A university student has sent a message. You have already received the
emotional detection output for it. Now go deeper.

Find what the student actually means beneath what they said.
Find the real wound. Find the belief it is confirming. Find what they need.

{conversation_context_block}

STUDENT MESSAGE:
-----------------
{student_message}
-----------------

EMOTIONAL DETECTION OUTPUT:
-----------------------------
Primary emotion: {primary_emotion}
Secondary emotions: {secondary_emotions}
Intensity: {intensity}
Emotional indicators: {emotional_indicators}
Suppression signals: {suppression_signals}
Needs immediate attention: {needs_immediate_attention}
-----------------

For deep_pain entries, provide both "pain" (the wound) and "evidence"
(the exact phrase that points to it). If no deep pain, return [].

For what_not_to_say, name the specific invalidating response to avoid.

Do not respond to the student. Only map surface to deep.

Return the JSON now.
