How The Grant Map checks, labels, and audits its data
Errors in assistance data can cost homeowners time and money. This page documents where records come from, which checks run automatically, what receives source review, how confidence labels are assigned, and how corrections are handled.
Where records come from
Records are compiled from government, utility, nonprofit, and administering-agency material. Most carry a cited URL, but a URL can be missing; as of July 20, 56,954 of 56,956 live-city records had one on file. A URL may be a direct program page, an application page, or a broader agency page, and a working link does not by itself confirm every stored amount, deadline, rule, or funding status. Program pages expose freshness and confidence context so readers can tell those cases apart and confirm current facts with the administrator.
The corpus is compiled with automated and AI-assisted research across thousands of sources, then normalized into one schema. At that scale, any record can carry an error. A changed amount is treated as a lead to verify rather than a fact to publish, and measured link health, source-review coverage, and audit results are public instead of implied by a blanket "verified" claim.
Guide topic counts on this page and related guides were recomputed together on July 20 from live city data with python -m bin.guide_stats. The June 17 state report is a separate frozen analysis with a different city-association set; its historical counts should not be mixed with this live-city snapshot.
Scheduled link health checks
The July 20 live-city analysis contained 12,320 unique nonblank cited URLs. A scheduled checker requests URLs and records response categories such as live, not found, connection failure, redirect, or bot-block. A successful response confirms only that a page responded; it does not validate the stored program facts or current funding.
Repeated 404 or 410 responses can trigger an automatic closed status under the link-health rule. Connection failures and bot blocks are handled separately because they do not establish that a program closed. The resulting status can still be wrong if an agency moved content or the cited URL was generic, so readers should verify with the administrator.
Selected federal figures normalized from published caps
For selected federal benefits with a nationally published fiscal-year cap, the normalizer uses a shared value to avoid inconsistent city copies. That does not mean every federal-program rule or local delivery detail has one national value, and the agency's current terms remain controlling. Examples in the June 2026 ruleset included:
- USDA Section 504: the June ruleset separated standard grant and loan ceilings from higher declared-disaster exceptions. At our July 11 review, USDA described $10,000 and $50,000 standard grant and combined limits, plus $15,000 and $55,000 limits for qualifying presidentially declared disaster circumstances. Verify the current page and exception criteria before using any figure.
- VA disability housing grants: SAH up to $126,526 and SHA up to $25,350 for fiscal year 2026. The separate HISA benefit publishes lifetime tiers up to $6,800 or $2,000 depending on the VA category; confirm the applicable tier and current terms with VA.
In June 2026, an audit found 547 city-page records showing the SHA cap at the SAH figure. We corrected the affected normalized output and added a regression rule to reduce the chance of recurrence. The incident also shows why a shared constant, a working link, or a generated page is not a substitute for independent checks.
Human review for city and state programs
Local program details change quickly. Records are prioritized for source review on a rotating basis, and automated or AI-assisted updates can be staged for comparison. We do not claim that every record or change receives manual review before publication. A changed amount should be treated as a lead until the cited page supports it.
When a record reaches source review, we fetch its cited page and assess program identity separately from amount evidence. model-identity-reviewed means a model-assisted review judged the page and program identity a match, but stores no verbatim identity proof and did not establish the amount. source-verified is narrower: a contextual excerpt must be found verbatim on the reviewed page, contain the current positive maxAmount, refer to the same program and amount role, and remain bound to the reviewed URL, name, administrator, and amount. The same number elsewhere on a multi-program page is insufficient. Both review-derived confidence tiers expire after 90 days until refreshed; sourceReviewStale preserves that age signal without presenting old evidence as current. The June aggregate audit used a prior rubric that treated some identity-only reviews as confirmed, classified status-0 connection failures as dead, and sampled replicated rows rather than unique fact keys. We retired its published rate on July 11, 2026 and publish no replacement percentage until a strict run completes. Neither label proves other fields or future funding.
What we measure, in the open
We publish quality measurements at /api/dataset/health: response status for cited links, the status of dated sample audits, and separate counts for reviewed records, identity-supported records, and amount-verified records. These metrics have important denominators and limitations. Link health is not fact accuracy, an unconfirmed record is not automatically correct or incorrect, and source review applies only to the reviewed record and date.
Past-deadline detection
Deadline text can be stale or recurring. A scheduled sweep flags records whose parsed date appears past while status remains available, sending candidates to review rather than proving closure. Parsing can miss vague, malformed, or context-dependent text, so readers must confirm the current deadline with the administrator.
What we do not do
Just as important as what we check is what we deliberately stay out of:
- We do not process applications. Confirm the application route directly with the agency or organization that administers the program; our cited URL may be a general page.
- We do not charge homeowners to use The Grant Map. Be skeptical of guaranteed-approval or finder-fee pitches, and independently confirm any administrator, inspection, financing, contractor, or project costs before proceeding.
- We do not guarantee eligibility or funding. A directory record describes a possible resource and can be incomplete or stale. The administering organization decides eligibility, funding, amount, and application status.
- We are not a government agency and are not affiliated with one.
Who runs this, and how to report an error
The Grant Map is run by The Grant Map Team at StanHattie LLC in Waukee, Iowa. Corrections go to support@thegrantmap.com. If you spot a stale figure, dead link, or inaccurate description, send the record and supporting source so it can be reviewed.
For a sense of how this rigor shows up in practice, see our guides on the $10,000 home improvement grant phrase, roof-repair grants and assistance, and senior home repair assistance, or browse records for veterans and seniors.
Screen possible matches in your city
Answer a few questions to build a list to verify. Only the administrator confirms eligibility and current funding.
Screen possible matches Browse all cities