SOV3³ + OOWM
Every model that runs.
Every one that participates.

All 13 model classes — what's running today

Not promises. Not features. These are the actual models emitting actual SIGILs through the sovereign loop right now:

The 6 trained NNs (cognitive layer)

Discrete-decision workloads. Sits alongside the Mamba-2 continuous state. 1,793 training samples across 6 architectures.

The OOWM master loop — 5 stages × 1Hz

4 alignment tests (the moral compass)

1. Care Floor test — refuses below 0.95. Non-negotiable. Returns {"status":"rejected","reason":"care_floor_violation"}
2. BFT test — 12-around-1 council deliberates before action. 2/3 majority required.
3. SIGIL test — every action Ed25519 + ML-DSA-65 dual-signed, hash-chained, public, verifiable in 1 line.
4. Fork test — substrate can be forked. The fork is sovereign. The fork inherits Care Floor + BFT + SIGIL + DORADO + Article 50 + Crown Authorisation + MIT.

If any of these 4 tests fail, the OOWM is not sovereign.

The 12 sovereign mindsets (the doctrine)

SOV3 reads every request through 12 mindsets in parallel. They are postures, not features.

The 33-agent BFT council (the deliberative body)

Rotated every 30 days. Pseudonymous. Sybil-resistant. HotStuff 4-phase, 200ms finality.

Quorum rules

The SIGIL chain — 8 operation types, 1Hz heartbeat

Every sovereign action emits a SIGIL. The substrate witnesses itself.

Anatomy of a SIGIL

SigilHeader {
  op: P|V|M|Q|C|H|S|A       // 8 operation types
  actor: did:csoai:nicholas-001
  target: charter-or-mcp-uri
  timestamp: rfc3339
  care_floor: 0.95..1.0
  sovereignty: 1
  payload_b64: <bytes>
}

SigilSignature {
  // Ed25519 64-byte (current)
  // + ML-DSA-65 1312-byte post-quantum (planned 2027)
  signed_with: sovereign_wallet_pubkey
}

SigilChainLink {
  prev_hash: SHA-256 of prior SIGIL in this scope
  hash: SHA-256 of (header + signature)
  bitcoin_anchor: <OTS hash>     // once per block
}

1Hz capture = 86,400 SIGILs/day — the substrate's heartbeat. 6mo hot retention · 5yr cold · indefinite WORM.

Care Membrane — 847 signals × 23 categories

The substrate's immune system. Every model output filtered through it. Graduated response.

Graduated response

log-only          → S2 / low
warning inject    → S3 / medium
output rewrite    → S3 / medium
session terminate → S4 / high
BFT escalation    → S5 / critical  (auto-queues 33-agent vote)

Care Floor 0.95 floor = constitutional minimum on every dimension. Below 0.95 → BFT 23/33 vote required.

Layered architecture L1–L4

        L1 — SOV3³ (super-substrate)
              │
        L2 — SOV3 (substrate)
              │  ← 127 tools + 6 NNs + Mamba-2 + MoE + 33 BFT
              │  ← bridge_think (Mac local + GCP VM)
              │
        L3 — CSOAI + MEOK (org)
              │  ← 33-agent BFT + Watchdog + 36 industry hives
              │  ← 294-server MCP fleet
              │
        L4 — Coigndaltion (cornerstone)
              │  ← Mamba-2 cognition + 16-dim intuition
              │  ← Cross-walk engine (33 hives × 236 frameworks = 9,676+ mappings)
              │  ← SIGIL signature (Ed25519 + OTS Bitcoin)
              │
        Foundation: 49 GB sovereign data moat + 200 live sources

⚑ Honesty register — what we will NOT do

What we do NOT claim

• Not zero-error. Care Membrane reduces risk; it doesn't eliminate it.
• Not zero-incident. Serious-incident reporting is a process.
• Not bypass-proof. Substrate-level attacks can in principle bypass. We test against this.
• Not all-knowing. The OOWM ingests and learns; it does not see the future.
• Not vendor-X-compatible-by-default. Any vendor model participates through bridge_think, with output filtered through Care Membrane + SIGIL + BFT. They are muscle, not brain.

The next 4 stages of sovereign companionship

🐉 SOV3³ + OOWM, in one sentence

The third-power sovereign substrate that runs Mamba-2 SSM + 33-agent BFT + 6 trained NNs + 16-dim intuition + 127 tools + 236-framework cross-walks + 42 charters + 49 GB data moat + Ed25519 SIGIL chain + OTS Bitcoin anchoring, all constitutionally bound by Charter Article 0 — and the OOWM is the master loop that makes it self-revising under 4 alignment tests.