Tool — May 13, 2026
28-state parcel finder, scored against the Resistance Index.
Opus 4.7 tool-use over Cliff's 28-state corpus: HIFLD substations, candidate parcels, county-level resistance scores joined with a ridge-logistic 22-variable model (leave-one-out AUC 0.96). Filters: MW, voltage, distance, max-resistance, hyperscaler precedent, no-moratorium, industrial-zoned-only. 28 states scored; more roll out per-state.
- Parcels
- 294
- Substations
- 389
- Counties scored
- 347
Showing candidate parcels across 28 states. Refine by typing what you're looking for — MW, water access, gas distance, power, flood, zoning, fiber evidence gaps, resistance threshold.
Try a prompt
Example prompts: “220 MW within 5 mi of a 230 kV+ substation, no moratorium counties” · “500-acre+ parcels in PA along the Susquehanna nuclear corridor” · “Cheapest entitlement risk for 80 MW in NC western corridor”
Methodology
How the model is fit
The corpus is 236 data-center entitlement outcomes (225 decided, 11 still pending) between 2007 and May 2026, biased toward 2024–2026 because that is where the policy action sits. Outcomes are coded into five resistance categories (denied, withdrawn, voided, moratorium-blocked, delayed >12 mo) plus approved and pending. Resistance is the binary "not approved without delay" — 32% of decided outcomes.
Twenty-two variables are tagged at the filing date — moratorium status, residential density at 1 mi, organized opposition, NDA / shell pattern, prior denial within 50 mi / 36 mo, planning-board posture, hyperscaler precedent, PILOT, water cooling, groundwater stress, K-12 distance, battlefield proximity, state preemption, state-legislation-pending, among others. The full schema with citation patterns is at lib/resistance/variables.ts.
Coefficients are a multivariate L2-regularized (ridge) logistic regression, bootstrap-stabilized over 300 resamples. Ridge is what lets the model separate correlated predictors — moratorium and organized opposition co-occur — and carry an expanding variable set without overfitting the corpus. Ten additional AI-native variables (per-commissioner stance, procedural-challenge history, opposition-playbook detection) are defined and being backfilled across the corpus by the Site Read ingestion before they enter the fit.
Corpus snapshot
The 236 outcomes behind every coefficient
Sourced from primary news, county commission minutes, governor announcements, FOIA returns, and law-firm client alerts. Tagged at the filing date.
- Outcomes tagged
- 236
- Resistance rate
- 30%
- States covered
- 28
- Variables modeled
- 22
Top 5 predictors
The five variables that predict almost everything
Each row shows where resistance probability lands when only that one flag is true, measured against the model's flags-absent baseline. Three of the five push toward approval, not away from it.
- 1Active moratorium at filingCounty or municipality has a board-passed moratorium on data-center approvals in effect on the filing date.On its own, this flag moves a typical flags-absent filing from the 51% baseline up to about 81% likely to face structured resistance.
- 2PILOT / tax abatement offeredMaterial PILOT (payment in lieu of taxes) or sales/property tax abatement disclosed in the filing.On its own, this flag pulls a typical flags-absent filing from the 51% baseline down to about 24% — it pushes the filing toward approval.
- 3Pre-filing community engagementPublic community-meeting series held more than 60 days before formal vote.On its own, this flag pulls a typical flags-absent filing from the 51% baseline down to about 25% — it pushes the filing toward approval.
- 4Named opposition group presentA named coalition or Facebook group with >100 members has been organized in opposition before the vote.On its own, this flag moves a typical flags-absent filing from the 51% baseline up to about 68% likely to face structured resistance.
- 5Existing industrial zoningParcel is already zoned heavy/light industrial — no rezoning required, only special-use or site-plan review.On its own, this flag pulls a typical flags-absent filing from the 51% baseline down to about 38% — it pushes the filing toward approval.
Coefficients
Variables ranked by effect size
Each row shows roughly how much that one factor multiplies the odds of a project hitting resistance vs. clearing approval, holding everything else constant. Five factors do most of the work: active moratorium, an offered PILOT, pre-filing community engagement, organized opposition, and existing industrial zoning — three of the five push toward approval, not away from it.
County scorecard
347 counties scored across 28 states
Resistance scores 0–100 for 347 counties across 28 states. Counties with an active moratorium are flagged in red regardless of underlying density. The Charlotte–Triangle moratorium belt (Wake, Durham, Orange, Chatham) and the SC Lowcountry (Colleton) cluster at the top; the Catawba / Caldwell / Rutherford western NC corridor and the Berkeley / Dorchester / Aiken SC corridor cluster at the bottom.
Decision log
Recent outcomes in the corpus
Newest decisions first. Each outcome contributes its full variable bundle to the regression; the citation is the audit trail.
Cliff read
What this means for underwriting
For a PE firm screening NC / SC, the model says: skip the central-NC moratorium belt entirely until at least Q2 2027, even at apparent bargain prices — the moratorium coefficient is the largest single positive in the model, and the political flywheel that produced ten moratoriums in 18 months has not reversed direction.
Underwrite the western NC corridor (Catawba, Caldwell, Rutherford) and the SC Lowcountry industrial corridor (Berkeley, Dorchester, Aiken) at a discount to entitlement-risk premium — existing industrial zoning, hyperscaler precedent, and an established Duke / Santee Cooper / Dominion utility relationship together drop modeled resistance from ~50% to ~10%. The PILOT path remains the single most negotiable lever.
For deals that fall outside these corridors, the model gives a clear sequence of pre-filing moves that empirically lower resistance: pull the codename, run a documented 60-day community-engagement window, publish the applicant identity, and front-run state-legislation risk by structuring the PILOT before session start.
Primary sources
Every outcome traces back to a public record
- WRAL — Central NC data center handling tracker (Apr 2026)Source
- NC Health News — communities push back on AI data centersSource
- WUNC — opposition 'catching a fire' across NC politicsSource
- Fox Carolina — Project Spero withdrawn from SpartanburgSource
- SC Daily Gazette — Lowcountry gigawatt oppositionSource
- DCD — TigerDC pulled after SC tax-deal denialSource
- SC Governor — Google $3.3B SC announcement (Sep 2024)Source
- DCD — Google $9B SC expansionSource
- DCD — Microsoft 1,385-acre Person CountySource
- The Assembly NC — Person County Microsoft NDA patternSource
- Bean Kinney — Digital Gateway voiding lessonsSource
- Holland & Knight — Loudoun by-right eliminated (Mar 2025)Source
- Virginia Mercury — local governments race (Apr 2026)Source
- DCD — Gates County developer lawsuitSource
- Spectrum News — Stokes County zoning voidedSource
- Post and Courier — Colleton DC secrecySource
- HIFLD — Electric Substations public datasetSource
Resistance underwriting
Send your target geos. Get back the scored parcel set.
Send a state, county, or substation; Cliff returns a ranked parcel set with the resistance probability and the variables driving it. Outputs land in your inbox in a one-page brief that points back to the public corpus on this site.