SOV3³ + OOWM

The sovereign sandwich. Open-weight models wrapped in a signed, evolving substrate — eight distinct model classes, four governed brains, every hop Ed25519-signed.

SOV3 running on VM 207 tools live 4 of 7 NNs need retraining handle_oowm_status = stub

① The Sovereign Sandwich

Every request flows: OFFLINE ⟶ SIGIL ⟶ SOV3 MIDDLE ⟶ SIGIL ⟶ ONLINE. Both halves Ed25519-signed.

☁
ONLINE
OpenAI / Claudefrontier · API
GLM-5.2Chinese cloud
Groqvote / fallback
275+ MCPsfederated external
S I G I L · E D 2 5 5 1 9 · S I G I L
🧠
SOV3 MIDDLE
207 MCP toolslive
Mamba-2 SSM16-dim state
7 trained NNs3 strong · 4 weak
Traibgle votingBFT
BFT council33 nodes · 22 quorum
Ed25519 SIGILevery hop
qwen3:30b-a3bMoE-LARGE (when loaded)
S I G I L · E D 2 5 5 1 9 · S I G I L
🔒
OFFLINE
qwen3:30b-a3bMoE · deep reason
qwen2.5:3bMoE · fast route
moondreamvision
deepseek-r1:7bCoT reasoning
BGE-M3embedding
Kokoro-82MTTS

② SOV3³ — Four Sovereign Brains

Same base OOWM, reconfigured as four governed brains. Live on the VM.

Brain 01

🛡 SOVEREIGN-COMPLIANCE
qwen3:30b-a3b
+ GLM-5.2
+ BFT vote
EU AI Act, UK AI Bill, GDPR Art 22. The live revenue brain.
⚠ EAT FROZEN

Brain 02

🛡 SOVEREIGN-DEFENSE
qwen3:30b-a3b
+ DeepSeek-R1
JSP 936 defensive doctrine. Designed, not firing.

Brain 03

🧠 SOVEREIGN-INTUITION
qwen3:30b-a3b
+ Gematria
+ 16-dim Mamba
Fast intuition. Pattern recognition. Signal early-warning.

Brain 04

🗣 SOVEREIGN-VOICE
qwen3:30b-a3b
+ Kokoro TTS
The spoken voice. MEOK OS voice dock uses this.

③ Eight Model Types — Every Class in the Estate

All distinct model architectures, all open-weight or open-source. No foundation model from scratch.

MoE
Mixture-of-Experts
  • Reasoning qwen3:30b-a3b
  • Fast route qwen2.5
Apache 2.0● Live
MoM
Mixture-of-Models
  • Vision moondream
  • Spatial zamba-large
  • VL qwen-vl
Open● Live
SSM
State-Space Model
  • Long mem Mamba-2
  • State-dim 16
Open● Live
Reasoning
Chain-of-thought LLM
  • deepseek-r1:7b
MIT● Live
TTS
Text-to-Speech
  • Kokoro-82M
  • Fallback Piper
Apache/MIT● Live
Embedding
Vector retrieval
  • Primary BGE-M3
  • Rerank BGE-reranker
MIT● Live
Trained NN (ours)
Governance signals
  • 7 models
  • 3 strong · 4 weak
CSOAI own◐ Mixed
Cloud ensemble
Vote / fallback
  • GLM-5.2 · Claude
  • Groq
API● Live

④ Seven Trained NNs — Honest Scorecard

Verified on MEOK MCP v3.0.0. No "best-of" cherry-picking — all 7 listed.

Model
Metric
Samples
Verdict
creativity_assessment_nn
r² 0.91
350
✓ strong
care_pattern_analyzer
mae 0.037
600
✓ strong
relationship_evolution_nn
mae 0.071
500
✓ strong
care_validation_nn
mae 0.19
19 ⚠
⚠ tiny sample
partnership_detection_ml
mae 0.22
19 ⚠
⚠ tiny sample
threat_detection_nn
acc 0.45
33
✗ weak — retrain
dependency_detection_nn
acc 0.22
50
✗ weak — retrain

⑤ Thirteen Task Categories — SOV3 OOWM Router

Each task gets routed to the optimal (brain, model, mindset) triple.

compliance
hybrid_sovereign · deepseek-r1:7b · wise
Reasoning wins on EU AI Act / GDPR
defence
hybrid_sovereign · qwen3:30b-a3b · bold
MoE-LARGE on JSP 936
sov3
hybrid_sovereign · deepseek-r1:7b · logical
BFT council + Traibgle voting
world
mom_large_offline · moondream · curious
Multi-modal for geo / sensor
physical
mom_large_offline · moondream · creative
Capillary + muscle simulation
sovereign
moe_large_online · qwen3:30b-a3b · careful
Ed25519 + crypto decisions
sovtown
mom_large_offline · moondream · diplomatic
NPC / agent interactions
deploy
moe_large_online · qwen3:30b-a3b · decisive
Vercel + PyPI deployments
ethics
hybrid_sovereign · deepseek-r1:7b · wise
Care principles + bias checks
companion
mom_large_offline · moondream · diplomatic
Orb HP + bond
reasoning
hybrid_sovereign · deepseek-r1:7b · logical
Explicit reasoning tasks
speed
mom_large_offline · qwen3:0.6b · focused
Sub-second real-time
default
hybrid_sovereign · qwen3:30b-a3b · wise
Safe default for unknown

⑥ Live Deployment

207
MCP tools
330
SOV3 tools
275+
external MCPs
33
BFT nodes
49 GB
data moat
3,000
tok/s OLM
13d
VM uptime
8
model classes
4
sovereign brains
7
trained NNs
13
task routes
661
MCP pkgs shipped

⑦ Honesty Flags — Do Not Overclaim

FLAG 01 · CRITICAL

handle_oowm_status is a stub

Returns True for mamba_warm, moe_loaded, moe_connected — hardcoded, not a live probe. Never cite as proof the OOWM is "running."

FLAG 02 · WARNING

qwen3:30b-a3b is DESIGNED, not loaded here

On this Mac, ollama list is empty. King-hive only carries llama3.2:1b/3b. 30B base is designed for the 192 GB M4 Mac — confirm there before claiming "live."

FLAG 03 · WARNING

4 of 7 trained NNs need retraining

creativity (r² 0.91), care_pattern (mae 0.037), relationship (mae 0.071) are strong. threat_detection (acc 0.45) and dependency_detection (acc 0.22) need real labelled data.

FLAG 04 · WARNING

"Bitcoin anchor" / "consciousness 0.775" are labels

Metaphor, not literal. Keep out of investor copy and the Series A deck.

FLAG 05 · CRITICAL

SOVEREIGN-DEFENSE brain is EAT-frozen

Per EAT directive 2026-07-02: no offensive / defence-capability work. Only ASSURANCE / GOVERNANCE / CYBER firing today.

A sovereign sandwich around open-weight models (qwen3-MoE + Mamba-2 + vision MoM), every hop Ed25519-signed, reconfigurable into four governed brains — plus seven small trained governance NNs (three strong, four honest).

The moat is the signed-governed substrate — not a from-scratch foundation model.