Watermark.
All sites
Switch site 38 sites · 11 basins 8 open

Each site is one Watermark investigation. The page color says what opens; the campus line is the build on the ground — two different clocks.

MAU Maumee Basin 9 sites
2MI The Two Miamis 9 sites
SE Southeastern Basins 12 sites
NE Northeast Basins 8 sites
The page — what opens on Watermark Open · Draftfull record Queued · Watchingpreview page Restrictedrequest access
The campus — the build on the ground Investigating Confirmed Under construction Operational
All 38 sites — the network index →

The defense nexus — what the corridor shows, and what it can’t

A #233 extension narrative, and the sharpest open question in the file. Like end-use-and-workloads.md it is hand-assembled analysis over cited records, and it ends at the question rather than closing it. Every statement carries its register: [verified] (read from a cited record), [inference] (a labeled reading of it), [open] (a question the record does not answer), [reference] (an outside-published spec). This page is built to a single rule from the project’s method: geographic adjacency, a capability, and a named market segment are an inferred connection — legitimate to raise as a question, never to assert as a finding. The reader is owed the discipline more here than anywhere else, because the subject is the one most easily turned into innuendo.

The plainest way to say what this report is: there is a defense installation near the campus, the developer has the credentials to do defense work, and an industry witness told a state committee that hosting the government is a normal line of business. Each of those is true. None of them, alone or together, shows that the Lima campus does defense work. Holding those two sentences at once — the facts are real, the conclusion is not earned — is the whole exercise.

It is worth saying who raised this first. The person who put the defense question to Ohio’s data-center committee is the relator behind this record, a cloud engineer who builds for regulated industries — and he framed it, on the record, as a question he could not answer: “I can speculate there. I think you probably understand that most of that is classified.” He called the broader pattern “likely speculative.” [verified: relator testimony, 2026-06-04] If the witness who introduced the thread labels it speculation, a page assembled from the public record has no business doing less.

———

the geography is real

About five and a half miles south of the campus sits the Joint Systems Manufacturing Center — the Lima Army Tank Plant — operated by General Dynamics Land Systems. It is not a rumor; it is on the parcel map. The corpus carries it as five contiguous parcels totaling ~384 acres, every one of them owned, in the auditor’s own field, by “UNITED STATES.” [verified: data/site/bundle/feeds/geo/jsmc.geojson] Measured against the campus parcels, the nearest edges are ~5.5 miles apart (centers ~6.6 mi) [verified: computed from geo/campus + geo/jsmc] — the two share a city and a corridor study area, not adjoining land. GDLS’s operation of the plant is a matter of public record [reference].

So the corridor contains, within a few miles of each other, the largest data-center build in the county’s history and one of the country’s two heavy-armor manufacturing plants. That is a striking coincidence of geography. It is also only that — two facts that share a map and five and a half miles of city, which is precisely the kind of connection the method says to raise as a question and stop.

———

the capability is real

The developer is Google [verified, #234], and Google can, as a technical and contractual matter, do high-authorization government work. The relator testified — with sources in his written submission — that Google has achieved IL-6, the DoD impact level for data classified up to SECRET [verified: relator testimony; reference: DoD CC SRG]. Google’s air-gapped Distributed Cloud appliance holds DoD IL5 [reference]. The federal market is not hypothetical for the industry, either: before the same committee, AWS named the Department of War and the CIA among “11,000 government agencies of all classification levels” [verified: hearing record, 2026-06-04 morning panel], and the relator tied the timing to Executive Order 14265 (signed 2025-04-09), which pressed defense primes to modernize cloud procurement [verified].

The capability has an economic edge the rest of the record sharpens. Government cloud runs 20–30% above commercial rates — the govcloud_premium axis of the economics-scenarios feed, pooling BCG and AWS GovCloud-vs-commercial pricing [reference]; an industry range whose application to this campus stays [open] by construction — and, the relator’s point, an authorized facility is closed to the community that subsidized it: “I cannot use that data center if it is a FedRAMP-compliant facility.” [verified: relator testimony] That is why the question is not idle. If the campus were a high-authorization enclave, the public-benefit math the abatement was scored on would not hold — a sealed federal supply chain seeds no local cluster (see end-use-and-workloads.md). But that is an if, and the record does not resolve it.

What that if would cost the public is now priced, and it is worth being precise about what the pricing is and is not. The economic ledger’s GovCloud / defense-hardened profile turns two knobs to the ends a hardened, cleared facility would imply — a larger real-property share and a leaner headcount — and lands at the top of the subsidy band (see the-economic-ledger.md, issue #1665). It is a labeled counterfactual: it says what such a facility would cost, and the model that produces it structurally cannot tag it [verified] or carry it above low confidence. It is not evidence of a defense nexus, and nothing here should be read as sharpening the case for one. It exists so that the reader who asks “and if it were?” gets a number instead of a hint.

———

proximity is not connection

Set the three confirmed facts down together: a defense plant nearby, a developer with the clearances, an industry that markets to the government. The temptation is to let them lean on each other until they look like a finding. The method forbids it, explicitly: name-proximity, temporal coincidence, and geographic adjacency are the signatures of an inferred connection, and an inferred connection is never a fact.

Nothing in the corpus ties the campus to the plant, to GDLS, or to a defense workload. There is no contract, no filing, no dated communication naming both. The proximity to the JSMC is geography; the IL-6 credential is a capability Google holds everywhere it operates, not a fact about Lima; the AWS testimony is about AWS. Each thread, followed honestly, ends without reaching the campus.

The committee record is consistent with that limit rather than against it. Google’s own witness testified to Ohio’s data centers and did not name Lima at all [verified, #234] — the silence is documented; what it means is [inference], not a finding.

———

the wall of “no records”

There is one place the record could have spoken, and the answer it gave is its own kind of fact. The public-records request asked the County for any communications between it and the DoD or its contractors — GDIT, GDLS — concerning the facility or the corridor. The County’s response: “No records.” It asserts that none exist on its side. [verified: bosc-prr-production-2026-06-05.response-index.yaml, item 2]

A clean negative is a result, and it should be stated cleanly: on the County’s account, there is no documented defense channel. But “no records” is not quite a no-link finding, because elsewhere in the same production the phrase did heavier lifting than it should have — conflating “we do not hold it” with “it does not exist” [verified: same index]. So the honest reading is narrow: the County holds no such records, or produced none. That forecloses one avenue. It does not establish that the connection is absent, and it does not establish that it is present. It leaves the question exactly where it was — open, and now with one door confirmed shut.

———

the money is a separate question

Everything above reads the nexus as geography — how close the plant is, what the developer is cleared for. There is a second reading the lens could not previously even state: whether the defense base distorts the local economy the campus lands in. It is worth running, because it comes back as a limit rather than a confirmation.

On the payroll record it does not. BLS QCEW 2024 annual averages for Allen County put federal-government employment at 321 jobs across 15 establishments, averaging $83,773 a year against the county-wide $58,790 — a real wage premium on a very small base [verified: data/reference/economics/baseline.yaml]. The location quotient is 0.33: the county’s share of federal employment is a third of the national one. An enclave economy would show the opposite.

The reason is structural, and it is why the number is a limit rather than a finding. The JSMC is government-owned and contractor-operated — the plant belongs to the Army, the workforce belongs to General Dynamics Land Systems. Those jobs book as private manufacturing, not federal employment, so the federal ownership row does not count the arsenal at all [inference]. QCEW also excludes uniformed active-duty entirely [reference: the model's standing coverage caveat]. The payroll series is simply not the instrument that would measure this.

Two threads stay open, and they are the ones worth pulling. The land is off the tax rolls: the five federally-held parcels carry $14,688,900 of assessed market value across 384.15 acres, and the United States owes no local property tax on any of it [verified: data/reference/allen-gis/parcels.defense.yaml]. And no abatement, PILOT, or public-benefit agreement tied to the JSMC footprint has been searched for or found [open] — unlike the campus’s own CRA, which is on record. The prime’s federal dollars are real but are the wrong scale to attribute here: General Dynamics Corp’s $301B in all-time prime-award obligations (88% defense-agency) is a national total resolved by UEI, not a Lima figure [verified: data/reference/usaspending/awards.yaml], and the same graph overlay that surfaces it appears on every site in the network. It is a cross-reference to follow, not a local finding.

where this stops

What would actually close it is small and specific: the facility’s authorization posture — whether it carries any FedRAMP or DoD impact-level authorization — is a single disclosable fact that would answer the end-use question and this one at once. It is not in the record. The relator’s sixth recommendation to the committee was that it should be [verified: relator testimony]. Until it is disclosed, the question is held open by two things at once: the classification that would keep a real defense use quiet, and the “no records” wall that keeps even the absence of one unproven.

The public instrument surfaces have now been searched for it, and they come back empty on every path [verified: federal-authorization-instrument-search.md, #1480]. USASpending and SAM.gov hold no award or recipient for the campus’s owner-of-record shell (Bistrozzi LLC / Bistrozzi Addition LLC); the FedRAMP Marketplace lists Google Cloud service-wide, at High, and names no Lima facility. That last point is the one worth holding onto, because it is structural, not circumstantial: Google’s public-sector compliance runs on Assured Workloads, which by Google’s own account “does not rely on physical infrastructure distinct from Google’s public cloud data centers” [reference: Google Cloud, FedRAMP High blog]. A FedRAMP or DoD-IL authorization in that model binds to the service, not to a building — so a facility-scoped instrument for Lima cannot exist by construction, and its absence proves nothing in either direction. The negative maps where the ceiling is; it does not lower it. The nexus stays [inference] and the search declines to force a link — the same discipline Urbana’s analogous search held to [reference: #1353].

That is an uncomfortable place to end, and it is the correct one. A defense nexus is not a finding of this record. It is a question the record raises by what it contains — a federal plant a few miles down the corridor, a developer cleared to SECRET, a procurement order, an industry that hosts the CIA — and cannot answer by what it withholds. The walk’s discipline was built for exactly this thread: to let the reader see the question clearly without being handed a conclusion the evidence has not earned. The honest end is the open one.

———

sources

  • The JSMC parcels (owner “UNITED STATES”, ~384 acres) — data/site/bundle/feeds/geo/jsmc.geojson
  • Relator testimony, 2026-06-04 (the IL-6 statement, the “speculation” framing, the FedRAMP-access point) — data/extracted/legal/select-committee-2026/hearings-audio/bosc-committee-testimony-2026-06-04.transcript.md
  • Relator written testimony (impact levels, EO 14265, the Ohio defense footprint, the six recommendations) — data/extracted/legal/select-committee-2026/relator-testimony/bosc-written-testimony-2026-06-01.md
  • AWS / DoW + CIA + “all classification levels”; hearing cross-read — data/extracted/legal/select-committee-2026/select-committee-2026.hearing-index.yaml
  • The “no records” defense-channel response — data/extracted/legal/prr-mandamus/bosc-prr-production-2026-06-05.response-index.yaml (item 2)
  • The #1480 documented negative search for a facility-naming federal-authorization instrument (FedRAMP / DoD IL / GSA-DoD award / cloud-region) — data/extracted/lima/federal-authorization-instrument-search.md
  • The county employment / wage baseline behind the capture reading (QCEW 2024, federal ownership row) — data/reference/economics/baseline.yaml
  • The federally-held parcels’ assessed value + acreage, and the typed scan registers — data/reference/allen-gis/parcels.defense.yaml
  • The prime’s federal obligations (national totals, resolved by verified UEI) — data/reference/usaspending/awards.yaml
  • The end-use frame this sits inside — end-use-and-workloads.md, DOSSIER.md