Reference
The primary documents were read in the original, and twice they overruled the essay.
The reference layer anchors Virginia's GS-5 rate order and the state watchdog study, the regional grid operator's auction results and its market monitor's counterfactuals, the seven-company pledge and its independent assessment, the nuclear and gas deals in their filings, two industry-funded cost studies, the Ohio tariff and Utah statute, the federal orders, and the polling and local record.
Mostly the sources bounded claims: contested models stayed contested, statuses stayed sorted (planned, operating, proposed), and the poll stayed scoped to what it measured. Twice they did more. The monitor's own report decomposed what an early draft had merged into a single dollar range. And two industry-funded studies falsified an early "nobody publishes this" claim — so the essay narrowed the claim and adopted the studies as the companies' strongest evidence.
What the references did
Changed
- Decomposed the headline dollar figures: $9.3 billion is one auction year's estimate; $23.1 billion is three years cumulative ($9.3, then $7.3, then $6.5 billion); the two are never a range.
- Corrected the auction shortfall to the monitor's revised figure and re-labeled it an auction shortfall against a planning standard.
- Falsified the universal receipt-absence claim (two industry-funded studies exist) and forced the narrowed form: no recurring, regulator-required, after-the-fact public reconciliation found.
- Corrected the essay's own note on the Virginia projections: $14 and $33–37 a month span two demand scenarios, not one.
Narrowed
- The Gallup 71% stayed triple-scoped: siting opinion, not cost opinion; national opinion, not local consent; local construction, not data centers everywhere.
- The Entergy savings figure stayed labeled as the utility's own claim in a pending filing.
- The pledge's "vague and legally toothless" verdict stayed attributed to its single independent assessment.
- The Ohio tariff stayed "in force unless struck" — a negotiated settlement on appeal, not an adjudicated finding.
Not imported
- No claim that households are currently subsidizing data centers — the essay's subject is the missing receipt, not a proven transfer.
- No claim that the pushback is winning or the buildout is slowing.
- No motive attribution to companies, regulators, or opponents.
- No claim that process transparency proves the essay.
Source ledger
Open any entry to see the outside constraint and what it could have changed.
RL-019-01: Virginia's GS-5 rate order, read in the original.
Why it matteredThe essay's "rung one" example — an adopted, enforceable rule — rests on order-level terms: eligibility thresholds, minimum demand charges, contract length, collateral.
Public sourcesSCC Final Order, Case PUR-2025-00058 (Nov. 25, 2025); SCC data-center fact sheet; notice of appeal.
ScopeSupports the 25-megawatt/75-percent thresholds, the 85/85/60 minimum charges on contracted demand, fourteen-year agreements, collateral, and the January 2027 effective date — every term verified against the order itself, with zero prose change needed. The order is under appeal; transmission- and generation-allocation questions remain open in later proceedings, and the essay says so.
RL-019-02: The Virginia watchdog study and its projections table.
Why it matteredThe essay's pairing discipline — fair under the tested rules, exposed in the modeled future — quotes both halves of one study.
Public sourcesJLARC Report 598, "Data Centers in Virginia" (Dec. 2024) — summary, chapter 4 and Table 4-2, and Appendix L for the approval-chain record.
ScopeSupports "paying their full cost of service" under 2024 rates, the $14 (half-forecast) and $33–37 (full-forecast) monthly generation-and-transmission projections for 2040 in constant dollars, and the by-right versus discretionary approval record in the essay's worked-example counties. A direct read of Table 4-2 corrected the essay's own earlier note about which scenarios the figures span.
RL-019-03: The auction results and the market monitor's counterfactuals.
Why it matteredThe shortage and every capacity-market dollar figure in the essay come from these documents — and their model status had to stay visible.
Public sourcesPJM 2027/28 auction results; Market monitor, Analysis of the 2027/2028 Base Residual Auction, Part A (Jan. 5, 2026); monitor's FERC comments on the capped auctions.
ScopeSupports the ~6,500-megawatt shortfall against the reliability requirement (the monitor's corrected figure), the $9.3 billion single-year and $23.1 billion three-year counterfactual estimates with their year-by-year decomposition, and the ~$13.8 billion capped-auction estimate nested inside the total. All are counterfactual models; PJM disputes the attribution share; the essay carries the dispute.
RL-019-04: The seven-company pledge and its independent assessment.
Why it matteredThe essay's opening object — a promise standing where a rule should be — is this pledge, and its enforceability verdict had to be attributed, not asserted.
Public sourcesWhite House Ratepayer Protection Pledge (Mar. 2026); Carnegie Endowment, "Beyond the Hype" (June 2026).
ScopeSupports the seven signatories, the March 2026 date, the voluntary/no-consequence character, the attributed "politically significant but vague and legally toothless" assessment, and the rough scale check that announced nuclear projects would cover less than one-fifth of projected 2035 data-center use. Single-assessment claims stay attributed.
RL-019-05: The deals, sorted by status in their own filings.
Why it matteredThe essay's honesty rule for commitments — planned, operating, proposed: all real, none interchangeable — required each deal's status from its own instrument.
Public sourcesDOE loan closing for the Pennsylvania nuclear restart; Talen–Amazon agreement (SEC filing); Entergy–Meta announcement; Louisiana PSC docket 32728.
ScopeSupports the planned 835-megawatt restart with its finalized $1 billion federal loan, the up-to-1,920-megawatt agreement on an operating plant through 2042, and the proposed Louisiana package in a pending regulatory filing with the ~$2 billion savings figure labeled as the utility's own claim. Statuses were re-checked at publication.
RL-019-06: Two industry-funded cost studies — the falsifiers adopted as evidence.
Why it matteredThese are the documents that killed the essay's early universal absence claim — and became its strongest pro-company evidence and half of its receipt argument.
Public sourcesE3, "Tailored for Scale" (Dec. 2025; Amazon-funded, funder-reviewed); E3, "Beyond the Headlines" (June 2026; developer-funded, funder-reviewed).
ScopeSupports the real-sites payments-exceed-costs projection, and the hypothetical 100-megawatt customer's $37 million base-case net benefit with its +$166 million to −$92 million sensitivity range. Funding and pre-publication review are disclosed in the studies' own front matter and quoted in the essay. Neither is an audit; both are forecasts, not the after-the-fact reconciliation the essay says is missing.
RL-019-07: The Ohio tariff and the Utah statute.
Why it matteredThe essay's claim that enforceable rules exist beyond Virginia — and its sequencing-clearance prior art — needed instrument-level support.
Public sourcesAEP Ohio data-center tariff; Utah Code § 54-26-302.
ScopeSupports the tiered minimum-payment formula rising toward 85 percent of reserved capacity in one utility's territory, the regulator's anti-cost-shifting rationale, and Utah's requirement of commission approval with express findings that the customer bears its incremental costs and existing ratepayers bear none. The Ohio order is a negotiated settlement under appeal; the essay's wording is "exists and binds unless struck."
RL-019-08: The federal actions and the pending proposals.
Why it matteredThe essay's "direction of travel" claim — make them pay more of their own cost, and keep building — rests on what regulators actually ordered versus what is merely proposed.
Public sourcesFERC's June 2026 show-cause orders to the six regional grid operators; PJM reliability-backstop proposal (June 30, 2026); Va. Code § 15.2-2209.5.
ScopeSupports the justify-or-reform orders (faster studies, clearer cost reporting, anti-stranding agreements), the backstop's proposed-not-adopted status, and Virginia's 2026 limited grid-information-sharing law for certain large projects. Proposals are labeled proposals throughout.
RL-019-09: The polling, the host county, and the federal framing.
Why it matteredThe essay's democratic-gap argument uses a national poll, a county's tax record, and the government's own strategy documents — each easy to over-read.
Public sourcesGallup: 71% oppose AI data centers in their local area; Loudoun County data-center FAQ; DOE, "Powering America's AI Future".
ScopeSupports the triple-scoped poll reading (siting opinion, not cost opinion; not proof any community withheld consent; not opposition everywhere), the major-share property-tax finding with the elected chair's constituent quote, and the economic-and-national-security framing as the government's own. None of the three settles the payer question, and the essay says so.
RL-019-10: The feasibility floor and the counter-narrative checks.
Why it matteredThe receipt proposal had to be checked against rate-making practice, and the essay's no-collapse and regional-cost claims against their strongest available sources.
Public sourcesLBNL, rate designs for large loads (Jan. 2025); SemiAnalysis on the cancellation narrative.
ScopeSupports "the pieces exist in rate cases, cost studies, and bill line items; the hard part is boundaries, then publication" and the attributed no-nationwide-collapse claim. The LBNL brief's no-industry-standard finding backs the standardization difficulty without proving the receipt impossible — one of the essay's published failure conditions covers exactly that.
What the sources do not do
Sources bound the claims; they do not prove the judgments. The essay's landing — a rule, a receipt, and a say before the public is asked to pay — is an argument, not a citation. The essay's absence claims are limited to what its disclosed searches covered, and its close publishes the findings that would prove it wrong.