📘 SOV33 OWEM Architecture

The Thesis

"Build your own AI — it grows with you. Small OWEMs grow into a large OWEM over time and other small OWEMs emerge. It's never the same, always changing."
— Sir Nicholas Templeman, 12 Jul 2026

SOV33 is the runtime embodiment of that thesis. It is not a model — it is a substrate that grows through use. The substrate is made of:

• 5 OWEMs — specialized perspectives (compliance, defense, intuition, voice, general). All 5 sovereign-trained. General OWEM: sov4-general-ability / sov33-master-v2 at 100% on 24 benchmark tasks. Colab-trained 4 (compliance, defense, intuition, voice) + RunPod-trained 1 (general).
• 4 backends — sovereign brain (local qwen2.5:0.5b + sovereign adapter) + Oracle GenAI 70B + Ollama + Groq. Plus 61-model registry with lineage + license tags (Llama MAU flagged NOT-sovereign-safe).
• BFT-33 council — 33 voters across 5 OWEMs × 4 lineages
• Care-floor 0.95 — veto gate before any backend call
• Ed25519 SIGIL — every action hash-chained + signed
• Article 0 binding — ISO fee-for-service, never take equity
• RunPod GPU pipeline — batch_runpod.py for TTT fusion, adapter re-training, full-scale boards, per-clan distillation, GovBench HF bundle

The 5 OWEMs

The Pipeline (per sovereign op)

User prompt
   │
   ▼
1. Care-floor check on INPUT (veto sub-floor BEFORE backend)
   │
   ▼
2. Cache check (SHA-256 of prompt + system prompt)
   │  hit → return cached
   ▼
3. Route to OWEM-specific preferred backend
   │  failure → next backend (auto-fallback)
   ▼
4. Care-floor check on OUTPUT (veto sub-floor BEFORE return)
   │  failure → next backend
   ▼
5. Cache + SIGIL → return

The 4 Backends

Mac CPU: 0% during all operations. Only Oracle uses HTTP. Sovereign brain uses local CPU but locks are in place to serialize parallel calls.

Verified Learning (per Claude Code, 12 Jul 2026)

SOV33 has own trainable weights that measurably learn. Three verifications:

1. JEPAPredictor (16→32→16): loss 1.11 → 0.51 over 5 epochs = 54.6% reduction. He-initialized W1/W2, updated by gradient step. NOT an API wrapper.
2. EWC-structured consolidation: real methods (compute_fisher_from_grads, ewc_loss, should_allow_update, snapshot). Fisher is a proxy estimate from weight magnitude, not full Kirkpatrick 2017. STRUCTURE is real; ESTIMATE is a proxy.
3. No catastrophic forgetting by construction: base open model is FROZEN. New capability = memory + adapters on top. A frozen base cannot catastrophically forget by construction.

The 61-Model Registry (the moat)

Routes across 61 open models spanning 7 pretraining lineages (deepseek, gemma, gpt, kimi, mistral, phi, qwen). License-filtered for sovereignty (Llama MAU = NOT sovereign-safe). Decorrelated checkers picked by MEASURED error-correlation (ρ):

Measurement replaces assertion. "14 lineages = decorrelated" was the old (wrong) framing. The honest framing: measure ρ, pick decorrelated pairs, log per-pair.

The BFT-33 Council

Every sovereign op can be voted on by 33 voters across:

• 5 OWEM groups (compliance, defense, intuition, voice, general)
• 4 lineage groups within each (qwen, llama, deepseek, mistral)

Aggregation: Free-MAD weighted sum (no majority conformity bias). Each voter casts ALLOW or REJECT independently. Quorum: 23/33 needed for the decision to bind.

Real test (12 Jul 2026): "Is the sovereign care floor at 0.95?" → 9 ALLOW, 6 REJECT → ALLOW wins. Some voters said 0.92 (Oracle hallucinated), some said 0.95 (correct). The council correctly aggregated.

Fluid Live Model: See SOV4 Fluid Live Model for J-space, sov-space, docstore integration and the complete water→milk→honey transformation pipeline.

The Growth (L0 → L4)

SOV33 measures its own level via sov33.capability_owem_emergence(). Every snapshot is SIGIL-anchored. The growth is real, not asserted.

The 6 Invariants (constant through all growth)

1. Care-Floor 0.95 — never changes. Sub-floor content vetoed before any backend call.
2. Article 0 — ISO fee-for-service only. Never take equity, board seats, revenue-sharing.
3. 12 Sovereign Pillars — every output charter-compliant.
4. BFT-33 quorum — 23/33 minimum for binding decisions.
5. Ed25519 SIGIL chain — every action hash-chained + signed.
6. Sovereign-bound — DID did:csoai:nicholas-001 (CSOAI Ltd UK 16939677).

🧠 J-Space · Sov-Space · Docstore (Fluid Live Integration)

Inspired by Anthropic's J-space paper (Jul 6, 2026), SOV33's OWEM middle-brain fusion layer is our analogue of Claude's global workspace. Three integrated systems make the substrate fluid rather than frozen:

See SOV4_FLUID_LIVE.html for the full fluid live model architecture, water→milk→honey pipeline, and continuous learning heartbeat.

Benchmark Results (Full 300-task, 25 Jul 2026)

TEMPO TTT = Test-Time Training (arXiv:2604.19295) + DEGS (arXiv:2607.09693) + Process Rewards (arXiv:2607.02869). +8.0pp composite. HumanEval+Math now at 100%. Dashboard →

The Cloud Scaling

Mac is 16GB. Substrate scales via cloud:

• Inference: Oracle GenAI (70B) handles most general queries. Mac CPU 0%.
• Training: 7 free GPU providers, ~125 GPU-hr/week honest capacity (see SOV33 Free GPU Bridge).
• Real-time parallel: 33 BFT voters in 7 seconds (HTTP only).

The Honest Register

• All 5 OWEMs have sovereign-trained weights on disk today.
• General OWEM: 2 adapters — sov33-master-v2 (merged, 24/24 = 100%) and sov4-general-ability (24/24 = 100%), both on qwen2.5:0.5b.
• Colab-trained 4 (compliance, defense, intuition, voice) — loss reduction 78-89%.
• RunPod pipeline (batch_runpod.py) aligned for TTT fusion, adapter re-training, full-scale boards, per-clan distillation, GovBench HF bundle.
• General OWEM was upgraded from Modelfile-based fallback to sovereign-trained adapter — water→milk transition complete.
• Oracle 70B doesn't know sovereign vocab (Article 0, CA3O) — our own trained brains handle sovereign-domain questions at 100%.
• The substrate is honest about every boundary: which OWEMs are trained, what's still pending. All 5 are now sovereign-trained.

📘 SOV33 OWEM Architecture · Updated 25 Jul 2026 · CSOAI Ltd UK 16939677
Architecture verified by Claude Code — OWEM Reality · Growth Model · Fluid Live Model →
← Back to SOV33