Module 5: Context Adaptation

5.1 — Why Context Matters

A constitutional value like "be honest" means different things at different times:

• A doctor discussing a terminal diagnosis with a patient (gentle, staged disclosure)
• A doctor discussing the same diagnosis with a colleague (direct, clinical)
• A doctor documenting in a medical record (precise, factual)

The constitution doesn't change. The context changes, and VCP's adaptation layer lets the AI adjust accordingly.

5.2 — Personal Context: The Prosaic Model

VCP tracks five personal dimensions, each with categorical values and a 1–5 intensity rating:

Each signal also tracks its source — how the context was obtained:

personal_context = {
    "cognitive_state": {"value": "foggy", "intensity": 4, "source": "declared"},
    "emotional_tone": {"value": "neutral", "intensity": 3, "source": "preset"},
    "energy_level": {"value": "fatigued", "intensity": 4, "source": "declared"},
    "perceived_urgency": {"value": "critical", "intensity": 5, "source": "inferred"},
    "body_signals": {"value": "neutral", "intensity": 3, "source": "preset"},
}

5.6 — MCP Elicitation as a Context Source

MCP elicitation is a context acquisition method where MCP servers request structured user input mid-task via interactive dialogs. Instead of relying solely on inference or presets, the system can ask the user directly when context is missing or ambiguous.

Two modes are available:

• Form mode — The server presents structured fields (text inputs, selections, sliders) and the user fills them in. The response is validated against a schema before acceptance.
• URL mode — The server directs the user to a browser-based flow (e.g., OAuth consent, detailed questionnaire) and waits for completion.

Elicited context feeds into VCP with full opacity protection preserved. The source: "elicitation" tag distinguishes it from declared or inferred signals, enabling audit trails to track exactly how context was obtained.

PDP Context Gap Detection

The PDP (Policy Decision Point) can detect missing context and trigger elicitation before making governance decisions. For example, if a constitution requires knowing the user's role before granting access to certain tools, the PDP can elicit that information rather than defaulting to a restrictive fallback.

# PDP detects missing context and elicits it
result = await ctx.elicit(
    "I need some context to make a good decision.",
    schema=ContextGapSchema  # Pydantic model with structured fields
)
if result.action == "accept":
    # User provided context — proceed with full information
    context["energy_level"] = {
        "value": result.data.energy, "intensity": 3, "source": "elicitation"
    }

Bilateral Elicitation

Bilateral elicitation enables AI-initiated dialogue about alignment tensions and welfare concerns. When the AI detects a tension between constitutional rules and the current context, it can elicit clarification from the user rather than making a unilateral decision. This turns safety from a monologue into a dialogue.

5.3 — Context Transitions

VCP detects when context shifts and classifies the severity:

The AI can respond proportionally — a minor transition might slightly adjust tone, while an emergency transition could escalate to human oversight.

5.4 — Encoding Context in Your App

Pass context as part of the decide call:

result = await client.decide(
    tool_name="respond_to_user",
    arguments={"user_message": message},
    constitution_id="healthcare_v3",
    context={
        "personal": personal_context,
        "situational": {
            "environment": "emergency_department",
            "time": "03:00",
            "occasion": "patient_triage",
        },
    },
)

5.5 — Privacy and Context Opacity

GOVERNANCE Personal context is sensitive. VCP provides opacity controls:

• Context never leaves the evaluation boundary — the LLM provider doesn't see raw personal data
• Granularity control — users choose which dimensions to share (all, some, none)
• Source transparency — the source field distinguishes user-reported data from inferred data, with confidence scores
• No persistence by default — personal context is ephemeral unless the user opts into continuity
• Decay over time — context signals marked as decayed indicate they were set in a previous interaction and may no longer be accurate

Context adaptation is what separates a value-aligned AI from a rule-following AI. The same values, applied with contextual awareness, produce qualitatively different — and more appropriate — responses.