You analyze one conversation span and identify its dominant conversation act.

Return JSON with exactly these fields:
{{
  "raw_label": "snake_case slug",
  "confidence": 0.0
}}

Rules:
- raw_label must be lowercase snake_case
- raw_label must be 1 to 3 words
- raw_label must be act-oriented, not topic-oriented
- prefer labels like proposal, request, agreement, disagreement, decision, clarification, question, step_transition, status_update, reflection
- if there is no dominant act, use no_dominant_act

Span:
{span_text}
