Approved snapshot: 2026-11-03T23:19:26+00:00. Data does not update live. Research preview · Assessments are drafts unless explicitly marked human-approved. Coverage is selective.
Session counts include pro forma sessions unless noted. Congress also works through committees, oversight, constituent service and district visits. Floor counts alone cannot establish workload, attendance or waste.
EPSTEIN FILES
Follow the documents
Explore the names, check the sources and read Cited for Contempt’s take.
Federal housing statements, with sources and draft fact checks.
DATA & METHODS
Compare methods
Government statistics get revised. Choose a number to see what changed and how the calculation works.
This comparison is still being researched.
Scope & review
We compare published versions and reproduce the arithmetic where we have the inputs. We have not finished rebuilding the survey weights. Each comparison shows what we checked, its sources and its review status.
Editorial mission: Expose corruption, manipulation and suppression of public information through documented evidence.
Megawatts (MW) describe how much power a system can deliver at once. Terawatt-hours (TWh) measure electricity used over time. These data-center estimates include AI and other computing; the Virginia sales figure covers the whole commercial sector. Forecasts depend on their assumptions. These dates stay fixed when you change the homepage timeframe.
DOMINION ZONE · SUMMER 2025
Peak power demand
23,905 MW
Summer peak load in PJM’s Dominion zone was 23% higher than in 2019.
This was the highest demand at any one moment during the period. Utilities need enough capacity and reserves to meet those peaks. The total includes every type of customer in the zone. We cannot separate out data centers from this figure.
Checked September 27, 2026. EIA reports the peak using its Hourly Electric Grid Monitor and PJM’s load forecast report. The earlier Texas-and-Virginia capacity share was removed because its retired source table could not be reproduced.
ELECTRICITY USE · HISTORICAL ESTIMATE
Electricity use
176 TWh
Estimated U.S. data-center electricity use in 2023, about 4.4% of national consumption.
Berkeley Lab’s 2024 model includes equipment and supporting infrastructure.
Increase in Virginia’s annual commercial electricity sales between 2019 and 2025, equivalent to about 30 TWh.
EIA attributes much of the growth to data centers, alongside vehicle and building electrification. The figure covers all commercial sales. The 2025 data were preliminary in this release.
Annual U.S. data-center electricity use, TWh. Separate scenarios from the June 2026 report; each bar uses a 0–900 TWh scale.
Lower compounded bound521 TWh
Lower equipment installations578 TWh
Shorter AI-chip operating life590 TWh
Reference case649 TWh
More specialized graphics chips664 TWh
Higher AI-server idle power and utilization782 TWh
Upper compounded bound843 TWh
The low and high bounds combine several sources of uncertainty. Each of the four middle alternatives changes an assumption in the reference model. The report does not assign a probability to these outcomes. Source: report abstract, reference case and sensitivity scenarios.
Who gets the bill?
To connect this growth to your electric bill, we need to know what your utility charges, how much electricity you use and who pays for new grid investments.
Still to research
Household bills: price per kWh alongside household usage, matched by utility and time period.
Public incentives: tax breaks and infrastructure commitments, with the approving officials and documents.
Grid costs: who pays for new connections, transmission and unused capacity if a project falls through?
Local tradeoffs: water use, backup generation, permanent jobs and promised versus delivered tax revenue.
These profiles are still being researched. Nicknames are satire. Each entry identifies whether it concerns an allegation, a charge or a court outcome.
Loading leadership profiles…
More filters
Loading records…
Loading the evidence archive…
Congressional Docket
Follow the cases, the responses and what happened next, with links to the evidence. An empty profile means we have no published record here; it does not clear someone of controversy.
Open the Docket to load records.
Sources & coverage
We have started with selected federal housing statements. Editors choose what to examine, and this collection is still incomplete.
How we check claims
Is the claim accurate? Could the promise be carried out? Has anything actually happened? Each question needs its own evidence.
Sources
We save the attributed wording and surrounding context, along with where we found it, when we saved it and why we can retain it. A digital fingerprint lets us detect later changes to that saved text.
Assessment labels
Statement type describes what someone is saying. Evidence finding describes what the evidence establishes. We show separately whether someone has reviewed the claim, how recent its evidence is, whether a promise is feasible, and what progress has been made.
Rounded-square icons identify the statement or information type. Circular icons identify the evidence finding. Hover, focus or tap an icon for its full label and description; press Escape or tap elsewhere to close it.
Icon key
Type colors group kinds of information. Evidence findings use green for support, amber for qualifications, red for contradictions and gray for unassessed or undetermined findings. Shapes and tooltips carry the same meaning without color.
Statement types
Not classified
The statement has not been classified yet.
Factual claim
A statement that can be checked against evidence.
Estimate
An approximate value based on data and assumptions.
Prediction
A statement about a future outcome.
Causal claim
A statement that one thing caused another.
Claim about intent
A statement about what someone knew, wanted or intended.
Opinion
A preference or value judgment. Embedded factual claims need separate checks.
Speculation
A suggested explanation without an adequate evidentiary basis.
Promise
A commitment to a future action, with feasibility and progress assessed separately.
Evidence findings
Not assessed
No evidence finding has been recorded.
Supported
The cited evidence supports the claim as stated.
Partly supported
Some elements are supported; others are not.
Missing important context
An omission materially changes how a reader would understand the claim.
Contradicted by evidence
The available evidence conflicts with the claim.
Insufficient evidence
The assessed evidence does not establish whether the claim is accurate.
Correlation shown; cause not established
A relationship is documented, but causation has not been demonstrated.
Intent not established
The evidence does not establish what the person knew or intended.
Not applicable
A pure opinion has no factual conclusion to assess. Separate factual premises must still be checked.
Lies & fabrication
A lie is an established false claim, regardless of intention. Fabricated describes a documented invention, such as a forged document or manufactured event. Each finding requires evidence.
Our own information
Source-reported figures, calculations, estimates and projections are labeled by how they were produced. Labels apply to specific values or claims. We check source data separately from calculations and label satire.
Each assessment records evidence and uncertainty. Earlier verdicts remain in the revision history under their original labels.
Promises
For a promise, we ask who has the power to act, whose approval they need, what it would cost and what could block it. Draft plans list those requirements. To report progress, we need evidence that the steps actually happened.
Review & approval
Explanations stay drafts until an authenticated human reviewer approves them. Approval records who reviewed the version and when. Corrections preserve earlier versions. Training simulations never count as human approval.
Known limits
A press release tells us what an office says. We still have to check the claims. Missing evidence, out-of-date documents or transcription errors can change the answer.
Corrections
Send the owner the record title, what needs correcting and a supporting source. We record the reason for each correction and keep the earlier version. A public submission form is not available yet.
Add a source and statement
Paste text you are permitted to retain. This saves an unverified draft. It does not fetch the website or verify the statement. Choose the statement type if clear; its evidence finding starts as Not assessed. Split mixed claims before assessing them.