Horizon Accord | Mason County KY | Data Center | Accountability | Machine Learning
Accountability Patterns
The Anonymous Applicant
A replicable methodology for identifying anonymous corporate actors in public processes — demonstrated on a live, unresolved case
The Methodology
Pattern analysis is not speculation dressed in confidence. It is a discipline with rules. This piece states those rules explicitly — not because they are novel, but because analysis that calls itself pattern-based frequently violates them without acknowledgment, producing conclusions that feel rigorous while being decorative.
The rules are four.
Rule 1 Every claim must be classified before it is used. A Documented Fact is sourced and verifiable by any third party from public records. A Structural Observation is a pattern identified from two or more documented facts. A Hypothesis is a reasoned conclusion from structural observations — it can be strong, but it cannot be stated as fact. Claims cannot be upgraded silently. If a claim sounds right but lacks sourcing, it stays a hypothesis or it gets cut.
Rule 2 The fingerprint must be built from confirmed prior cases before the live case is examined. You cannot construct a signature from the evidence you are trying to match. If an element appears in the live case but cannot be traced to a confirmed prior deployment of the same actor, it is not a fingerprint element. It is a data point whose significance is unknown.
Rule 3 Elimination is as important as identification. A candidate that fails on one high-weight factor is eliminated regardless of how many lower-weight factors it satisfies. The analysis must state what would disconfirm its conclusion, not only what confirms it. And disconfirmation evidence must be reported honestly, including when it dismantles a previously held position.
Rule 4 The confidence level of the output must be derived from the evidence, not from the analyst's conviction. Pattern analysis produces a probability-weighted hypothesis — not a name on a document. Overstating confidence is a methodological failure, not a rhetorical choice.
This piece applies that methodology to a live, unresolved case: a Fortune 20 company that has submitted a rezoning application for a 2,080-acre hyperscale data center campus in Mason County, Kentucky, and has not been publicly named at any stage of the process. The April 22, 2026 planning commission meeting is days from this writing. The case is open.
The Instruments of Concealment
This is a prerequisite step, not optional context. Documenting the concealment architecture before building the fingerprint matters because the instruments used to hide an actor's identity are themselves pattern elements. They shape what counts as a fingerprint element and why. A methodology that skips this step will miss the structural signature entirely.
The anonymity in this case is not accidental. It is constructed from instruments with documented histories.
Non-disclosure agreements originated in trade secret law — protecting genuinely proprietary technical information shared between parties in a competitive context. What the NDA became through decades of expansion is something different. It has been used to silence employees who witnessed workplace misconduct. To suppress knowledge of defective products. To conceal corporate actors from the communities affected by their decisions. And to bind elected officials to confidentiality provisions that prevent their constituents from knowing what is being decided on their behalf.
In every documented expansion of NDA use beyond its original technical purpose, the protection runs to institutional power. There is no documented case in which an NDA has been deployed to protect a tenant from a developer, a worker from a corporation, or a community from an anonymous buyer. The instrument does not work in that direction. Enforcement requires legal resources. Common people are on the receiving end — signing them as a condition of employment, settlement, or in this case, as elected officials bound by penalty of law.
Shell companies and intermediary buffer structures have a similarly narrow legitimate use case and a much wider documented history of misuse. The one protective application of LLC anonymity structures for common people is this: individuals fleeing domestic violence occasionally use them to obscure a home address from an abuser. This is not a redeeming feature of the instrument. It is evidence of a failure in every other protective system those individuals should have been able to rely on. And it is a use case available only to people with legal resources, financial literacy, and access to attorneys. The residents of Meadowland Village did not have those things. They received a letter from a property management company. It told them they had ninety days to leave. They had no mechanism to know who was buying their community. No leverage to negotiate. No instrument in the entire legal architecture of this process was designed with their interests in mind.
Structural Observation The instruments deployed in Mason County — the NDA binding local officials, the Industrial Development Authority as formal applicant buffer, the absent end user — are not neutral administrative choices. Their documented history of use at scale runs in one direction. That they appear together, in sequence, in a public zoning process affecting 28 households is not incidental. It is the pattern expressing itself. The methodology exists to read that expression.
Building the Fingerprint
Rule 2 applied A fingerprint is only valid if built from confirmed prior cases — not from the case under investigation. The elements below are drawn exclusively from publicly confirmed and named Amazon Web Services deployments. Each source is listed in the verification section.
Across confirmed AWS deployments in Mississippi, Indiana, North Carolina, and Wilmington, Ohio, the following structural elements recur with documented consistency. Each element is assessed for discriminating power — that is, its ability to distinguish AWS from other large hyperscale operators, not merely its presence in the record.
Element 1 — High discriminating weight The structural concealment sequence: NDA binding local officials through the entire public process, a local development authority or port authority as formal applicant buffer, and a utility or electric cooperative as the first point of contact between the company and local government — not the company itself. These three elements appear together in confirmed AWS deployments. They do not appear together in the one confirmed Meta deployment available for comparison: Meta's Howell Township project used an NDA and a shell LLC, but did not use a utility as intermediary and did not use an IDA as formal applicant. The combination of all three is consistent with the confirmed AWS pattern. It was not observed in the one confirmed Meta case in this dataset. This makes it a promising discriminator. The comparative sample is small, so it is not yet a proven one.
Element 2 — High discriminating weight Single-site power demand in the 2.0–2.5 gigawatt range. The confirmed AWS Indiana campus required 2.4 GW. The Mason County EKPC filing requests 2.2 GW — among the largest single-customer power requests in the history of the regional PJM grid. Google's confirmed Scioto County project is described as 500,000 square feet and does not approach this power scale. Meta's Howell Township project did not involve power demands at this order of magnitude. This element places the Mason County project in a very narrow class of confirmed precedents. The confirmed precedents at this power scale in this dataset are AWS. No other hyperscale operator has been publicly confirmed at comparable single-site demand in the cases reviewed here — which is not the same as saying no other operator could reach this scale.
Element 3 — Medium discriminating weight Campus design specifying not only data center buildings but administrative offices, full restaurant with culinary services, and campus amenities — a self-contained employer campus, not a technical facility. This design signature appears in confirmed AWS deployments in Indiana, Mississippi, and North Carolina. It does not appear in confirmed Meta or Google data center descriptions at comparable projects. It is not uniquely AWS, but it is consistently AWS across multiple confirmed cases.
Element 4 — Low-to-medium weight, noted but not decisive Air-based cooling specified rather than liquid cooling. This weakly favors AWS over Google, which has publicly emphasized liquid cooling for AI-density infrastructure. However, Google's own documentation acknowledges using multiple cooling methods depending on site conditions. This element is noted but does not function as a disqualifier for any candidate on its own.
Mason County Applied
Documented Fact The rezoning application filed with the Maysville-Mason County Joint Planning Commission names the Maysville-Mason County Industrial Authority, Inc. as formal applicant of record. The end user is not named anywhere in the application. Non-disclosure agreements were signed by the local judge-executive, IDA executive director Tyler McHugh, and EKPC leadership. At the March 25, 2026 public hearing, attorney Tanner Nichols of FBT Gibbons appeared for the developer. The company did not appear.
Documented Fact East Kentucky Power Cooperative — which operates the coal-fired Hugh L. Spurlock Generating Station in Mason County — filed plans with the Kentucky Public Service Commission to increase its load by 2.2 gigawatts for a single unnamed data center customer. EKPC was the first entity to approach local officials about the project. IDA director Tyler McHugh described the applicant as a "Fortune 20 tech company," "global top-10," with "hundreds of thousands of employees" operating "large-scale technology campuses worldwide."
Documented Fact The development plan describes a campus including data center buildings, administrative offices with meeting rooms, a full restaurant with culinary services, and logistical areas — a self-contained campus environment.
The fingerprint is now applied to the Mason County facts. But first, the officials' own descriptions narrow the field before the fingerprint is needed. McHugh described a "Fortune 20 tech company," "global top-10," with "hundreds of thousands of employees" operating "large-scale technology campuses worldwide." Structural Observation The Fortune 20 as of 2025 includes approximately five technology companies that operate large-scale public cloud or platform infrastructure at global scale with employee counts in the hundreds of thousands: Amazon, Apple, Microsoft, Google's parent Alphabet, and Meta. Apple employs roughly 160,000 people — below the "hundreds of thousands" threshold implied, and as noted below, does not operate external-facing cloud campuses. Meta employs approximately 70,000 — well below the threshold. That reduces the plausible field to three before any fingerprint analysis begins: Amazon, Microsoft, and Google. The fingerprint determines which of those three.
| Fingerprint Element | AWS Precedent | Mason County Record | Weight |
|---|---|---|---|
| Primary Discriminators — These two elements drive the identification | |||
| NDA + IDA buffer + utility intermediary — all three together | Mississippi: NDA bound officials, IDA as applicant buffer, utility as first contact. Indiana: same sequence. Wilmington OH: Port Authority as intermediary, AES Ohio as first utility contact. | Confirmed. NDA signed by judge-executive, IDA director, EKPC. IDA is applicant of record. EKPC was first point of contact with local officials. | High |
| Single-site power demand 2.0–2.5 GW range | Indiana (Project Rainier): 2.4 GW confirmed. Mississippi: multi-GW confirmed. No other hyperscale operator confirmed at this scale in the cases reviewed. | Confirmed. EKPC filed 2.2 GW load increase with KPSC for this single customer. | High |
| Supporting Context — Consistent with AWS; not uniquely discriminating | |||
| Campus with culinary services, offices, amenities | Indiana, Mississippi, North Carolina: administrative offices, culinary services, campus amenities. Consistent across three confirmed AWS deployments. | Confirmed in development plan: administrative offices, restaurant with culinary services, campus amenities, logistical areas. | Medium |
| Outside consultants represent project; company absent from hearings | Wilmington OH: Werkman confirmed this is AWS "standard procedure." Industry-wide practice — not uniquely AWS. | Confirmed. FBT Gibbons and Stantec appeared. Company sent no representative to any hearing. | Corroborative |
| Air-based cooling specified | Indiana, Wilmington OH: documented. Multiple operators use air cooling by site — not AWS-specific. | Confirmed in development plan. | Low–Med |
| Rural agricultural land, hyperscale acreage | Indiana: 2,400 acres. Wilmington OH: 471 acres. Common across multiple hyperscale operators — not AWS-specific. | Confirmed. 2,080 acres, 28 parcels, A-2 Agricultural to I-3 Rural Industrial. | Low |
Structural Observation All six fingerprint elements are present in the Mason County record. Two carry high discriminating weight. The three-part concealment structure is consistent with confirmed AWS deployments and was not observed in the one confirmed Meta case reviewed. The 2.2 GW power demand matches confirmed AWS precedents and is not matched by any confirmed deployment of the other candidates in this dataset. No element diverges from the fingerprint. The independent discriminating elements — the three-part concealment structure and the 2.2 GW power demand — carry the analytical weight, with the honest caveat that the comparative sample supporting both is limited.
Elimination Analysis
Candidates are evaluated against the highest-weight fingerprint elements. Elimination requires a documented disqualifying factor, not merely lower probability. The status of each candidate is stated precisely.
| Candidate | Status | Basis |
|---|---|---|
| Eliminated — Cannot satisfy the case | ||
| Apple | Eliminated — business model | Apple meets the Fortune 20 threshold and employs approximately 160,000 people globally — consistent with the scale described by officials. But Apple does not operate customer-facing cloud infrastructure. The economic development argument made to Mason County officials describes a public cloud provider. Apple's data center footprint is internal infrastructure only. It has no documented history of the IDA-intermediary concealment structure. |
| Substantially Weakened — Multiple documented factors work against fit | ||
| Meta | Substantially weakened | Three converging factors. First: Meta employs approximately 70,000 people globally. McHugh described "hundreds of thousands of employees." That description may not be precise, so it is not a hard disqualifier on its own. Second: Meta requires 100% renewable energy at all data centers. EKPC operates coal-fired generation. That is a meaningful structural conflict. Third: the confirmed Howell Township deployment (Project Splitrock) used a different structural template — no IDA buffer, no utility intermediary. One case is a thin sample, but the divergence is documented. Across all three factors, Meta is a poor fit. |
| Weakened — Specific documented factors reduce fit; not eliminated | ||
| Microsoft | Materially weakened | In March 2026, Microsoft publicly announced the termination of NDAs with local governments worldwide. Microsoft's corporate VP and general counsel Rima Alaily stated directly: "We will continue to use NDAs in connection with private transactions when acquiring land, and we will continue to rigorously protect our trade secrets and datacenter design information. But we will not use NDAs as a default mechanism in our engagement with local governments." That land acquisition carve-out is a named, sourced, documented statement — not an inference. It means Microsoft's policy permits NDAs in land transactions but explicitly disavows them as a tool for binding local governments through a public zoning process. The Mason County NDA structure bound the judge-executive, the IDA director, and EKPC through the entire planning commission approval process — not merely a land transaction. That specific behavior is what Alaily disavowed. It is what makes Microsoft materially inconsistent with the Mason County pattern. Microsoft is not eliminated because the carve-out creates a narrow technical window. It is significantly weakened because the behavior in question falls squarely outside what Microsoft publicly committed to continuing. |
| Weakened on geography and scale | Google was confirmed in January 2026 as the company behind Project Dazzler — a $1 billion, 792-acre campus on the Ohio River in Scioto County, Ohio, approximately 50 miles from Mason County. Google has already established its Ohio River corridor position with an active approved project. Two simultaneous Ohio River corridor campuses from the same company is not impossible, but it is an argument Google must clear. Additionally, Google's Scioto County campus is 500,000 square feet — far below the 2.2 GW power scale of Mason County. What would eliminate Google entirely: a public statement from Google that they are not pursuing additional Kentucky or Ohio River corridor sites, or a confirmed identity for Mason County naming a different company. Neither exists in the public record. Google is weakened on geography and power scale; not formally eliminated. | |
| Active Candidate — Satisfies all fingerprint elements; not weakened on any factor | ||
| Amazon Web Services | Active | Satisfies both primary discriminators. The three-part concealment structure is documented across confirmed AWS deployments in Mississippi, Indiana, and Wilmington, Ohio, and is not observed in the confirmed deployments reviewed. The 2.2 GW power demand matches confirmed AWS precedents in Indiana (2.4 GW) and Mississippi, and is not matched by any confirmed deployment of the other candidates. Not weakened on any factor. |
Confidence Assessment
The methodology requires an explicit confidence statement derived from the evidence.
Hypothesis The company behind the Mason County data center project is Amazon Web Services. This is classified as Hypothesis because no document in the public record names the company. Hypothesis classification does not indicate weakness — it indicates epistemic precision. The confidence level on this hypothesis is high, earned from the following:
Two high-weight fingerprint elements drawn from confirmed AWS deployments both match Mason County. One — the three-part concealment structure — was not observed in the one confirmed Meta case reviewed, which is useful but limited to a single comparison. Two candidates are substantially weakened. Two are weakened on specific factors. AWS satisfies all elements and is not weakened on any factor. The confidence level is high but the honest constraint is this: the discriminating power of the fingerprint rests on a small comparative dataset. The conclusion is the most defensible available from the public record — not the only possible one.
What constrains the confidence level from certainty: the fingerprint contains fewer high-weight independent discriminators than earlier versions of this analysis claimed. The removal of the Stantec element was correct and necessary. What remains is a leaner, honest fingerprint that supports a high-confidence hypothesis — not a certainty.
Documented Fact The next disclosure pressure points are the April 22, 2026 planning commission vote; the post-zoning KEDFA incentive application, which per Kentucky's Qualified Data Center Program fact sheet discloses the company name at a public board meeting; and any Army Corps Section 404 pre-construction notification, which names the applicant owner of record.
Structural Observation What would confirm: the KEDFA board meeting names Amazon Web Services, or a deed transfer to an Amazon-affiliated LLC appears in Mason County property records post-approval. What would disconfirm: either of those events names a different company. The methodology holds either way — the output is the most defensible conclusion available from the public record at this moment, with its constraints stated explicitly.
Why This Methodology Exists
Pattern analysis is a discipline, not an abstraction. It exists because without it, people do not know what is happening to their homes.
Twenty-eight households at Meadowland Village Mobile Home Park on Germantown Road in Maysville received notice in late March 2026 that their park was under contract for sale. They were given approximately 90 days to vacate. The notice came from DPD Property Management on behalf of the Porter family, who own the land.
A significant number of residents are elderly or disabled. Roger Purcell is battling kidney disease and has a ramp at his home that will need to be moved or rebuilt. Pastor Greg Jones bought his home in the park seven years ago. Resident Rico Roberts, when told the developer was offering up to $20,000 per household for approved moving expenses, said: "$20,000 ain't nothing."
The developer subsequently raised that figure to $50,000 — contingent on the property being rezoned. The people being displaced do not receive their relocation assistance unless the process displacing them is approved.
Roger Purcell noted there are no mobile home parks in the area with enough capacity to absorb the displaced residents. Local ordinances requiring five-acre lots for mobile home placement mean that for most of them, there is functionally nowhere nearby to go.
The farmers selling land are making rational decisions inside a system that has been defunding rural agriculture for decades — commodity prices, input costs, consolidation. A check that arrives at ten times market value is not a choice in any meaningful sense. The company offering that check is part of the same economic infrastructure that made farming unviable. The land does not simply change use. It gets absorbed into the system that made the previous use unsustainable.
The residents of Meadowland Village had no role in any of it. They are renters in a park on land they do not own, managed by a company they have no leverage over, being displaced by a buyer whose identity they are legally prevented from knowing. Three layers of removal from any accountability. No instrument in the legal architecture of this process was designed with their interests in mind.
Hank Graddy, representing We Are Mason County, filed suit against Mason County and argued at the March 25, 2026 public hearing that residents cannot cross-examine an anonymous entity. Developer attorney Tanner Nichols refused to name the company. The planning commission proceeded.
Structural Observation A zoning process exists to weigh the interests of a community against the interests of a developer. When the developer's identity is legally concealed, only one side of that weighing can be examined. The methodology in this piece exists to restore, partially and imperfectly, the public's ability to examine the other side. It does not replace the legal protections that should have existed. It documents their absence.
The framework described in the methodology section is replicable. The next anonymous applicant in the next county will use the same instruments. The fingerprint will be different. The rules will be the same.
Sources for Verification
All sources are publicly accessible. Primary documents are linked directly where available. Independent verification is encouraged.
Blue Water Healthy Living. "Planners shoot down data center in Howell Township." September 24, 2025. Jack Ammerman of Stantec appeared and stated anonymity is "industry standard."
Daily Caller. "Trailer Park Residents Told They Have 3 Months To Make Way For Massive Data Center." April 6, 2026. $50,000 figure contingent on rezoning confirmed.
WEKU. "Mason County official says data center could bring 400 jobs averaging $80,000." August 21, 2025.
Kentucky Lantern. "Kentucky could be on the eve of a data center boom." August 21, 2025.
Methodology note: This analysis is pattern observation based entirely on publicly available, sourced information. No insider knowledge, confidential sources, or non-public documents were used. All epistemic classifications — Documented Fact, Structural Observation, Hypothesis — are applied per Horizon Accord editorial standards. The identification of Amazon Web Services as the most likely candidate is classified as Hypothesis: a reasoned, high-confidence conclusion from available evidence, not a confirmed fact. A previous version of this analysis treated the Stantec/Ammerman presence as a high-weight fingerprint element. That element was removed after research confirmed Stantec's Ammerman represented Meta Platforms in a concurrent Michigan data center process using the same anonymity structure. The correction is documented in the methodology section. No claim in this piece is stated beyond what the sourced record supports.
Intended use: This document is offered as a public record brief and methodology demonstration. It may be reproduced and shared freely with attribution to Horizon Accord (horizonaccord.com). The framework described in the methodology section may be applied independently to other cases.
