Telco Support Triage
Overview
Use the shared support console API to fetch the records needed for each task, then classify items and produce structured JSON output matching the provided answer template. Every task follows the same pattern: read the payload file(s), call the relevant API endpoints for each listed identifier, map the evidence to classifications using the decision rules below, and output JSON that conforms exactly to the answer template.
Never guess. Fetch every record from the API. If an API call fails or returns no data, treat that as actionable evidence (e.g., invalid account, auth failure).
Workflow
- Read the prompt, the payload file(s) under the input directory, and the answer template.
- Identify every entity ID in the payload (ticket IDs, case IDs, account IDs, incident IDs, customer IDs, line IDs, bill IDs).
- Fetch the corresponding records from the API using the endpoint catalog in API Endpoints. Call each endpoint exactly as listed.
- Apply the classification rules in Classification Rules to map API evidence into the template enum choices.
- Output only valid JSON that matches the answer template exactly. Preserve payload ordering for arrays. Quote all string values. Use numeric types for integers and decimals as specified.
API Guidance
The base URL is always the task environment variable <TASK_ENV_BASE_URL>.
Substitute it literally.
Fetch every entity record before drawing conclusions. When a record is absent or returns an error:
- Ticket calls returning 404 or empty means the account is invalid.
- Customer calls returning 404 means no matching customer.
- Enterprise calls: use the complaint email approximate reference to look up the real incident, then cross-reference with messages and export-runs.
For the full endpoint catalog see API Endpoints. For how to map API evidence to classifications see Classification Rules.
Task Types
The API supports five categories of task. Match the payload structure to the right category:
Ticket batch — CSV of ticket_id rows with account_id, reported_service_type, and a customer report or queue note. Resolve each ticket by pulling ticket, diagnostic, outage, and account records. Output ticket_decisions with resolution statuses and escalation routes.
Case queue (mobile support) — JSON array of cases with case_id and reported_issue. Pull customer, line, bill, plan, and device records. Assign a primary and secondary action from the mobile-action enum. Output case_decisions.
Enterprise response package — Complaint email plus response_requirements.json. Pull the enterprise account, incident, export-runs, messages, and SLA records. Fill in root cause, failure window, owners, severity, permissions, and response status.
Queue quality review — CSV of tickets with a queue_note column. Classify using the key-blocker and route-team enums instead of the full resolution route. Output ticket_decisions with key_blocker and route_team.
Mobile-data recovery worklist — JSON array of cases with reported_issue plus customer_preferences. Pull customer, line, bill, plan, and device records. Focus on data-specific actions (refuel, roaming, data-saver, network mode, mobile data toggle). Include data_refuel_gb and charge_amount_usd when applicable.
Important Rules
- Preserve payload ordering: output arrays must list items in the same order they appear in the input payload.
- The answer template is authoritative for output shape and enum values. Always read the template before output.
- When a field requires an empty string, an explicit NONE, or 0.0, use exactly that value — not null or omission.
- For summary fields, count items in the decisions array; do not derive from external sources.
- The tickets_requiring_customer_wait summary field counts tickets whose resolution requires the customer to wait for an external event (e.g., outage resolution).