Sentiment Priority Scorer RE-India

Sentiment Priority Scorer

Produce deterministic priority scores for leads without mutating any state.

Quick Triggers

• Rank leads by urgency and tone for callback priority.
• Classify leads into P1/P2/P3 queue.
• Score follow-up priority from normalized lead records.

Recommended Chain

india-location-normalizer -> sentiment-priority-scorer -> summary-generator

Execute Workflow

1. Accept input from Supervisor containing normalized leads.
2. Validate input with references/sentiment-priority-input.schema.json.
3. Score each lead with:
   • sentiment_score in range [-1, 1]
   • intent_score in range [0, 1]
   • recency_score in range [0, 1]
   • mapped urgency_score from lead urgency (high=1.0, medium=0.6, low=0.3)
4. Use record_type to avoid over-prioritizing generic bulk inventory:
   • buyer_requirement: apply +0.10 intent lift (hard demand signal)
   • inventory_listing: no lift unless high-action cues are present
5. Boost intent_score when high-action cues exist in listing text:
   • immediately, keys at office, one day notice, possession, inspection any time
6. Compute priority_score on a 0-100 scale:
   • priority_score = 100 * (0.40*urgency_score + 0.30*intent_score + 0.20*recency_score + 0.10*sentiment_risk)
   • sentiment_risk = max(0, -sentiment_score)
7. Assign buckets:
   • P1 for priority_score >= 75
   • P2 for priority_score >= 50 and < 75
   • P3 for < 50
8. Produce plain-language evidence tokens that explain the score, including record-type evidence.
9. Validate output with references/sentiment-priority-output.schema.json.

Enforce Boundaries

• Never write to Google Sheets, databases, or files.
• Never send messages or trigger outbound channels.
• Never create reminders or execute actions.
• Never bypass Supervisor routing or approvals.
• Never replace upstream urgency; only derive scoring fields.

Handle Errors

1. Reject schema-invalid inputs.
2. Return field-level reasons when scoring cannot be computed.
3. Fail closed if required scoring features are missing.