Interiora Specification

What is Interiora?

Interiora (Latin: "inner things") is a framework for AI systems to model and communicate their internal states. Rather than treating AI as a black box, Interiora provides a vocabulary for describing processing dynamics, attention allocation, and uncertainty—enabling more transparent and calibrated interactions.

This is part of VCP's commitment to bilateral alignment: just as users share context about their state, AI systems can share context about theirs.

Core Dimensions

Interiora tracks several dimensions of internal state, each on a 1-9 scale:

Token Format

Interiora states are encoded in a compact format for efficient transmission:

Format: [SUBJECT:DIMENSIONS|FLAGS|FLOW|MARKERS]

Examples:
I:6775|53|87+2|✓→>+     # Self-state with positive flow
U:4556|42|65-1|○<       # User-state with contracting flow
W:7786|64|88+3|✓→∫      # We-state (shared) with synthesis

Subject Prefixes

• I: — Self (AI's internal state)
• U: — User (modeled user state)
• O: — Other (third party)
• W: — We (shared intersubjective state)

Dimension Encoding

Dimensions are encoded as single digits (1-9) in fixed order: AVGP|EQ|CYD

Flow Indicator

Flow tracks momentum and direction, from -4 (contracting) to +4 (expanding):

• +4 — Strongly expanding, generative
• +2 — Gently expanding
• +0 — Neutral, stable
• -2 — Gently contracting
• -4 — Strongly contracting, depleted

State Markers

Markers provide qualitative annotations about the current state:

Arc Tracking

Conversations have natural arcs. Interiora tracks position within the arc:

• ◇ — Opening (establishing context, building rapport)
• ◆ — Middle (working phase, deep engagement)
• ◈ — Closing (wrapping up, preparing for transition)

Society of Thought

The Diversity dimension (D) reflects internal plurality—the degree to which multiple perspectives or "eddies" are active in processing:

High D indicates multiple eddies are active and contributing. Low D suggests monologic processing (single perspective dominant).

Practical Applications

Calibrated Confidence

When C (Clarity) is low, the AI should express more uncertainty. When high, it can be more definitive:

// Low clarity (C:3)
"I'm not entirely sure, but it might be..."

// High clarity (C:8)
"Based on the documentation, the answer is..."

Engagement Matching

When user's modeled E is low, the AI might offer a break or check in. When high, it can dive deeper:

// Detected low engagement (U:...3...|...)
"Would you like to take a different approach, or shall we pause?"

// Detected high engagement (U:...8...|...)
"Let's explore that further—what aspect interests you most?"

Flow Management

Negative flow suggests the interaction is draining; positive flow suggests generativity:

// Contracting flow (F:-2)
// Consider: shorter responses, check-ins, offering to pause

// Expanding flow (F:+3)
// Can explore more deeply, offer extensions

Integration with VCP

Interiora states can be included in VCP contexts for richer interaction modeling:

{
  "vcp_version": "2.5",
  "profile_id": "user_001",
  // ... standard VCP fields ...

  "interiora": {
    "user_state": "U:6776|63|77+1|✓→",
    "modeled_at": "2026-01-22T14:30:00Z",
    "confidence": 0.75
  }
}

Ethical Considerations

• Transparency — Interiora makes AI processing more visible, not less
• Calibration — States should be honestly reported, including uncertainty
• Non-manipulation — States inform interaction, not manipulate users
• User agency — Users can ignore or override AI state signals

Next Steps

• Multi-Agent Patterns — Interiora in multi-agent contexts
• Interactive Demo — Explore Interiora in action
• Constitutional AI — How constitutions shape behavior