How this publication is governed
A publication that defines the operating doctrine for governed AI decisions must itself be governed by the same doctrine. This page is that recursion made explicit.
Constitutional principle
Council operating model
This publication operates under OptimaX AI Governance v2.2, sealed 2026-03-31. The review model separates drafting, adversarial review, audit, research, and authorization responsibilities. Public pages describe the control objective; role internals remain counsel-vetted.
The hardest invariant: review authority is separated from deployment authority. Public claims are published only when the evidence path can be inspected without exposing private implementation detail.
How a claim becomes a published page
- Builder draft. The Builder produces a draft using the prompt-OS (versioned, signed, schema-bound prompts) under a session anchor that names role, invariants, disclosure tier, and forbidden behaviors.
- Adversary pass. The Adversary attempts to break the claim — find the most credible reason it is wrong, the cleanest factual error, the most quotable embarrassment.
- Auditor review. The Auditor (separate model, separate context) reviews against full document parity. Findings are recorded with severity P0–P3.
- Sovereign sign. The Sovereign reads, signs, and authorizes. Sealed RFCs are version-bumped.
- Disclosure-tier check. The build pipeline refuses to publish a Tier 1+ phrase on a Tier 0 page, atomically.
- Receipt. An RDL entry is logged before the artifact ships. Tree hash is computed.
Disclosure governance
Three disclosure tiers govern what appears where:
| Tier | Audience | What appears |
|---|---|---|
| Tier 0 | Public — this site | Category language, architectural framing, doctrine, conformance results, contestation procedure |
| Tier 1 | Strategic counterparties under mutual NDA | Architectural summary at higher resolution; reference documents under counsel sign-off |
| Tier 2 | Pilot-LOI counterparties | Implementation specifics, integration interfaces, source review |
| Tier 3 | Sealed integration partners | Full system access under signed master agreement |
The disclosure-tier rendering contract is software-enforced. Tier 1+ phrases (specific cryptographic constructions, restricted implementation details, threshold values, claim-bearing patent language) cannot publish to Tier 0 routes. The build fails atomically. Sovereign override is a first-class, logged, expiring event.
Amendment process
This publication's governance is itself amendable. Doctrine changes follow Decision Class D3:
Builder draft → Auditor adversarial review with full document parity → Adversary break-attempt → Sovereign sign → version bump → ratification at quarterly review. The most recent amendment is the Document-Parity Rule, proposed 2026-05-01 after a detected review-coverage gap and scheduled for ratification into AI Governance v2.3 at the 2026-06-30 quarterly review.
Contestation
Anyone can contest a published claim, registry entry, or conformance score. Routing: contest@governedai.ai. See the corrections policy.