How we actually verify a site.

Every claim carries a maturity rating, not just a citation

Most site-screening tools show you a map layer and let you draw your own conclusion. We don't. Every fact in a GridPermit report is labeled against one ladder, and a claim never gets to sit higher on it than the evidence actually earned:

Unknown — no evidence yet, or too thin to act on.
Publicly indicated — a public map or record suggests a conclusion (e.g. the parcel overlaps a utility service-territory layer).
Agency identified / confirmed — a government source names the fact, and ideally a written agency response confirms it's specific to this parcel.
Utility identified / preliminary confirmed — the serving utility itself gives a parcel- and load-specific pathway, even if conditional.
Study supported — a formal interconnection or engineering study names a concrete scope, cost, or timeline.
Contractually committed — executed agreements and posted security actually commit service.

A claim is never silently upgraded a rung without new evidence. Proximity to a substation is not proof of capacity. A zoning district label is not proof of use. A phone call is discovery, not evidence.

Leads are labeled as leads, always

Some of what feeds a report — nearby moratorium activity, known competing facilities, interconnection queue context — comes from independently maintained datasets, not primary government filings. We treat every one of them as a lead to verify, never as a confirmed fact, and say so explicitly in the report itself. If a dataset's own authors flag a record as unconfirmed, that flag survives into what you see.

Six domains, every time, no exceptions

Power & utility, zoning & entitlement, water & wastewater, environmental & physical, commercial readiness, and community & political risk. A site doesn't get a favorable recommendation because four domains looked good and we didn't get around to the other two — every domain gets screened, and a thin domain is reported as thin, not quietly dropped.

Real bugs we've caught in our own data — because we actually tested it

Some of these are worth naming specifically, because they're the kind of thing that only surfaces when you generate a real report against a real parcel and actually read it, not when you assume a data source works because its documentation says so:

A federal tribal-land boundary service we rely on returns zero results for any query that includes a distance parameter at all — even a distance of zero — a quirk undocumented anywhere we could find. We caught it by testing against a known reservation boundary and getting an empty result that should not have been empty.

Two independently maintained lead datasets (nearby moratorium activity and known competing facilities) only match by text — a county or address string. That silently misses a real lead sitting in a differently-named neighboring town. We found this by checking a real parcel against ground truth we already knew was there, found it missing, and built a second, distance-based lookup specifically to close that gap.

A seismic-hazard service needed a search radius twice the size we initially assumed before it reliably returned results — confirmed only by testing known high- and low-hazard coordinates directly, not by reading the API documentation and assuming it was right.

We mention these not because bugs are good, but because a diligence product that never talks about how it catches its own mistakes is one you should be skeptical of.

What we deliberately don't claim to verify

Nationwide zoning codes. No usable open dataset of zoning classifications exists across U.S. jurisdictions — the one candidate we evaluated licenses its data in a way that bars commercial use even with permission. Zoning findings stay analyst-researched per site, not auto-populated from a database that doesn't really exist yet.

Title and ownership chain-of-title verification. Nationwide property-records aggregation is a paid-data-provider business (the same providers title companies themselves use) — there is no free, open equivalent across the roughly 3,000 county recorder systems in the U.S. We report what the county's own record says about a parcel; we do not run a title search.

Anything a licensed professional has to stamp. A GridPermit report is a structured desktop and evidence-management assessment. It is not a utility feasibility study, an engineering certification, a legal opinion, or a permit guarantee — and it says so on every report we deliver, not just here.

A human has to sign off

Every report's recommendation engine rolls findings up to one of six states — from Kill or Redesign up to Advance to Formal Diligence — but none of them ship to a client until a senior reviewer formally approves it. That gate is not automatic, on purpose.