What this dataset does not contain
Every omission below is deliberate and has a stated reason. None is a gap waiting on more effort: each is either a fact nobody publishes, or a number we could only produce by making one up.
The rule: a figure is published only when a named source states it. Where a source states something adjacent, the column is named after what the source actually said.
Megawatts, per facility
Not obtainable
datacenterHawk defines commissioned power as capacity "currently leased or are available to be leased". Both halves are facts about a lease, and there is no public registry of commercial leases.
| Source checked | What it returned |
|---|---|
| PG&E KnownLoad layer 19 | -99.99 on all 30,758 rows — the column exists and is empty |
| ERCOT large-load queue | 8 records, all aggregate. Zero customer names, zero counties |
| County building permits (11 ingested) | Publish square feet. Not one publishes power |
| Air permits (TCEQ, VA DEQ) | Backup diesel nameplate, not IT load. No bulk feed |
And not converted from floor area either. Hawk’s own stated reason for measuring in power is that the watts-per-square-foot ratio is unstable, so applying a constant would reproduce the number while destroying the reason it exists.
Lease rates
Out of scope and unobtainable
Hawk carries wholesale, retail and three hyperscale tiers per market-year. These come from broker and operator conversations, not from any record.
Vacancy rate
Depends on an unobtainable term
Available power over commissioned power. "Available" means available to lease, which is a leasing fact. Total floor area is public; how much of it is unlet is not.
Absorption
Depends on an unobtainable term
Hawk: "how much capacity was leased up by customers over a specific period of time." It is the change in leased power. Building counts and floor area moving between stages are published instead — that answers what is being built without pretending to answer what got leased.
Carrier-on-net fiber
The free sources answer a different question
Hawk lists the carriers present in a building’s meet-me room.
| Source checked | What it returned |
|---|---|
| FCC Broadband Data Collection | Reports availability at a location, not a carrier on-net in the building. Bulk download is account-gated, and no account was created |
| FiberLocator | The commercial vendor Hawk itself names. Route data is licensed |
Colocation vs hyperscale
Not stated by any source
Hawk splits these into two entities with different capacity vocabularies. No county record says which a building is, so the field is left null rather than guessed from the operator name.
Utility parent company and ticker
Not in the source
HIFLD names the operating utility only. "Virginia Electric & Power Co" is Dominion’s operating subsidiary, and the mapping to a listed parent is not published there, so parent is emitted as null rather than inferred.
Broker submarket names
Substituted, and labelled
Hawk’s submarkets are the names brokers use. Those are not published anywhere authoritative, so counties are used instead — the only sub-geography every publisher agrees on.
A second snapshot
Not yet, and not faked
The archive holds exactly one capture of each county layer, so the change feed is empty and will stay empty until the counties republish. No synthetic second snapshot has been fabricated to make the diff engine look exercised; it is tested against hand-built fixtures, and each of the nine change classes was verified to fail when removed from the engine.
What is included
grid/export/hawk/SCHEMA-DIFF.md. Shaped after Hawk OpenAPI v1.3.1, fetched 2026-09-11.