Retro-Score Report: Data Center Sites and Consent-Path Evidence
This audit compares projects with known public review outcomes against operating sites. Matching rules were set in advance for state, date and planned scale. Three exploratory pairs show a +1.0 mean Site Viability difference. Crooked Creek’s 200 MW input was supplied by the user and has not been independently confirmed. The old is_optimal source-flag group is kept only to trace the study’s history.
frozen legacy is_optimal source-flag rows retained for study lineage only. This is not the current qualification set or active published inventory.
state-aligned operational controls with cited vintage and MW evidence. The roster was fixed before scoring.
eligible plan-scale matched pairs. The Crooked Creek 200 MW value is a user-directed exploratory assumption, not utility-confirmed or independently corroborated.
Scores use today's model (site-viability formula v6.0.0, higher score = better site) at each project's day-one location, with land-acquisition / announcement date context. This outcome-enriched retrospective roster is a small descriptive current-model pilot and hypothesis-generating only. The numeric viability screen is reported separately. The comparative study retains the frozen 1,548 rows as a legacy is_optimal source-flag cohort, never as current receptor-gate survivors or active published inventory. Project names, model findings, analyst context, and citations are in the report.
Inside the report
Outcome-selected case studies
Eight projects selected for documented adverse consent-path outcomes. The roster is descriptive, not representative of the market or a failure-rate estimate.
Severity, not misleading direction
Site Viability remains higher-is-better. Every component is labeled as exposure severity so a high grid score cannot be mistaken for a good grid.
What the audit supports
Three exploratory plan-scale pairs produce a +1.0 mean paired Site Viability difference. This is descriptive, not causal or an operational-load estimate.
The named projects
The 2,100-acre Virginia corridor voided in court. The Texas campus behind a $100M federal suit. The desert campus a city council killed unanimously. And five more.
Evidence status
Retrospective evidence is not a head-to-head result.
The report documents what the current model surfaces when applied back to known cases. The predeclared matcher publishes a three-pair exploratory plan-scale estimate using an explicitly unverified, user-directed 200 MW Crooked Creek input. Community Readiness and every other dimension remain cohort-conditional descriptive context, not causal or predictive evidence. This pilot does not establish that coordinate-level evidence outperforms broader-area context or another framework. Those claims require the preregistered, independently documented prospective comparison now published in our methodology.
Read the prospective validation protocol →Download the separate historical-vintage PDF →What the Retro-Score Report examines
The Retro-Score Report looks at sites with a known public review history. That history may include approval, a dispute, a withdrawal or an ongoing review. We screen each location and compare selected projects with operating sites. The aim is to ask better early questions, not to claim the model predicted what happened.
We chose cases because of their known outcomes and public records. They are not a random sample of all proposed data centers. They can teach us about site constraints. They cannot tell us the national chance that a new project will fail, proceed or face a lawsuit.
How the matching audit works
The frozen audit follows rules set before matching. Each pair must match the state, project vintage and project scale. Operating sites form the comparison group. A pair counts only when the rules and known inputs support it. We disclose unmatched cases and missing facts; we do not fill them with convenient substitutes.
The page shows three exploratory plan-scale pairs and their mean paired Site Viability difference. That is a small descriptive study, not a tested forecast. Check the sample size, how well the sites match and the input limits before using the result for a new site.
Why source quality and scale matter
An announced capacity, a planned capacity and an operating load are not the same thing. A plan does not prove a site's current load. Crooked Creek's 200 MW input is user-directed and not independently corroborated. This matters because project scale is used to match sites.
We keep the frozen legacy source-flag cohort so readers can trace the study's history. It is not the current qualification set or the active published site inventory. Keeping it fixed prevents changes to today's database from silently changing the study denominator.
A current-model screen is not a historical forecast
Scoring an old site with today's model does not show what an analyst knew then. A historical claim needs dated sources, a set time window and a method that can be rerun for that window. Read the separate historical-vintage study on its own terms. Do not treat this current-model comparison as that study.
A score does not prove why a project was approved, denied, disputed or withdrawn. Power, water, public services, local review and community facts can guide further work. They do not explain the project's history on their own or replace the decisions made by the bodies that reviewed it.
How to use the findings in site diligence
Keep three things separate: a checked fact, a modeled signal and an open question. Use the public records to choose the right next reviewer: the utility, an engineer, a lawyer, a local official or the outreach team. A nearby power line does not prove spare capacity. A readiness signal does not prove local support.
For a new project, screen the actual address or coordinates with current sources. Compare the issues, not just the scores. Do not assign an old case's outcome to new land. Local review, design, timing and the authority of each source can change what a finding means.
Sources, scope, and further reading
Use the frozen audit notes and downloads to check the study, each project and its public sources. Keep the descriptive matching audit separate from the historical-vintage report. Read each report's denominator, inputs and limits before quoting a result.
This is a screen to support professional judgment. It is not engineering feasibility, legal advice, a permit decision or a promise of approval. Our research and diligence pages explain the method and how a current site screen fits alongside expert review.
Explore the supporting research · Understand data center site diligence
The names, the scores, the receipts.
Read the seven-page report. It explains how cases were chosen and where the method falls short. Review the current-model score table, each project and its public sources.