Methodology

Every number cites a source; claims that cannot be verified are refused and recorded.

How a claim becomes a booked fact, why we hold so many claims out, what the depth grades mean, and what the prediction ledger will grade in public.

Source provenance on every fact

Nameplate Data aggregates public filings, utility and county records, regulatory testimony, corporate disclosures, and trade press. Each published dollar and megawatt carries the source document behind it, along with that source's review status. To see what a figure rests on, you open the source.

Sources are reviewed individually before the facts they support are booked. The review status stays visible in the workbench, so a figure resting on an unreviewed source reads differently from one resting on a checked filing.

Refusals are a first-class output

A dollar or a megawatt gets a source document, a scope check, and a confidence grade, or it does not get booked. When a claim can't be verified against a project-specific source, we decline to count it and record the refusal, with a written reason and the list of sources we checked.

The common failure modes we hold out for:

  • Program-level totals announced across multiple sites, with no public split — booked to a site only when a site-level figure exists.
  • Secondary estimates that trace to trade press rather than to the company's own announcement or a filing.
  • Scope substitution — a contracted PPA capacity reported as site load, a multi-phase total reported as phase one, an interconnection request attached to the wrong campus.
A refused claim is kept and shown, not quietly dropped. Megawatt claims fail more often than cost claims — scope substitution is the usual reason.

Evidence depth tiers A–D

Every project carries a depth grade on the same published criteria: does it have a verified MW figure, a sourced cost estimate, power evidence (a named utility or interconnect filing), and an approved primary source. The grade is a description of the evidence, not a verdict on the project.

TierWhat it means
AVerified MW, a sourced cost estimate, power evidence, and an approved primary source — the fullest public evidence a project can carry.
BMost of the evidence spine present, with one or more named gaps.
CPartial evidence; several load-bearing facts still unsourced or unverified.
DThin public evidence — an announcement with little behind it yet.

Only a handful of projects reach tier A — that is the state of public evidence in this market, not a defect in the sample. The grade tells you how much weight the rest of a project's numbers can bear.

Scores are model indicators, not probabilities

Scores and verdicts are model-derived indicators on a 0–100 scale. They are not calibrated probabilities, forecasts, or investment advice, and they should not be the sole basis for any investment or financing decision. They summarize the evidence; they do not predict an outcome.

Nameplate Data is a publisher of research and data. It is not a registered investment adviser and does not provide individualized investment advice. When we say a project's power reality looks weak, that is an analytical opinion grounded in named evidence you can open — not a claim about what will happen.

The prediction ledger

Every month, every project's status and risk scores freeze into an append-only table the model cannot rewrite. When rumored and announced projects resolve — energized, cancelled, or quietly shelved — the earlier calls can be graded against what actually happened.

The first model report card lands in 2026-Q4 and grades those frozen calls against outcomes. Until then the scores are published as indicators, not calibrated probabilities — the report card is where calibration gets demonstrated.

What we will and won't claim about queues

Per-project load interconnection-queue data is confidential in every ISO we have scoped. The public named queues are generation only; the public load-queue reports are aggregated and anonymized by design. So the honest promise is aggregate queue-reality context per ISO and per-project corroboration evidence — land use, rate cases, co-location signals — never a claim of a specific project's load queue position.

An announced state figure and an ISO's tracked large-load request pile describe different things: the request pile is anonymous, duplicative, and unbuildable in aggregate; the announced-and-sourced figure is what survives name-and-source discipline. Neither tells you which specific projects energize; matching a named announcement to project-specific corroboration is the work the database does.

See it in the workbench

Read the criteria, then check a project against them.

Open the workbench and walk any project's sources yourself, or start with the weekly notes.