Research & Methodology

County Sentiment vs. Site-Level Diligence

Broad county sentiment gives useful background for a site review. It does not, on its own, provide a validated prediction of what will happen to a specific project.

Site-level diligence asks what is true at this candidate location, not just across the county. It records local facts, review steps and open questions, with their sources. Our prospective study will test whether these extra facts improve decisions beyond county context. That benefit is not an established result.

Evidence at the Coordinate Level

Sensitive Receptors

Shows whether nearby land uses make specialist diligence necessary before the project advances.

Emergency Response Capacity

Shows whether local response conditions warrant deeper operational review.

Approval Path

Clarifies who decides, which approvals matter, and where delays could emerge.

Evidence Quality

Makes clear what is supported, how current it is, and what remains unknown.

Model Validation & Evidence

Our Retro-Score study reconstructs coordinate-level signals for a retrospective, outcome-enriched case set. It is useful as descriptive and hypothesis-generating evidence, not as a prospective validation result.

We make no comparative performance or superiority claim against county context or other assessment frameworks. The preregistration below locks the design needed to support only those future claims that its independently documented results justify.

Public preregistration · ADCI-CVS-2026-01 · Version 1.1.0 · August 23, 2026

County context and coordinate-level evidence: prospective validation protocol

Status: preregistered; not enrolling; no validation result exists. This protocol is published before cohort enrollment and outcome review. Material changes will be issued as a dated, versioned amendment rather than overwriting this record.

Version 1.1.0 amendment, August 23, 2026—before enrollment: added deterministic enrollment dates, discovery cadence, and the requirement to lock and publish both score timestamps before any outcome event.

The prespecified question

Among consecutively enrolled, independently documented proposed U.S. data-center sites, does a frozen coordinate-level evidence score add discriminative value for a prespecified adverse siting outcome beyond a county-context score calculated from the same pre-outcome information set?

This is an observational validation question. It is not a claim of causation, approval probability, financial return, legal advice, or suitability for an individual project.

Cohort and exclusions

One proposed U.S. data-center campus per publicly documented coordinate or parcel centroid.
Enrollment window: September 1, 2026 at 00:00 UTC through August 31, 2027 at 23:59:59 UTC.
Every Monday at 17:00 UTC, reviewers archive every result since the prior run from the fixed U.S. Google News RSS query: “(data center OR datacenter) (proposed OR campus OR rezoning OR permit OR acquired).”
Every unique project is screened; the query, run time, result URLs, and disposition are published unchanged. Eligible projects cannot be skipped.
The first qualifying announcement, land transaction, permit filing, or government agenda record inside the window is the index event; enrollment occurs within seven days of discovery.
Exclude operating facilities, routine approved-campus expansions, duplicate phases, projects without a defensible location, and projects already adverse at enrollment.

Same sites, same information clock

County context: conditions shared broadly across the county or larger jurisdiction.
Coordinate-level evidence: project-specific conditions that can distinguish one candidate from another.
Both arms use the identical sites, index dates, outcome horizon, source-access window, direction, and missing-data rules.
Neither arm may use post-index hearings, litigation, withdrawals, later legislation, or later data revisions.
Both scores are produced in one run within 72 hours of enrollment and before any primary-outcome event. A pre-lock adverse event causes a documented exclusion.

Inputs and date lock

County context

Broad jurisdiction evidence available when the project enters the study.

Place evidence

Site-specific conditions and decision context available at the same point in time.

Audit fields

Each input identifies its source, date, geographic basis, and any quality or coverage limitation.

One immutable score snapshot per project will record scoring formula v6.0.0, weights, continuous score, tier, threshold status, full input manifest, and static-proxy/API-failure flags. Validation snapshots are frozen before any outcome event and never refreshed after lock. Before adjudicators receive the site, the public registry records each arm’s scored_at timestamp, input as-of ceiling, immutable manifest hash, and scoring version.

Outcome and independent evidence

The primary endpoint is a publicly documented cancellation, withdrawal, rejection, blocking, or legal/administrative halt within 24 months of the index date.

Official planning and zoning records, agendas/minutes, permit records, court orders, developer statements, and contemporaneous reporting will be archived with source dates. Two blinded reviewers classify outcomes; a documented third-review process resolves disagreements. Built/approved, active, delayed, contested, sued, and lost-to-follow-up remain separate statuses.

Analysis before results

The primary comparison uses continuous scores and paired discrimination/calibration estimates with bootstrap confidence intervals. Region, project scale, and stage are prespecified strata; no score weights will be fitted or re-fitted on this cohort.

A score of 50 or below is reported only as a pre-existing secondary product screen, with sensitivity, specificity, predictive values, and uncertainty—never selected after seeing outcomes. The target is 150 eligible sites; any insufficient event count and its resulting precision limits will be reported.

What will be published—and what this cannot answer

Public record

We will publish the de-identified enrollment registry, screening and exclusion log, outcome-source list, fixed-query search log, both scored_at timestamps, input as-of ceilings, immutable manifest hashes, protocol amendments, flow counts, missingness and fallback rates, score distributions, uncertainty, deviations, and null or negative results after outcome lock.

Known limitations

Public records can be incomplete, delayed, revised, or uneven across jurisdictions, and a parcel centroid may not be a final facility footprint. The study is observational and cannot establish that any signal caused an outcome or predict a particular project’s approval, financial return, or legal result.

Public protocol record

De-identified enrollment registry

Enrollment and score locking are reported here separately from future outcome follow-up. The registry withholds project names, coordinates, and score values while outcome review remains blinded.

The enrollment registry is temporarily unavailable. The preregistration above remains the controlling protocol record.

Ground Your Diligence

Move beyond generalized sentiment and test the site-specific conditions around a proposed coordinate.

Score a Site

Already have an account? Sign in