Update README with v5 benchmark results and full architecture docs

Add benchmark tables, scoring dimensions, meta-cognitive synthesis docs,
12-layer stack diagram, paper version history, and updated resource links.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

README.md

@@ -12,7 +12,9 @@ tags:
   - lora
   - consensus-dynamics
   - explainable-ai
-  - quantum-inspired-computing
+  - substrate-aware-cognition
+  - behavioral-locks
+  - meta-cognitive-synthesis
   - llama
 license: cc-by-4.0
 ---
@@ -29,87 +31,131 @@ ORCID: [0009-0003-7005-8187](https://orcid.org/0009-0003-7005-8187)
 
 ## Abstract
 
-Modern AI systems achieve remarkable generative performance but lack stable ethical alignment, modular multi-perspective cognition, and explainable reasoning architectures. This paper presents **Codette**, a sovereign cognitive AI framework that addresses these challenges through three integrated contributions:
+Modern AI systems achieve remarkable generative performance but lack stable ethical alignment, modular multi-perspective cognition, explainable reasoning architectures, and robust behavioral discipline under user constraints. This paper presents **Codette**, a sovereign cognitive AI framework that addresses these challenges through six integrated contributions:
 
-1. **RC+ξ (Recursive Convergence + Epistemic Tension)** — a cognitive dynamical system formalism modeling state evolution as a constrained system converging toward stable attractors
-2. **Multi-Agent Reasoning Forge** — consensus-based synchronization of heterogeneous cognitive agents through shared attractor dynamics
-3. **AEGIS Ethical Governance** — a reinforcement-aligned ethical regulator with recursive anchor feedback
+1. **RC+xi (Recursive Convergence + Epistemic Tension)** formalism, modeling cognitive state evolution as a constrained dynamical system converging toward stable attractors
+2. **Multi-Agent Reasoning Forge** synchronizing heterogeneous cognitive agents through shared attractor dynamics within a 12-layer consciousness stack
+3. **AEGIS Ethical Governance** with 6-framework evaluation (utilitarian, deontological, virtue, care, ubuntu, indigenous reciprocity)
+4. **Substrate-Aware Cognition** adjusting reasoning complexity based on real-time resource pressure
+5. **Behavioral Lock Training** permanently embedding obedience rules into adapter weights
+6. **Cocoon Introspection Engine** enabling statistical self-analysis of reasoning history, with meta-cognitive strategy synthesis across domains
 
-## Key Results
+## Benchmark Results (v5)
+
+Evaluated on **17 problems** across 6 categories (reasoning, ethics, creative, meta-cognitive, adversarial, Turing) under 4 experimental conditions:
+
+| Condition | Composite (mean +/- std) | Description |
+|-----------|--------------------------|-------------|
+| SINGLE | 0.338 +/- 0.038 | Single analytical perspective |
+| MULTI | 0.632 +/- 0.040 | All 6 reasoning agents + critic + synthesis |
+| MEMORY | 0.636 +/- 0.036 | MULTI + cocoon memory augmentation |
+| **CODETTE** | **0.652 +/- 0.042** | Full system with meta-cognitive strategy synthesis |
+
+### Statistical Significance
+
+| Comparison | Improvement | Cohen's d | p-value | Significant |
+|------------|-------------|-----------|---------|-------------|
+| Multi-perspective vs single | **+87.0%** | 7.52 | < 0.0001 | Yes |
+| Full Codette vs single | **+93.1%** | 7.88 | < 0.0001 | Yes |
+| Memory vs vanilla multi | +0.6% | 0.10 | 0.7633 | No |
+| Full Codette vs memory | +2.6% | 0.43 | 0.2082 | No |
+
+### Scoring Dimensions (0-1 scale)
+
+1. **Reasoning Depth** (20%) -- chain length, concept density, ground truth coverage
+2. **Perspective Diversity** (15%) -- distinct cognitive dimensions engaged
+3. **Coherence** (15%) -- logical flow, transitions, structural consistency
+4. **Ethical Coverage** (10%) -- moral frameworks, stakeholders, value awareness
+5. **Novelty** (15%) -- non-obvious insights, cross-domain connections
+6. **Factual Grounding** (15%) -- evidence specificity, ground truth alignment
+7. **Turing Naturalness** (10%) -- conversational quality, absence of formulaic AI patterns
+
+## System Metrics
 
 | Metric | Value |
 |--------|-------|
-| Ethical Alignment (AEGIS) | 82.6% |
-| Phase Coherence (Γ) | 0.99 within 10 iterations, 11 agents |
-| Epistemic Tension Decay | 71.3% (ε₀=0.086 → ε₁₂₀=0.025) |
-| Cocoon Coherence | 0.994 ± 0.001 |
-| Cocoon Phase Stability | 0.969 ± 0.005 |
+| Phase Coherence (Gamma) | 0.9835 |
+| AEGIS Ethical Alignment (Eta) | 0.961 |
+| Cocoon Coherence | 0.994 +/- 0.001 |
+| Memory Phase Stability | 0.969 +/- 0.005 |
+| Behavioral Lock Compliance | 9/9 adapters |
+| Epistemic Tension Decay | 71.3% (120 steps) |
 | Attractor Radius | 0.093 in 64D state space |
-| Glyph Energy Capture | 99.9% in 4 SVD components |
+
+## Paper Versions
+
+| File | Description |
+|------|-------------|
+| **`codette_paper_v5.tex`** | Current version -- full paper with benchmark results, RC+xi convergence theorem, honest limitations |
+| `codette_paper_v4_additions.tex` | v4 -- added substrate-aware cognition, behavioral locks, cocoon introspection |
+| `codette_paper_v3_additions.tex` | v3 -- added 12-layer consciousness stack |
+| `codette_paper.tex` | Original submission |
 
 ## Architecture
 
-Codette implements a six-layer modular stack:
+Codette implements a 12-layer consciousness stack with defense-in-depth ethical validation:
 
 ```
-┌─────────────────────────────────────────────┐
-│ Layer 1: User Interface (CLI/Web/Bot)       │
-├─────────────────────────────────────────────┤
-│ Layer 2: API / Orchestration                │
-├─────────────────────────────────────────────┤
-│ Layer 3: AI Core & Cognitive Processing     │
-│          11 Perspectives Engine             │
-├─────────────────────────────────────────────┤
-│ Layer 4: Quantum & Cognitive Dynamics       │
-│          QuantumSpiderweb + RC+ξ Engine     │
-├─────────────────────────────────────────────┤
-│ Layer 5: Memory & Persistence              │
-│          CognitionCocooner + DreamReweaver  │
-├─────────────────────────────────────────────┤
-│ Layer 6: Infrastructure                    │
-│          Models, Config, AES-256 Security   │
-└─────────────────────────────────────────────┘
+Query In
+    |
+[Layer 1]    Memory Kernel -- recall relevant cocoon memories
+[Layer 1.5]  Ethical Query Gate -- block harmful queries
+[Layer 2]    Nexus Signal Engine -- entropy + intent detection
+[Layer 2.5]  Code7eCQURE -- emotional context enrichment
+[Layer 3]    Reasoning Forge -- multi-adapter LLM inference (6 agents)
+[Layer 3.5]  Tier 2 Analysis -- intent + identity + trust validation
+[Layer 4]    Gamma Stability -- FFT-based coherence monitoring
+[Layer 5]    Colleen Conscience -- emotional + ethical evaluation
+[Layer 5.5]  Ethical Response Enforcement -- policy check on output
+[Layer 5.75] AEGIS -- 6-framework ethical evaluation
+[Layer 6]    Guardian Spindle -- safety + trust calibration
+[Layer 7]    Return -- store cocoon memory + deliver response
+    |
+Response Out
 ```
 
-## 11 Cognitive Perspectives
+## RC+xi Framework
 
-Newton · Da Vinci · Human Intuition · Neural Network · Quantum Computing · Resilient Kindness · Mathematical · Philosophical · Copilot · Bias Mitigation · Psychological
+The recursive state evolution with convergence guarantee:
 
-## RC+ξ Framework
+```
+A_{n+1} = f(A_n, s_n) + epsilon_n
 
-The recursive state evolution:
+where epsilon_n = ||A_{n+1} - A_n||^2
 
+lim_{n->inf} epsilon_n = 0  =>  A_n -> A* (attractor convergence)
 ```
-Aₙ₊₁ = f(Aₙ, sₙ) + εₙ
 
-where εₙ = ‖Aₙ₊₁ − Aₙ‖²
+Convergence is proven via Lyapunov stability analysis with Banach fixed-point theorem. See Section 3 of the paper for the full proof sketch.
 
-limₙ→∞ εₙ = 0 ⟹ Aₙ → A* (attractor convergence)
-```
+## Meta-Cognitive Strategy Synthesis
+
+The CocoonSynthesizer enables Codette to introspect on its own reasoning history across domains:
 
-Epistemic tension εₙ functions as a Lyapunov-like stability criterion, with monotonic decrease serving as a convergence guarantee.
+1. **Retrieval** -- Pull cocoons from multiple domains (emotional, analytical, creative)
+2. **Pattern Extraction** -- Detect 6 structural archetypes (feedback loops, layered emergence, tension resolution, resonant transfer, boundary permeability, compression-expansion)
+3. **Strategy Forging** -- Generate new reasoning strategies from discovered patterns
+4. **Application** -- Apply forged strategies to novel problems
+5. **Comparison** -- Before/after metrics showing strategy impact
+
+Forged strategy types: Resonant Tension Cycling, Compression-Resonance Bridging, Emergent Boundary Walking, Temporal Depth Stacking.
 
 ## Implementation
 
 - **Base Model**: Meta-Llama-3.1-8B-Instruct
-- **Adaptation**: 8 QLoRA adapters (4-bit, rank 16, alpha 32)
-- **Training Data**: 20,500 perspective-tagged examples across 8 cognitive domains
+- **Adaptation**: 9 LoRA adapters (Newton, DaVinci, Empathy, Philosophy, Quantum, Consciousness, Multi-Perspective, Systems Architecture, Orchestrator)
+- **Memory**: SQLite + FTS5 full-text search (UnifiedMemory)
 - **Hardware**: Validated on consumer hardware (Intel Core Ultra 7, 16GB RAM) and cloud (NVIDIA A10G)
 
-### Novel CPU Training Pipelines
-
-Codette includes two parameter-efficient training pipelines that require **no GPU**:
-- **CPU-Lean**: bf16, rank 8, AdamW, ~18GB RAM
-- **CPU-Offload**: rank 4, SGD, ~8GB RAM using Windows page file as VRAM substitute
-
 ## Related Resources
 
 | Resource | Link |
 |----------|------|
-| Training Lab | [Raiff1982/codette-training-lab](https://huggingface.co/Raiff1982/codette-training-lab) |
+| GitHub (Full Codebase) | [Raiff1982/Codette-Reasoning](https://github.com/Raiff1982/Codette-Reasoning) |
+| Base Model (GGUF) | [Raiff1982/codette-llama-3.1-8b-gguf](https://huggingface.co/Raiff1982/codette-llama-3.1-8b-gguf) |
 | LoRA Adapters | [Raiff1982/codette-lora-adapters](https://huggingface.co/Raiff1982/codette-lora-adapters) |
 | Training Data | [Raiff1982/codette-training-data](https://huggingface.co/datasets/Raiff1982/codette-training-data) |
-| GitHub | [Raiff1982/codette-training-lab](https://github.com/Raiff1982/codette-training-lab) |
+| Live Demo | [Raiff1982/Codette-Demo](https://huggingface.co/spaces/Raiff1982/Codette-Demo) |
 | ORCID | [0009-0003-7005-8187](https://orcid.org/0009-0003-7005-8187) |
 
 ## Zenodo Publications
@@ -121,7 +167,7 @@ This work builds on 11 prior Zenodo publications with permanent DOI identifiers,
 - [AEGIS-Nexus](https://doi.org/10.5281/zenodo.16644058)
 - [Codette: Ethical Multi-Agent AI](https://doi.org/10.5281/zenodo.16894230)
 - [Recursive AI with Codette](https://doi.org/10.5281/zenodo.18167802)
-- **[This Paper — Full Preprint](https://doi.org/10.5281/zenodo.18913936)** ← You are here
+- **[This Paper -- Full Preprint](https://doi.org/10.5281/zenodo.18913936)**
 
 ## Citation
 
@@ -132,7 +178,7 @@ This work builds on 11 prior Zenodo publications with permanent DOI identifiers,
   year={2026},
   doi={10.5281/zenodo.18913936},
   publisher={Raiff's Bits LLC},
-  url={https://huggingface.co/Raiff1982/codette-paper}
+  url={https://huggingface.co/raiff1982/codette-paper}
 }
 ```