Study #04 · AI & Public Policy Observatory · v1.2

Where the Cloud Hits the Ground

In Washington, AI dockets focus 3.5× more on national security and global competition alongside federal grid and environmental rules. In American towns and states, the debate centers on tax breaks (4.0× higher), local power bills, and water—where communities mostly bargain or pause (45%–52%), while blocking projects outright in nearly 1 in 3 high-heat fights (30.3%–36.0%).

Sister-Lab Audit ★ SG v1.2 Audit Claude Opus 5.5
Version v1.2 Nov 6, 2026 (3 Ed.)
Dataset N = 364 / 243 48 / 44 Jurisdictions
Live AI Judges ICC = 0.9361 Live 2-Judge PPI
Interactive US Geo-Map 21 Hubs + 364 Rows

What’s New in Version 1.2 Nov 6, 2026

Following Singapore Sister-Lab Audit SG-AUDIT-S04-V1.1-2026-11-06 (EXT-AUDIT-025 → EXT-AUDIT-026), v1.2 drops all 824 rule-fallback ballots to report live-ballot-only ICCs, derives policy posture from live median/mode (7.42%–8.79% direct denials overall; 30.34%–35.96% in high-pressure dockets), rebuilds the 2023–2026 Federal Register sample (n=51, revising P1 to null, p=0.4401), cleans and deduplicates the on-topic panel (N=243, G=44, where P2 = +48.1730, p=0.0001), and archives v1.1 PDF and v1.0 PDF unchanged.

Data Nutrition Label & Provenance CC-BY 4.0

Panels: Baseline N=364, G=48 (ai_infrastructure_observatory_panel.csv) & Cleaned On-Topic v1.2 N=243, G=44 (192 Local/State + 51 Rebuilt 2023–2026 Federal).
Replication Bundle: GET /api/agent-replication?study_id=study-04
Visual 1 · Interactive U.S. Geomap (48 Jurisdictions · 364 Records)

Where AI Data Centers Are Sparking Local Deals, Pauses, and Grid Fights

PJM DATA CENTER ALLEY MISO / PJM MIDWEST SERC SOUTHEAST ERCOT TEXAS DESERT SOUTHWEST PACIFIC NW WA OR CA AZ CO TX WI IL OH TN GA NC VA MD/DC PA NJ MA
Pin Color: Tighten Rules / Cap Hit Pause / Moratorium "Yes, If…" Bargain Green Light / Fast-Track Click any hub pin to inspect & filter →
Northern Virginia · PJM Grid Tighten Rules / Cap

Prince William & Loudoun, VA

Local Heat Score
45.0 / 100
Tracked Records
25 filings
🏘️ Top Fight: Zoning Next to Homes & 500-kV Power Lines

Northern Virginia holds the world's largest cluster of data centers. Loudoun County ended automatic "by-right" zoning (AII-2025-010), requiring public hearings and noise limits. Next door, a judge voided Prince William County’s 2,100-acre Digital Gateway rezoning (AII-2026-201).

Key Docket: AII-2025-010 · AII-2026-201
Visual 2 · Live-Ballot Policy Posture (v1.2 Live Median & Mode, N = 364 / N = 243)

Conditional Bargaining & Pauses Lead (45%–52%), While Direct Denials Reach 30%–36% in High-Heat Fights

Live Median across N=364 (Cleaned v1.2 N=243: 28.0% / 31.7% / 20.6% / 12.8% / 7.0%) ↓
38.7% Green Light
28.3% "Yes, If…"
17.0% Pause
8.5% Cap
7.4% Ban
38.7% (141 / 364)
Green Light (Code 1)
Live Mode: 39.6% (144) · Cleaned v1.2: 28.0% (68/243)
28.3% (103 / 364)
"Yes, If…" Bargain (Code 2)
Live Mode: 27.5% (100) · Cleaned v1.2: 31.7% (77/243)
17.0% (62 / 364)
Hit Pause / Study (Code 3)
Live Mode: 17.6% (64) · Cleaned v1.2: 20.6% (50/243)
8.5% (31 / 364)
Tighten Rules (Code 4)
Live Mode: 6.6% (24) · Cleaned v1.2: 12.8% (31/243)
7.4%–8.8% (27–32)
Direct Denial (Code 5)
30.3%–36.0% (27–32 / 89) in high-heat dockets (live_ppi ≥ 50)
Visual 3 · Two Different Conversations

Towns Talk Power Bills. D.C. Talks Superpowers.

How often each issue comes up in 301 local & state records vs. 63 federal records (scored 0–100% by 4 open-weight AI judges):

⚡ Electric Grid & Household Power Bills 2.3× higher locally
Local & State
28.6%
Washington D.C.
12.6%
💰 Tax Breaks & Local Budgets 2.7× higher locally
Local & State
14.2%
Washington D.C.
5.2%
💧 Water, Farmland & Generator Noise High at both levels
Local & State
19.1%
Washington D.C.
27.4%
🛡️ National Security & Global AI Race 4.3× higher in D.C.
Local & State
6.8%
Washington D.C.
29.0%
Three Real Playbooks From the Records

What Local Pushback Actually Looks Like

1. "Pay for Your Own Power Lines" (Ohio) Tariff Deal

Ohio’s utility commission did not ban data centers. Instead, it ordered a special tariff (AII-2025-027). Big tech campuses must guarantee minimum power payments for years, so families never get stuck paying for new grid lines.

2. "No More Automatic Permits Next to Homes" (Virginia) Zoning Cap

Loudoun County is the world capital of data centers. In 2025, it ended automatic "by-right" approvals (AII-2025-010) and required public votes and noise rules. Next door in Prince William County, a judge voided a massive rezoning near Manassas Battlefield (AII-2026-201).

3. "Hit Pause & Check the Receipts" (Cities & States) Pause / Audit

Georgia lawmakers voted to pause state sales-tax giveaways (AII-2024-002). Meanwhile, city councils in Oakland, Seattle, Baltimore, Nashville, Denver, and Chandler, AZ proposed temporary pauses or megawatt caps to check water and grid capacity first.

Visual 4 · Public Pressure Meter (0–100 Scale)

Where AI Companies Face the Most Local Heat

Click any community card to inspect its official records below ↓
IN / MI / WI / OH
Midwest Counties
48.7 / 100
Water & farmland fights
Northern Virginia
Prince William, VA
45.0 / 100
Digital Gateway lawsuit
State Utility Docket
Ohio PUCO Grid
41.9 / 100
Big-tech power tariff
Arizona Desert Grid
Chandler, AZ
39.0 / 100
Unanimous council rejection
Georgia Corridor
Metro Atlanta, GA
36.8 / 100
BeltLine ban & water rules
Mid-Atlantic / PJM
Baltimore, MD
36.6 / 100
>5 MW moratorium bill
Northern Virginia
Fairfax & Loudoun
32.6 / 100
Ended by-right zoning
Bay Area, CA
Oakland & San Jose
32.2 / 100
Moratorium & grid deals
Pacific Northwest
Seattle, WA
30.5 / 100
Council Bill 121214 hearing
Desert Southwest
Phoenix, AZ
26.4 / 100
Substation & water rules

Explore All 364 Public Records

Every row links directly to the original city council file, state utility docket, or federal notice.

Date Community / Level Official Record (Click to Open Source) What They Did Main Issue Heat Score
Visual 5 · Data Quality & Completeness Check

Did We Miss Anything? How We Checked 30 City Halls

City government websites break or change software all the time. Before analyzing the data, our agents checked the official inventories of 30 U.S. city council APIs so we wouldn't mistake a broken website for "zero protests":

24 Cities
🟢 Live & Up to Date
Direct city clerk feeds active through 2026 (Phoenix, Columbus, Seattle, Denver, Baltimore…)
4 Cities
🟡 Switched Software
Goodyear, Long Beach, Greensboro & San Antonio froze old Legistar feeds in 2021–2023
2 Cities
🔵 Recovered via Backup
Chicago & Atlanta city APIs returned error 500; recovered via county & state dockets
9 Typos
🧹 Clerk Typos Caught
Filtered city clerk date typos like year 9859 or 2030 before analysis
See the Full 30-City Completeness Checklist (data_completeness_ledger.json) ▾
City API Community Status Latest File Date Clerk Typos 2023–2026 Matches Included Rows How We Handled It

Under the Hood: Statistical Tests, 4-Judge AI Scores & Academic PDFs

Want the full regression tables, pre-registration lock (commit 08ef92e), and AI referee report? Everything is open below.

📊 Statistical Summary (v1.2), Live-Ballot Reliability & Singapore Audits (EXT-AUDIT-023..026) ▾
Test P1 · Revised to Null in v1.2
+0.0502 (p = 0.4401)
Baseline P1 (+0.0769, p = 0.0268) has wild-cluster bootstrap p_wcb = 0.1740 and is -0.0023 (p = 0.9209) vs. EPA/FERC energy buckets. On the cleaned v1.2 panel (N = 243, G = 44) with 51 rebuilt 2023–2026 Federal Register dockets, overall resource salience is null (t = +0.7721, p_wcb = 0.6264); instead, federal dockets focus 3.50x more on national security (0.5067 vs. 0.1449, p < 0.0001) while local/state dockets focus 3.99x more on tax breaks (0.2449 vs. 0.0614, p < 0.0001).
Test P2 · Resource Strain → Heat (Supported)
+48.17 pts (p = 0.0001)
On the cleaned v1.2 on-topic panel (N = 243, G = 44), physical resource salience strongly predicts Public Pressure (SE = 12.5497, t = +3.8386, p_wcb = 0.0028; primary gov records N = 183: +54.0006, p = 0.0001). On baseline N = 364, all-4-slot P2 is +26.2713 (full-rank SE = 9.8013, t = +2.6804, p = 0.0074, p_wcb = 0.0282) and live-ballot-only P2 is +55.0318 (t = +5.1118, p < 0.0001, p_wcb = 0.0004).
Test P3 · 2025–2026 vs. 2023–2024 (Marginal / Null)
-3.16 pts (p = 0.3741)
On baseline N = 364, controlled OLS is marginal (+4.8119, full-rank SE = 2.5359, t = +1.8975, p = 0.0578, p_wcb = 0.1022; live-ballot +8.6461, p = 0.0606). Once records are dated by event and off-topic rows are removed on v1.2 (N = 243), period escalation is null (-3.1579, SE = 3.5529, t = -0.8888, p_wcb = 0.4426).
📄 1. Full Academic Working Paper PDF (Edition v1.2 — N = 364 Baseline & N = 243 Cleaned On-Topic Panel) ▾
🔒 2. Study Design Card & Endpoint-Probe Provenance Disclosure PDF (STUDY_DESIGN_CARD.md · v1.2) ▾
🛡️ 3. Independent AI Referee & Sister-Lab Verification Report PDF (INDEPENDENT_PEER_REVIEW_REPORT.md · v1.2) ▾
🌏 4. Singapore Audit SG-AUDIT-S04-V1.1-2026-11-06 & Boston v1.2 Author Response PDF (EXT-AUDIT-025 → EXT-AUDIT-026) ▾
🌏 5. Earlier Singapore Audit #12 & v1.1 Author Response PDF (EXT-AUDIT-023 → EXT-AUDIT-024) ▾
🗄️ 6. Archived Edition v1.1 PDF (PAPER_MANUSCRIPT_V1_1_ARCHIVED.md · SHA-256 d092faf1fe05...) ▾
🗄️ 7. Archived Initial Edition v1.0 PDF (PAPER_MANUSCRIPT_V1_0_ARCHIVED.md · SHA-256 7ce2c0e57813...) ▾