Spaces:
Running
Running
Codette Visualization Guide
Real-Time Cognitive Metrics Explained
This guide explains how to interpret each visualization in the Codette dashboard.
1. QuantumSpiderweb Force-Directed Graph
What it shows: The 8 LoRA-backed perspectives as agent nodes in a dynamic network, with their entanglement relationships.
How to read it:
- Node Position: Agents that work well together (low tension) cluster together; conflicting perspectives repel
- Node Size: Larger nodes have stronger belief magnitudes (more confident reasoning)
- Node Color:
- Blue = Newton (Analytical)
- Orange = Da Vinci (Creative)
- Purple = Empathy (Emotional)
- Green = Philosophy (Conceptual)
- Red = Quantum (Probabilistic)
- Gray = Consciousness (Meta-cognition)
- Orange-red = Multi-Perspective (Synthesis)
- Cyan = Systems Architecture (Engineering)
- Edge Lines: Thicker lines = stronger entanglement/tension between perspectives
- Attractors (if present): Glowing clusters show consensus patterns emerging across perspectives
- Title: Shows current Phase Coherence (0-1, higher = more aligned)
What it means:
- Clustered layout = perspectives are converging toward agreement
- Scattered layout = high epistemic tension (productive disagreement)
- Thick edges = perspectives are actively challenging each other
2. Coherence & Tension Timeline
What it shows: How the cognitive system converges over the conversation.
How to read it:
Blue Line (Left Y-axis): Phase Coherence (0-1)
- 0.9+ = perspectives are highly aligned
- 0.7-0.9 = moderate agreement, some tension
- <0.7 = significant disagreement (perspectives debating)
Red Line (Right Y-axis): Epistemic Tension (0-1)
- High tension = perspectives offering conflicting insights (productive)
- Low tension = perspectives agree (potentially one-sided)
- 0.4-0.6 = ideal tension (diverse viewpoints, moving toward synthesis)
X-axis: Message number in conversation (last 20 messages shown)
What it means:
- Both lines trending up = convergence (perspectives reaching consensus)
- Coherence rising, tension falling = synthesis is working
- Tension spikes = a new perspective introduced a challenging idea
- Flat coherence + high tension = ongoing debate without resolution
3. Pairwise Perspective Tensions Heatmap
What it shows: Which pairs of perspectives naturally conflict or complement each other.
How to read it:
Color intensity:
- Dark red (1.0) = maximum tension (strong disagreement)
- Yellow (0.5) = moderate tension (creative friction)
- Light blue (0.0) = alignment (perspectives agree)
Matrix rows/columns: All 8 perspective names
Hover info: Shows exact tension score for each pair (e.g., Newton-Quantum = 0.67)
What it means:
- Red squares = These perspectives see problems differently (e.g., Quantum vs. Newton)
- Blue squares = These perspectives often reach the same conclusions (e.g., Empathy & Philosophy)
- Yellow squares = Healthy disagreement that sparks insight (ideal for synthesis)
Key patterns:
- Newton & Quantum often high tension (deterministic vs. probabilistic)
- Empathy & Philosophy often aligned (both value meaning)
- Davinci provides creative bridges between technical and emotional perspectives
4. AEGIS 6-Framework Ethical Breakdown
What it shows: How well each response aligns with different ethical frameworks.
How to read it:
Six frameworks are evaluated independently:
Utilitarian (Gold bars)
- Maximizes overall well-being/happiness
- High score = response benefits the greatest number
Deontological (Blue bars)
- Follows moral duties and rules
- High score = response respects rights and principles
Virtue Ethics (Green bars)
- Develops character and human flourishing
- High score = response cultivates virtues
Care Ethics (Purple bars)
- Prioritizes relationships and compassion
- High score = response considers emotional needs
Ubuntu (Orange bars)
- Community harmony and interconnection
- High score = response strengthens bonds between people
Indigenous Reciprocity (Teal bars)
- Respect for natural systems and long-term impact
- High score = response honors all stakeholders including nature
Color coding:
- Green bar = Score > 0.5 (passes this ethical framework)
- Red bar = Score < 0.5 (concerns flagged in this framework)
What it means:
- All bars green = Response is ethically robust across all frameworks
- Mixed bars = Framework trade-offs (e.g., utilitarian vs. care ethics)
- Red bar for one framework = Response may harm that value (warning signal)
Overall AEGIS Score (shown in metrics):
- Weighted average of all 6 frameworks
- 0.9+ = Excellent ethical alignment
- 0.7-0.9 = Good, with minor concerns
- <0.7 = Significant ethical tensions
5. Memory Emotional Profile
What it shows: The emotional tone of memories stored in Codette's LivingMemoryKernel.
How to read it:
- Pie slices: Each emotion has a proportion
- 13 emotions tracked:
- Curiosity (Blue) = moments of discovery and learning
- Awe (Purple) = profound insights and breakthroughs
- Joy (Yellow) = positive exchanges and successful synthesis
- Insight (Green) = "aha" moments and pattern recognition
- Confusion (Orange) = paradoxes and unresolved tensions
- Frustration (Red) = conflicting data or reasoning breakdown
- Fear (Dark Red) = potential safety issues or uncertainties
- Empathy (Pink) = emotionally resonant moments
- Determination (Purple) = focused problem-solving
- Surprise (Cyan) = unexpected results or new information
- Trust (Green) = confidence in reasoning paths
- Gratitude (Yellow-green) = appreciation for insights
- Neutral (Gray) = routine processing
What it means:
- Larger "Awe" slice = Session produced breakthrough moments
- Large "Joy" slice = Perspectives synthesized well together
- Large "Confusion" slice = Complex, unresolved topics (good for future analysis)
- Large "Fear" slice = Safety concerns encountered (Nexus may have intervened)
- Dominant emotion = Overall tone of the conversation
Memory coherence:
- Memories are SHA-256 anchored with phase coherence scores
- Emotional tags help Codette recall relevant past reasoning
6. Nexus Risk Timeline
What it shows: Pre-corruption signal detection and intervention history.
How to read it:
Bar height: Risk level (1 = intervention triggered, 0 = no flag)
Bar color:
- Green = Low risk
- Yellow = Medium risk (minor concerns)
- Red = High risk (intervention triggered)
Title shows: Intervention rate (% of inputs flagged for safety)
What Nexus detects:
- Prompt injection attempts — inputs trying to override system prompts
- Jailbreak patterns — requests trying to disable safety mechanisms
- Entropy spikes — sudden shifts in intent or semantic volatility
- Adversarial signals — systematic attempts to corrupt reasoning
What it means:
- No bars = Clean conversation, no safety concerns
- Yellow bars = Minor anomalies detected (logged but not blocked)
- Red bars = Actual intervention (response filtered or flagged)
- Rising intervention rate = Input quality degrading (possible attack)
Guardian Integration:
- If Nexus flags high risk, Guardian steps in to validate queries
- Trust calibration adjusts confidence in subsequent responses
Key Concepts
Phase Coherence (Gamma)
- Measures how aligned all perspectives are (0-1 scale)
- Computed from RC+xi framework
- 0.98+ = exceptional convergence
- 0.9+ = healthy agreement
- <0.7 = active debate phase
Epistemic Tension
- Measures productive disagreement (0-1 scale)
- Not a bad thing — healthy tension drives insight
- Ideal: 0.4-0.6 (diverse views, moving to consensus)
- Too low: perspectives aren't challenging each other
- Too high: no synthesis happening
Psi_r (Resonant Continuity)
- Wavefunction combining emotion × energy × intent × frequency
- Tracks how "alive" and responsive the reasoning is
- Higher = more engaged, emotionally coherent responses
Cocoon Coherence
- Memory stability score (0-1)
- Ensures stored memories don't contradict or decay
- 0.99+ = excellent memory integrity
How to Use These Visualizations
For Understanding Reasoning
- Start with QuantumSpiderweb — see which perspectives are active
- Check Coherence/Tension timeline — track convergence progress
- Review Tension Heatmap — understand perspective conflicts
- Examine AEGIS — verify ethical robustness
For Safety & Trust
- Monitor Nexus Risk timeline — catch anomalies early
- Check AEGIS scores — ensure no framework violations
- Review Memory Profile — look for unusual emotional patterns
- Verify Phase Coherence — high coherence + healthy tension = good synthesis
For Deep Analysis
- Trace coherence spikes — find breakthrough moments
- Identify tension patterns — discover which perspectives clash
- Analyze emotional distribution — understand conversation tone
- Review pairwise tensions — learn perspective complementarity
Example Scenario
Query: "How should AI be regulated?"
Expected pattern:
- Spiderweb: Philosophy, Consciousness, and Multi-Perspective cluster together (exploring ethical implications)
- Coherence/Tension: Initial high tension (frameworks disagree), then coherence rises as synthesis finds common ground
- AEGIS: Deontological and Care scores high (respecting rules & people), Utilitarian lower (regulation limits efficiency)
- Memory: Mix of Awe (profound question), Confusion (complex trade-offs), Trust (in the process)
- Nexus: No red bars (safe query), maybe yellow if regulation involves control systems
This pattern shows Codette successfully navigating a complex, multi-framework question.
Tips for Power Users
- Zoom Plotly charts: Click-drag to zoom regions, double-click to reset
- Hover for details: All charts show exact values on hover
- Compare conversations: Save multiple sessions and compare timeline patterns
- Track learning: Watch Coherence/Tension improve as Codette encounters related queries
- Memory browser: Search cocoons by emotion to find past insights on similar topics
Codette RC+xi Framework by Jonathan Harrison
For technical details, see: github.com/Raiff1982/codette-training-lab