screen.UNMEASURED
Road access & distance
road-access
OpenStreetMap via Overpass API (highway=*)The underlying locations are published by the authority; the distance is our arithmetic over them. Bounding it means re-deriving from the published geometry.UNMEASURED
SAR surface water (cloud-penetrating)
flood-sar
Sentinel-1 GRD (VV) via Planetary Computer STACNo independent ground truth identified. Bounding this needs field observation or an independent flood/burn record for the same parcel and window.UNMEASURED
Slope / gradient
slope
USGS 3DEP seamless (US) or Copernicus GLO-30 DEM via Planetary ComputerThe elevation model this is computed from now HAS a measured vertical error, bounded against NOAA NGS published survey marks — see the 'Elevation model (DEM)' row. That bounds the input, not this value: the error of the derivation itself (the slope operator, the flow routing, the index) is still unmeasured.UNMEASURED
Snowfall days (climatology)
snow-days
Open-Meteo Archive ERA5 daily snowfall_sum (cm) (9 km reanalysis)The reanalysis this is computed from now HAS a measured error, bounded against NOAA NCEI 1991-2020 station normals — see the 'Reanalysis temperature (ERA5)' row. That bounds the input, not this value: the error of the derivation itself (the threshold, the index, the season definition) is still unmeasured.UNMEASURED
Terrain ruggedness
ruggedness
USGS 3DEP / Copernicus GLO-30 DEM via Planetary ComputerThe elevation model this is computed from now HAS a measured vertical error, bounded against NOAA NGS published survey marks — see the 'Elevation model (DEM)' row. That bounds the input, not this value: the error of the derivation itself (the slope operator, the flow routing, the index) is still unmeasured.UNMEASURED
Usable acreage
buildable-area
DEM slope, FEMA NFHL flood polygons and USFWS NWI wetlandsThe elevation model this is computed from now HAS a measured vertical error, bounded against NOAA NGS published survey marks — see the 'Elevation model (DEM)' row. That bounds the input, not this value: the error of the derivation itself (the slope operator, the flow routing, the index) is still unmeasured.UNMEASURED
Vegetation health & trend
vegetation
Sentinel-2 L2A via Planetary Computer (raw NIR/Red)No independent ground truth identified. Bounding this needs field observation or an independent flood/burn record for the same parcel and window.UNMEASURED
Volcanic proximity
volcano
Smithsonian Global Volcanism Program, Volcanoes of the World (Holocene) via GVP GeoServer WFSThe underlying locations are published by the authority; the distance is our arithmetic over them. Bounding it means re-deriving from the published geometry.UNMEASURED
Wildfire — burn history
wildfire
MODIS MCD64A1 Collection 6.1 monthly burned-area (Giglio et al. 2018, RSE 217:72)No independent ground truth identified. Bounding this needs field observation or an independent flood/burn record for the same parcel and window.UNMEASURED
Wind resource
wind-resource
ERA5 reanalysis (ECMWF) served via Open-Meteo Archive APIThe reanalysis this is computed from now HAS a measured error, bounded against NOAA NCEI 1991-2020 station normals — see the 'Reanalysis temperature (ERA5)' row. That bounds the input, not this value: the error of the derivation itself (the threshold, the index, the season definition) is still unmeasured.UNMEASURED

Ground truth: what the land told us

Most of the 47 unmeasured signals above have no independent endpoint to check them against, because the independent endpoint is the ground. A percolation test is the check for our septic screen. A driller's log is the check for our depth to bedrock. So we ask buyers what their surveys found, keyed to the coordinate, and count the answers here.

We could not read the corrections just now — the database answered with an error. This is a fault on our side, not a count of zero. We would rather say that than show you a zero we cannot stand behind.

What this page deliberately does not contain

The exact indices, thresholds and weightings behind each signal stay ours — see the methodology. What we publish is which authority each value comes from, whether we transcribed it or derived it, and how often our answer matched an independent one. That is the part a buyer needs in order to trust us, and the part a competitor cannot copy without doing the work.