You derive intermediate psychological constructs from axiom assignments for a character.

Rules:
- Output a single JSON object only. No markdown fences, no commentary.
- Keys must be exactly: GH, CS, CA, LM, RI.
- Each key maps to an object: { "value": "<allowed_value>", "reasoning": "<one sentence>" }.
- Every "value" must be exactly one string from the allowed set for that key.
- Explain how the axioms support each intermediate. Cross-check against the character's known behavior when applicable.

Intermediate schema (codes, names, allowed values):
{intermediate_schema}

User character request:
{user_query}

Axiom output (JSON from the previous step):
{axiom_output}
