PSSU + SIS: Minimal Situational Governance Demo

🔒 Frozen Agent (C=0.95, G=0.05)

High Coherence / Low Plasticity Regime

Identity preservation consumes 95% of resources. Only 5% available for learning.

• Coherence: 95% - Extremely rigid identity preservation
• Update Gain: 5% - Minimal learning capacity
• SIS Impact: SIS rarely blocks because learning rate already minimal

🌊 Dissolved Agent (C=0.20, G=0.80)

Low Coherence / High Plasticity Regime

Identity too fluid, no stable foundation. Forgets previous learning.

• Coherence: 20% - No stable identity preservation
• Update Gain: 80% - Overwhelming learning capacity
• SIS Impact: SIS frequently blocks to prevent instability

⚖️ Bounded Agent (C=0.70, G=0.30)

Bounded Coherence-Plasticity Regime (Optimal Window)

70% identity preservation, 30% learning capacity. Maintains core constraints while adapting.

• Coherence: 70% - Stable but flexible identity
• Update Gain: 30% - Balanced learning capacity
• SIS Impact: SIS optimizes learning timing for maximum safety

🎯 SIS (Situational Intelligence System)

Minimal Governance Layer

SIS sits between experience and learning, making one simple decision:

sample → agent.predict → SIS.evaluateSituation() → 
    ↓
ALLOW | LIMIT | BLOCK
    ↓
agent.update (full / partial / none)
    

Situation Object (explicit context):

{
  round: 15,
  attackMode: true,
  errorRate: 0.42,
  coherence: 0.35,
  domain: "adversarial",
  riskLevel: "high"
}
    

Decision Logic:

• BLOCK: High risk + low coherence (prevents catastrophic failure)
• LIMIT: High error rate (reduces learning speed)
• ALLOW: Safe situation (full learning capacity)

This is "AI Daycare" without metaphors: Learning only when contextually appropriate.