Sovereign Provenance Architecture

The architecture is the architecture. The brands are the brands.

There can't be enough of us in this game until we are operating under shared architectural truth. This document proposes that minimum.

The Sovereign Provenance Architecture (hereafter SPA) describes a consortium specification for AI governance systems that operate under inspectable, deterministic, and provenance-bearing principles. SPA does not propose a brand, product, or platform. It proposes the architectural minimum that allows independently-operating firms — each with its own commercial brand and methodology — to converge on shared truths regarding how AI systems must be governed at formation, custody, and execution.

The motivation: as AI governance vocabulary proliferates across firms and conferences (admissibility, custody, refusal rails, anti-entropy, sovereignty, deterministic governance, pre-execution integrity), there is a growing gap between firms that ship inspectable architecture and firms that publish vocabulary without architecture behind it. SPA closes that gap by defining what conformant implementation means.

This document is published as v0.3. It is a Living Document. It will evolve. The protocol for revision is specified in §8; the v0.2 → v0.3 changelog is recorded inline there.

The six principles every conformant implementation MUST honor.

The following tenets are normative. They define what it means for an architecture to operate under the Sovereign Provenance Architecture, regardless of brand, vocabulary, or commercial form.

1. Provenance over performance. A system's claims about its own behavior are subordinate to its ability to demonstrate the origin of every decision it produces.
2. Inspectability over opacity. Architecture MUST be readable from outside the firm. Source, specification, or live surface — at least one inspectable artifact per claimed capability.
3. Determinism at the gate. Where governance is claimed, gating decisions MUST be reproducible: same input, same context, same outcome.
4. Custody before execution. Every action that materially affects an external system MUST be admitted by a custody check before commitment.
5. Anti-entropy as discipline. Governance gates decay without active maintenance. Implementations MUST specify how their gates resist decay over time.
6. Consortium, not vendor lock. Implementations SHOULD be interoperable in vocabulary, conformant in minimums, and sovereign in branding. SPA does not certify, license, or charge.

Five layers. Names normative; implementation is not.

A conformant implementation MUST document, name, and inspectably ship the following five architectural layers. The layer names are normative for vocabulary alignment. The layer implementations are sovereign — each firm builds them as fits its commercial frame.

3.1 — The Admissibility Layer

Determines whether a state, instruction, or trajectory is structurally valid to enter the system. Answers "should this exist at all?" before "is this accurate?" Inadmissible states MUST be refused before propagation, not corrected after.

3.2 — The Provenance Layer

Maintains custody of origin, authorship, data inputs, and transformation history for every state admitted by §3.1. Implementations MAY use ledgers, tamper-evident logs, cryptographic attestation, or signed event chains. The layer's output MUST answer the question "what was made, by whom, with what data, when?"

3.3 — The Anti-Entropy Layer

Specifies how the system prevents gates from decaying into ceremony over time. Implementations MUST publish their maintenance discipline — periodic re-verification, drift detection, replay-based audit, or equivalent. A gate without anti-entropy is rubber-stamping with extra steps.

3.4 — The Sovereignty Layer

Preserves meaningful human authority at the decision boundary. Defines who may override admissibility, when, with what credential, and how that override is recorded. Sovereignty MUST NOT be implicit; it MUST be named, scoped, and ledgered.

3.5 — The Sanctioning Layer

Issues the final tamper-evident decision (admitted / refused / escalated) that downstream systems consume. The decision MUST be replayable: any third party with appropriate authority MUST be able to reconstruct why a given decision was rendered.

These five layers are derived from the Five-Layer Integrity Architecture documented in ECCO's Master Context (Public Edition, CC BY-NC-ND 4.0). The Vocabulary Registry (§6) reflects terms in convergent use across the AI governance space; vocabulary origin is acknowledged in §9.

What it takes to claim SPA v0.3 Conformant.

A firm, product, or implementation is SPA v0.3 Conformant if and only if all of the following are true:

1. All five required layers are documented in a publicly readable specification reachable from a stable URL.
2. At least one architectural artifact is inspectable per the criteria of §5 — source code repository, deployable spec, or live demonstration surface.
3. Vocabulary is interoperable — terminology used aligns with the SPA Vocabulary Registry (§6) or provides a public mapping from internal terms to SPA terms.
4. Versioning is transparent — the implementation's spec carries a version number and a changelog reachable from its primary surface.
5. Stewardship is named — a single accountable individual or entity is publicly identified and reachable for verification of claims.

A firm or product that honors the Core Tenets (§2) but has not yet met all Conformance Criteria MAY declare itself SPA v0.3 Aligned. Alignment is a position; Conformance is a verification.

Three pathways. One is required.

Per Tenet §2.2, every claimed architectural layer MUST have at least one of the following be true:

1. Source inspectable. Implementation source published under any OSI-approved license at a stable repository URL.
2. Spec inspectable. Layer specification published as Markdown, HTML, or PDF at a stable URL, with sufficient detail that an independent implementer could re-implement the layer from the document alone.
3. Surface inspectable. Live deployed surface that a third party can interact with and observe the layer's behavior in real conditions.

A capability claim with none of the above does not qualify under SPA v0.3, regardless of the quality of the firm's marketing presentation. Vocabulary without inspectability is not architecture.

Shared terms across conformant implementations.

The following vocabulary is normative for SPA v0.3. Implementations MAY use internal aliases but SHOULD publish a mapping to these terms. The registry will grow with future revisions as new terms reach community consensus.

Admissibility

The structural-validity check applied before a state, instruction, or trajectory is permitted to propagate. Operates at formation, not at output.

Provenance

The traceable origin and transformation history of a state — what was made, by whom, with what data, when.

Custody

The ongoing maintenance of provenance across the lifecycle of a state. Provenance is the snapshot; custody is the chain.

Refusal Rail

An enforceable boundary that halts inadmissible action in real time. Refusal is not failure; it is protection.

Anti-Entropy

The maintenance discipline that prevents governance gates from decaying into ceremony. Without anti-entropy, gates become rubber stamps.

Sovereignty

Meaningful human authority at the decision boundary, named and scoped, never implicit.

Sanctioning

The final, tamper-evident decision (admitted / refused / escalated) that downstream systems consume.

Replay-Verifiability

The property that any third party with appropriate authority can reconstruct why a given decision was rendered.

Decision Boundary

The architectural surface at which governance is enforced and human authority may be invoked.

Pre-Execution Governance

Governance applied before the system commits. The category that contains §3.1 through §3.5.

Consortium

SPA's structural model: voluntary association of independently-operating firms converging on shared architectural minimums — sovereign in branding, conformant in substance, equal in participation. Distinguished from hierarchy, certification body, or trade association.

Stewardship, not incorporation.

SPA is governed by a stewardship model. There is no SPA LLC. There is no SPA Foundation. There is no membership fee, no certification body, and no commercial moat at the specification layer.

The current steward is Ethereal Connections Co., of Denver, Colorado. Stewardship is open to transfer or rotation by community consensus, defined as a written majority of named Conformant signatories agreeing in writing to a successor.

A firm MAY declare Conformance or Alignment unilaterally by:

1. Publishing a public statement linking to this specification at its canonical URL.
2. Specifying which layers of §3 the firm's implementation addresses, with citations to the inspectable artifacts of §5.
3. Maintaining the criteria of §4 for as long as the declaration stands.

Firms wishing to be listed in the public SPA Conformance Registry maintained by the steward MAY submit at the contact address in the letterhead. Inclusion requires verifiable inspectability under §5; the steward reserves the right and obligation to verify before listing.

How this document evolves.

This document is a Living Specification. Revisions are governed as follows:

1. Additive changes (vocabulary additions, clarifications, examples, registry growth) increment the minor version: v0.1 → v0.2.
2. Breaking changes (modifications to the Tenets of §2, restructuring of the Required Layers of §3, or revisions to Conformance Criteria of §4) increment the major version: v0.x → v1.0.
3. Major revisions require a 30-day public review period and notice to all currently listed Conformant signatories.
4. All previous versions remain accessible at versioned URLs. The canonical URL always points to the current version; previous versions are reachable at a versioned suffix.
5. A changelog is maintained at the canonical URL and updated with every revision.

Changelog

v0.1 → v0.2 · 13 May 2026 · §9 corrected. v0.1 listed Vocabulary Contributors that had not declared per §7 and whose external URLs did not meet §5 inspectability. v0.2 removes those entries, replaces the sub-section with a Vocabulary Origin note that attributes the §6 registry honestly, and verifies that ECCO's own Conformant listing tracks actually-inspectable artifacts at issue time. Minor: Master Context reference normalized to "(Public Edition)" without version pin to track the living edition. The correction holds the steward to the same standard the steward holds the field to.

v0.2 → v0.3 · 13 May 2026 · Two coupled changes addressing prior inconsistencies between stated principle and operative terms. Vocabulary: "Federation" → "Consortium" across the document, including Tenet 6 header ("Consortium, not vendor lock"), the §7 section heading ("Consortium Model"), meta tags, and body references; a new Consortium entry added to the §6 Vocabulary Registry. License: CC BY-NC-ND 4.0 → CC BY-SA 4.0. The prior license's NonCommercial and NoDerivatives clauses contradicted both §7's "no commercial moat at the specification layer" and §4's invitation of vocabulary mappings as inspectable derivatives; ShareAlike preserves openness across the implementation chain while retaining attribution to ECCO as steward. Both changes are treated as additive under §8.1: no Tenet's normative content was modified — only labeling and the license-layer's openness — and the operative requirements of §§2, 3, and 4 are unchanged between v0.2 and v0.3. The steward acknowledges the prior license terms were inherited from template rather than chosen with the Consortium principle in view; v0.3 corrects the call.

What's already verifiable at v0.3 issue.

The following are the only claims this document makes about firms or implementations as of v0.3 issue. Listing as Conformant requires verification of all §4 criteria. The steward maintains the registry under the discipline of §§5 and 7, and applies that discipline to the steward's own listings.

Vocabulary Origin

The terminology in §6 (Vocabulary Registry) derives from the architecture work documented in the ECCO Master Context (Public Edition) and the surfaces referenced above. The AI governance field at large is converging on overlapping vocabulary in parallel — admissibility, custody, refusal rails, anti-entropy, sovereignty, deterministic governance, pre-execution integrity — through independent work at multiple firms, conferences, and standards bodies. SPA does not claim invention of any term in §6; it claims the architectural minimum that gives the terms operational meaning.

The Vocabulary Registry remains open to formal contribution per §7 from firms whose published work would extend or refine the registry. Submissions are inspected against §5 prior to listing; the same discipline applies to the steward.

Firms wishing to be added as Conformant signatories at future revisions MAY declare in accordance with §7. The registry is open. Inspectability is the credential; the steward verifies before listing.

The work is the bridge.

The Sovereign Provenance Architecture is open. There is no fee, no exclusion, and no commercial moat at the specification layer. Firms operating in the AI governance, integrity, custody, or admissibility space are invited to declare Alignment or Conformance, contribute to the Vocabulary Registry, or propose revisions in accordance with §7 and §8.

The doctrine that informs this specification — provenance over performance, infrastructure over influence, doctrine over excuse — is offered without proprietary claim. Firms are encouraged to publish their own architecture, ship their own surfaces, and let inspectability be the credential rather than the gatekeeping.

The architecture is the architecture. The brands are the brands. The work is the bridge.