Support Console Resolver
How to navigate a telecom support console API, resolve tickets, cases, and enterprise incidents from records rather than assumptions, and produce decisions that conform to a provided answer template.
Core Principles
Every decision is an API lookup, never a guess. The reported issue text in a payload gives you a starting point, but the resolution must come from the API records. When you see a ticket, case, or incident identifier, fetch its full record. Then follow every cross-reference it contains.
Parallelize all independent fetches. When you have a batch of N items, fetch all individual records, diagnostics, and related resources simultaneously in one wave. Serial fetches waste time and risk hitting rate limits unnecessarily late.
Conform to the template, not to habit. Every task provides an
answer_template.json with strict fields, enum values, and ordering
requirements. Read it before you write the output. Enum values in the output
must appear exactly as written in the template. Do not invent synonyms.
How to Approach Any Task
- Read the prompt to understand the role and identify the task type.
- Read every payload file. Note which entity IDs are present and which cross-references you will need to follow.
- Read the answer template. Note every field, its type, its allowed enum values, and any ordering constraints.
- Identify the relevant API endpoints for this task type using the API Endpoints reference.
- Fetch all primary records in a single parallel wave: every individual ticket, case, or incident in the batch, plus any list endpoints (outages, export-runs, messages) that cover all items.
- From the primary records, extract every cross-reference (account_id, customer_id, line_id, bill_id, etc.) and fetch all secondary records in a second parallel wave: diagnostics, troubleshooting records, customers, lines, bills, plans, devices, SLA records, and enterprise accounts.
- For each item, apply the decision rules in Resolution Patterns to map the API evidence to template fields.
- Build the output JSON matching the template structure, field order, and enum constraints exactly.
- Compute summary counts by tallying the decisions you already made; do not calculate them independently.
Task Types and Their API Footprints
The prompt and payloads signal which workflow applies. A single task may combine patterns; when it does, apply each pattern to the corresponding subset of items.
Ticket Resolution
Prompt signals: "support operations analyst", "queue-quality analyst", "service tickets", "ticket batch", "queue snapshot", "SLA handoff", "same-day batch", "classify each ticket"
Payloads: CSV with columns that include ticket_id and account_id
(typically also reported_service_type, customer_report, or queue_note).
API footprint: Tickets, diagnostics, troubleshooting, outages. Use API Endpoints.
Decision logic: Resolution Patterns.
Contact-Center Case Triage
Prompt signals: "contact-center lead", "mobile support queue", "case queue", "choose the support operation"
Payloads: JSON with a cases array containing case_id and
reported_issue. May include customer_preferences.
API footprint: Cases, customers, lines, bills, plans, devices. Use API Endpoints.
Decision logic: Resolution Patterns.
Enterprise Incident Response
Prompt signals: "enterprise support lead", "export complaint", "structured response package", "client export complaint", "formal response"
Payloads: Complaint email (.txt) with client identity and incident
reference, plus response requirements (.json) with field list, permission
users, and naming conventions.
API footprint: Enterprise incidents, accounts, export-runs, messages, SLA. Use API Endpoints.
Decision logic: Resolution Patterns.
Mobile Data Recovery
Prompt signals: "mobile-data recovery analyst", "data recovery", "mobile data worklist", "data refuel", "carrier update"
Payloads: JSON with a cases array plus customer_preferences mapping
case IDs to accepted refuel amounts and plan-change preferences.
API footprint: Cases, customers, lines, devices, plans. Use API Endpoints.
Decision logic: Resolution Patterns.
Domain Model Overview
The console models a multi-tenant telecom provider. Each entity carries cross-references you must follow to reach a decision. The full reference is at Domain Model.
Key relationship chains to follow:
- Ticket to
account_idto diagnostic, troubleshooting, outage matching - Case to
customer_id+line_idto Customer to Line to Bill, Plan, Device - Enterprise Incident to
enterprise_account_idto export-runs, messages, SLA, account details
When an API response includes an ID you do not yet have, fetch that resource. Empty or null IDs signal that relationship does not apply.
Parallel Fetch Strategy
Structure your API calls in two waves:
Wave 1 (primary records): Every ticket, case, or incident listed in the
payload, fetched by individual GET. Also fetch the list endpoints that serve
all items: /api/outages, /api/enterprise/export-runs,
/api/enterprise/messages.
Wave 2 (secondary records): Every diagnostic, troubleshooting record, customer, line, bill, plan, device, SLA, or enterprise account referenced by Wave 1 results. Gather all IDs first, then fetch everything in one parallel batch.
After Wave 2 you have all the evidence needed to make decisions. Do not make additional API calls during decision-writing unless the template demands a field that none of your fetched records covers.
Output Rules
- Preserve item order as directed by the template. Common directives are "preserve payload order" (CSV row order) or "preserve ascending case_id order." Follow them literally.
- Use exact enum strings. If the template defines
RESOLVED | PENDING_ACTION | ESCALATED | FAILED, do not writeClosed,Pending, or any other variant. - Empty strings for absent references. Use
""for fields likeoutage_id,bill_id, orescalation_teamwhen no value applies. - Zero for absent numerics. Use
0.0or0forcharge_amount_usd,data_refuel_gb, or similar fields that do not apply to the decision. - Boolean false for absent conditions. Set
diagnostic_needed,latency_issue,stability_issue,bandwidth_issue, andcarrier_update_requiredtofalsewhen the API evidence does not support them. - Derive summary counts from the decisions array. Tally the actual
final_resolution_status,final_route, or equivalent category values in the decisions you produced. Do not estimate or compute summaries independently; that guarantees drift. - Match summary field names to the template. Some templates name the
summary object
batch_summary, othersqueue_summaryorworklist_summary. Use the name the template provides.
Common Pitfalls
- Resolving from the reported issue text instead of the API. The payload text hints at which endpoints to query; it is not the resolution itself.
- Fetching records one at a time. Batch all independent GETs into parallel waves.
- Stopping at the first record. A ticket's
account_idleads to account validation; a case'sline_idleads to the line, the plan, and the device. Follow every cross-reference the API gives you. - Inventing enum values not in the template. The template is the schema contract. Read it before writing output.
- Computing summaries that do not match the decisions. Count from the decisions array you built; never recalculate.
- Using a non-existent base URL. The base URL comes from the prompt
(often as
<TASK_ENV_BASE_URL>or a specific host:port). Do not hardcode it. If the prompt uses a placeholder, substitute it from the environment.