The Sovereignty Foundation
The evidence record
Artificial Governed Intelligence is not a slogan — it is a documented chain, and every link is dated
A category earns its name when the term corresponds to a defined technical architecture rather than an aspiration. The record below is deliberately layered: applicant-created terminology, filed technical disclosure, independent government timestamps, substantive examination outcomes, working software, machine-verifiable evidence, and a conformance institution. No single layer proves the category. The accumulation is the argument — and each layer is independently checkable, because every date below was fixed by a third party or a filing, not by this page.
The chain
The architectures enter the federal record before a word is published. Independently verifiable government timestamps on the underlying mechanisms.
Government timestampThe category is named and argued in public: “Toward Artificial Governed Intelligence” (May 11), the ontological restatement “AGI = Governed Intelligence” (May 15), and a white paper already naming RITA, SAGE, and Mirror (May 10). Third-party platform dates.
Applicant-created terminology · publicly datedThe architecture becomes running software: 208 crates, ~155,000 lines, 3,958 tests across the governance chain, capability warrants, execution boundary, and evidence stores.
Working implementationThe entire operating system extracted into claims, diagrams, and a 280-signature mathematical registry; the estate's recurring structures — deterministic evaluation, chained evidence, separated trust boundaries, fail-closed dispatch — filed two months before the Standard abstracted them.
Filed technical disclosureThe architecture's properties abstracted into eleven testable requirements across three maturity tiers — normative language derived from mechanisms already filed and already running, not the other way around.
Normative specificationEvery gated function call conditions propagation on a mathematical predicate; the witness is the authorization. Worked end-to-end on the issued patent's pipeline: 974 captured observations, twenty claim-property tests, twelve witnesses in an append-only ledger, deterministic across hosts — and four rejection tests proving the negative case: an out-of-bounds value is refused and does not propagate.
Execution-level enforcement · machine-verifiable evidenceThe evidence methodology is not bespoke to one system: a machine-readable cell contract (status · captured-time · metrics), hardware-fingerprint scoping, four live measurement cells, and a documented one-file path for an outside developer to add another. An extensible conformance architecture, not a pile of internal tests.
Conformance methodology · extensibleSubstantive examination crossed: claims covering semantic identity authentication, governance thresholds, permit / delay / escalate / deny decisions, and cryptographically verifiable provenance with downstream verification and no central callback.
Issued claims · substantive examinationThe behavioral role-inference engine crosses substantive examination without an intervening rejection: a finite-state dispatch indicator — permit, restrict, prevent, defer — whose record changes the entity’s operational disposition downstream. “Governed” with structural consequences, in an allowed claim’s own words.
Allowed claims · substantive examinationThe Standard, the conformance corpus with two scaffolded cells, the Charter separating standard-setting from certification, the process, the production pilot cohort — and the Known Deltas register, disclosing the reference implementation’s own shortfalls before certification of anyone is offered.
Certification institution · public reviewAn independent implementation passes the published corpus, the Foundation re-derives the sealed results, and the grant lands in the public ledger. The chain closes where it aimed from the first essay: the claim is checkable.
The mark, earnedNo single layer proves the category.
The accumulation is the argument.