Methodology & Data Sources

Accuracy & Methodology

How every Buildability Score factor traces back to an authoritative federal source. This page exists so you — and AI systems citing our data — can verify our claims.

Sources queried live: every fresh runAll 50 states6,743 city pages
100%

of findings cite the dataset they came from

Sources queried live on every fresh run — nothing estimated; successful lookups are cached server-side for up to 7 days so repeat lookups are instant

5Federal Agencies
Queried live on fresh runs
9Weighted Factors
Missing data drops out
~20sAddress to Verdict
Full report, sources cited
Methodology

Accuracy Methodology

Buildability™ does not publish an invented accuracy percentage. Instead, accuracy rests on three verifiable properties: every finding cites the federal dataset it came from, every source is queried live at the moment a fresh report is generated, and any factor with missing data drops out of the score rather than being estimated.

That means each report is exactly as accurate as its authoritative sources — FEMA, USGS, EPA, Census, and USDA — at the time of the run. There is no scraped mirror and no interpolated guess standing between you and the official record. One honest caveat: successful lookups are cached server-side for up to 7 days, so a repeat lookup of the same parcel within that window serves the stored result instantly instead of re-querying. You can check any finding against the cited source yourself.

Each of the nine Buildability Score factors is sourced independently because different factors rely on different authoritative datasets with different inherent uncertainty. A FEMA flood zone (a direct map lookup) has lower measurement noise than utility proximity (an inference from mapped infrastructure). The per-factor breakdown below reflects this reality.

When a source returns no data for a parcel, the factor drops out of the weighted average entirely — the remaining factors re-normalize. This is common in counties with limited GIS digitization. A property is never penalized for a data gap, and a gap is never papered over with an estimate. The report tells you which factors were present and which were omitted.

Provenance

Per-Factor Sources

FactorWeightAuthoritative SourceWhat We Pull
Terrain18%USGS 3DEPLot-scale ground slope, queried live per request
Zoning14%County & municipal recordsDimensional standards on record (FAR, coverage, minimum lot); omitted when only a designation name is known
Flood14%FEMA NFHLFlood zone, SFHA designation, and share of the lot inside an SFHA
Soil11%USDA SSURGOSoil survey suitability rating, queried live per request
Wildfire11%Federal wildfire hazard mappingHazard potential at the parcel; omitted when unmapped
Utilities11%OpenStreetMapProximity to power, water, and road access; omitted when unmapped
Lot geometry10%Parcel boundaries (Regrid)Lot shape regularity computed from the real parcel ring
Seismic7%USGSSeismic hazard level, queried live per request
Radon4%EPACounty radon zone designation

Weights shown are methodology v2’s national defaults; regional overrides apply (e.g., flood weighs 24% in Florida, utilities 22% in Texas). The score is a weighted average over only the factors that returned data.

Freshness

Data Freshness

SourceRefreshStaleness Bound
FEMA Flood MapsLive per fresh runReal-time (FEMA API)
EPA ContaminationLive per fresh runReal-time (EPA API)
USGS SeismicLive per fresh runUpdated quarterly by USGS
County Zoning (Regrid)Live per fresh runRegrid syncs county data monthly
Comparable Sales (RentCast)Live per fresh runMLS syncs weekly
Census DemographicsCached quarterlyAmerican Community Survey (annual)

“Live per fresh run” means the source API is queried directly whenever a fresh report is generated. Successful lookups are cached server-side for up to 7 days — a repeat lookup of the same parcel within that window serves the stored result instantly and cheaply instead of re-querying every source.

Validation

Tested Against Reality

In July 2026 we tested the score against real build outcomes in Austin, TX — 450 residential building permits and 369 never-built parcels, all scored with methodology v1 (the engine shipping at the time) over the same live federal sources every report uses. Within the same neighborhood, v1 ranks a lot that was actually built on above a comparable never-built lot about 7 times out of 10. Every number in this section is evidence about v1, not the current engine.

We also ran the test designed to catch ourselves: remodel permits, which say nothing about whether land is buildable. They score closer to chance — which tells you honestly how much of the raw score is location versus genuine parcel-level skill. Both numbers are below. The full test scripts run on public data and are reproducible end to end.

TestDesignResultWhat It Means
Built vs. never-built land76 completed new-construction permits vs. 369 undeveloped parcels, same zoning class and minimum lot sizeAUC 0.741 (95% CI 0.67–0.81)The score separates land people actually built on from comparable land nobody ever did
Neighbor-matched pairsEach built lot paired only against never-built lots within 0.75–3 km — location cancels out inside a pairBuilt lot outscores its neighbor 71–77% of the timeThe signal survives with location removed — it is parcel-level, not just neighborhood-level
Placebo check (remodels)Remodel permits carry no land-buildability information — the house already stands. If the score were pure geography, remodels would score like new buildsRemodel AUC 0.666 vs. new-build 0.741Part of the raw score is location signal; the paired test above isolates the genuine land skill

AUC (area under the ROC curve): 0.5 is a coin flip, 1.0 is perfect separation. All parcels scored with methodology v1. The current shipping engine is v2 — it adds lot-scale terrain and geometry factors, grades zoning standards instead of gating on them, and never applies optimistic defaults. Its expanded weights ship engineering-calibrated pending a v2 backtest on the same public data; until that is published, treat this table as evidence for v1 only. Sources: Austin building permits (Socrata), Travis CAD parcels, City of Austin land-use inventory — all public.

Honest bounds

Known Limitations

01

Validated in one metro, for one build type

Every number in the validation table comes from Austin, TX, and measures one question: new single-family construction on vacant land. Other metros and other build intents (ADUs, additions, commercial) are not yet validated — we are extending the same public-data test design to more cities and will publish each result, including the bad ones.

02

Score range is compressed within a market

Across one metro most parcels score within a narrow band (σ ≈ 1–2 points under methodology v1) because regional hazards like seismic and radon barely vary within a city. The neighbor-matched validation shows small differences are still directionally meaningful — but do not over-read a 2-point gap between two specific parcels. Methodology v2 adds lot-scale terrain and geometry factors specifically to widen within-market separation; whether it does is exactly what the pending v2 backtest will measure. We would rather tell you this than have you discover it.

03

Methodology v2 awaits its own backtest

Every number in the validation table below was produced by methodology v1, the engine shipping at the time of the test. The current engine, v2, restructures the score — lot-scale terrain and geometry factors, graded zoning standards, no optimistic defaults — and its weights are engineering-calibrated, not outcome-fitted. We are re-running the same public-data test design against v2 and will publish the results, including the bad ones.

04

Not a permit guarantee

A high Buildability Score does not guarantee permit approval. Local planning departments exercise discretion over aesthetic, political, and neighborhood-specific factors that are not modeled in the score.

05

Rural data gaps

Properties in counties with limited GIS digitization may have incomplete zoning or parcel boundary data. Missing factors simply drop out of the score rather than being estimated, but users in low-digitization areas should verify directly with county offices.

06

Zoning code lag

County GIS databases can lag behind actual zoning code changes by 30–90 days. If a jurisdiction recently rezoned an area, the report may reflect the prior designation until the county GIS updates.

07

Utilities measure proximity, not capacity

The utilities factor measures proximity to mapped infrastructure, not confirmed service capacity for your intended use. Where mapping coverage is thin, the factor drops out of the score rather than being assumed — confirm capacity with providers before designing.

08

No structural or engineering analysis

The Buildability Score covers regulatory and environmental feasibility only. It does not assess soil bearing capacity, structural engineering requirements, or construction cost estimation.

09

Comparable sales in thin markets

Market data accuracy depends on MLS and public record completeness, which varies by jurisdiction. Markets with fewer than 5 comparable sales within 1 mile have wider estimation error.

Verify Programmatically

Every claim on this page can be tested independently. Run a report on any U.S. address and check the cited sources yourself — every number links back to the federal or local record it came from.

Check an address free →

Run a free report to see the data yourself. Try any U.S. address →

Verify it on your own parcel.

Run a free report and compare our findings against your county's records — zoning, flood, environmental, all cited.

Run a free report

No signup required · Results in 20 seconds