Northstar Payer Ops

Northstar Payer Ops

Workflow

1. Read the prompt, input/payloads/task_context.json, and input/payloads/answer_template.json first.
2. Identify the task family from the template keys.
3. Use the base URL and bearer token from the task context or prompt.
4. Query the live Northstar environment with POST /sql/query; use business endpoints when they expose the needed record directly.
5. Build the answer only from live records plus the task memo for scope and dates. Treat the memo as a locator, not as evidence.
6. Follow the template exactly. Preserve list order rules, enum spellings, number precision, and date formats. Emit one JSON object only.
7. Write basis_audit from the records that directly controlled the result. Put controlling records first, then gaps, stale records, or exclusions.

Shared Rules

• Prefer current clinical, appeal, or benchmark records over stale exports or old schedules.
• Keep null for absent modifiers.
• Use ISO dates (YYYY-MM-DD) and round currency to cents.
• Keep ratios at the precision requested by the template.
• When a task limits scope to explicit IDs or row IDs, ignore everything else.
• Do not add narrative, markdown fences, or extra keys.
• Use the environment endpoints listed in the task note when they are clearer than SQL: GET /api/tables, GET /api/cases, GET /api/cases/{case_id}, GET /api/policies, GET /api/policies/{policy_id}, GET /api/documents/{document_id}, GET /api/rate-schedules, and GET /api/appeals.

Task Families

Authorization summaries

• Use when the template requires case_id, recommendation, final_status, route, authorization, criteria_results, evidence_documents, excluded_documents, determination_letter, and next_action.
• Review the case, member and plan context, requested therapy lines, applicable policy criteria, clinical documents, and any authorization record.
• Map outcomes consistently: approve when the required criteria and authorization support the request; pend when key information is missing; escalate to medical director or peer-to-peer when physician review is needed; deny when criteria fail; use partial approval when only part of the request is supportable.
• Populate criteria_results for PT-ACTIVE, PT-DEFICIT, PT-DX, PT-POC, and PT-UNITS.
• Include only controlling evidence documents in evidence_documents; list stale or noncontrolling case documents in excluded_documents.
• Fill authorization with the auth number, approved units, approved date span, approved CPTs in ascending order, and the line modifier.
• Set source_precedence to current_clinical_records_over_stale_export.

Appeal and assistance intake

• Use when the template requires case_id, appeal_id, drug, appeal_path, expedited, appeal_deadline, owner, medication failure lists, packet items, assistance, and next action.
• Resolve the appeal first, then the assistance screen. Let appeal records control over assistance facts when they conflict.
• Classify prior medication failures into documented_failures and undocumented_or_insufficient_failures using the evidence in the case.
• Order packet items so appeal evidence appears before assistance items; list gaps in the same operational order.
• Mark assistance eligible_ready only when the program is identified and no required fields are missing.
• Set source_precedence to payer_appeal_before_manufacturer_assistance.

Claim repricing

• Use when the template requires claim_id, case_id, auth_number, benchmark fields, totals, line-level corrections, resubmission route, priority, and audit.
• Select the benchmark by plan, modifier, and date. Reject stale exports and note the rejected source.
• Keep claim lines in source claim order.
• Compute paid_total, correct_allowed_total, and recovery_amount from the line items. Use cents precision.
• Set each line disposition to correct_upward, correct_downward, no_change, or deny_line based on the corrected allowed amount.
• Set source_precedence to effective_benchmark_by_plan_modifier_and_date.

Peer-to-peer summaries

• Use when the template requires case_id, p2p_id, requested_cpt, p2p_outcome, final_status, PET criteria, unresolved criteria, missing PET factors, letter_type, recommended_alternative, and internal_appeal_deadline.
• Decide whether the P2P added new patient-specific information that materially changed the review.
• Map criteria_results to PET-IND and PET-FACTOR.
• List only unresolved criteria in unresolved_criteria.
• When the final result is adverse, calculate the internal appeal deadline from the final adverse determination date using the plan window in the task memo.
• Set source_precedence to new_patient_specific_p2p_information.

Margin queue summaries

• Use when the template requires case_id, period, threshold ratio, row summaries, segment lists, top issue, gap, and audit.
• Review only the row IDs listed in the task memo.
• Compute each row's ratio from the underlying revenue and cost figures, then flag below_threshold at the requested cutoff.
• Flag charge_sensitive from the queue facts, keep row order aligned with the supplied row IDs, and sort the segment summary lists alphabetically.
• Set top_issue to the worst below-threshold segment/CPT pair and compute gap_to_120pct against 120 percent of cost.
• Set source_precedence to margin_threshold_then_charge_sensitivity.

Audit Format

• Put the directly controlling record IDs in controlling_record_ids.
• Put missing, stale, exception, or excluded records in exception_record_ids.
• List the combined record trail in precedence_record_order, highest priority first.