Single tool with critical definition gaps. Tool name 'arbitrate_dispute_case' lacks action verb prefix (should be 'arbitrate_dispute' or similar). Description is present (67 chars) but minimal and lacks WHEN/WHY context. Input schema visible with 5 parameters, all with types and descriptions, but descriptions are terse (avg 45 chars, below 72-char baseline). No output schema documented, LLM cannot plan downstream actions. No error handling guidance. Tool performs state-modifying arbitration (verdict + settlement action) but description does not indicate this is a consequential decision. Parameters lack constraints (e.g., claim_reason is free-form string, not enum of valid claim types). No pagination, no idempotency guidance.

• Document the complete output schema with all fields (arbitration_id, order_id, evidence_proof_hash, days_since_delivery, within_policy_window, eligibility_score, arbitrated_verdict, recommended_settlement_action) and their types/descriptions.
• Convert claim_reason to an enum parameter with valid values: ITEM_DEFECTIVE_MOTOR_STALL, ITEM_NOT_RECEIVED, ITEM_NOT_AS_DESCRIBED, DAMAGED_IN_SHIPPING, WRONG_ITEM_SENT. Add description: 'The category of dispute claim. Affects eligibility scoring.'
• Expand tool description to 100-150 chars explaining the arbitration logic: 'Evaluates refund disputes using carrier tracking, delivery timeline, and photographic evidence. Returns a binding verdict (AUTO_REFUND_APPROVED, PARTIAL_STORE_CREDIT_OFFERED, or DISPUTE_REJECTED) and recommended settlement action.'
• Add parameter constraints: days_since_delivery (1-365), tracking_delivered (boolean, no default needed), photo_proof_url (valid HTTPS URL or null). Document that score >= 90 → auto-refund, 60-89 → partial credit, <60 → rejected.
• Add error handling: validate photo_proof_url format, return clear error if unreachable, and document the fallback behavior. Include recovery guidance: 'If photo evidence is unavailable, dispute may be rejected. Provide valid URL or contact support.'