Payer Operations Review
Follow this general workflow for any payer-operations review task.
Step 1: Orient to the task
Read the user's prompt, the payload answer template, and any task context payload. The task context gives you the target business ID, reporting date, and domain notes. The answer template defines the exact JSON shape.
Identify:
- The root business record type (case, claim, appeal, P2P, queue)
- The payer domain (utilization management, pharmacy, payment integrity, finance)
- Template enums that constrain your choices
Step 2: Discover the environment schema
The environment runs at the base URL from the prompt. The SQL endpoint is POST /sql/query and uses the bearer token provided. Read references/environment-interaction.md for endpoint details and interaction patterns.
Start by listing all tables:
SELECT name FROM sqlite_master WHERE type='table' ORDER BY name
Follow with a schema inspection of each table and spot-checks of sample rows. Also use REST endpoints to pull records by known IDs.
Step 3: Pull every record that could matter
Once you know the schema, retrieve all records linked to the target business ID. Use both SQL and REST endpoints as appropriate.
Cover these record types as they appear in the schema:
- Case records -- member, plan, provider, request lines, status
- Policy/criteria records -- criteria definitions for the service domain
- Clinical documents -- evaluation findings, trial results, P2P transcripts
- Authorization records -- auth numbers, approved units, dates, CPT codes
- Rate schedules / benchmarks -- rate rows for relevant CPT codes and dates; check which is current versus stale
- Appeal records -- status, deadline, packet requirements
- Drug trial / formulary records -- documented failure rows for the target drug
- Queue rows -- only the rows listed in the task context
- Claim lines -- line-level paid amounts, units, modifiers
Pull broadly, then narrow. Do not stop at the first record you find.
Step 4: Evaluate against policy criteria
For each applicable criterion from the policy records:
- Read the criterion definition
- Check whether clinical documents satisfy it
- Classify each criterion as
met,not_met,unclear, ornot_applicable
When all criteria are met: the default is approve.
When any criterion is not met or unclear: escalate or deny according to the policy. Document which criteria failed and what gap each represents.
Appeal tasks: separate documented failures (drugs with clear trial records) from undocumented or insufficient failures (drugs mentioned but without adequate trial documentation).
Payment integrity: compare rate-schedule versions and effective dates. Use the current schedule; reject the stale one. Apply unit-level pricing to each claim line.
P2P tasks: the central question is whether the P2P discussion introduced new patient-specific information that materially changed the review. If it did not, pre-P2P criteria evaluation controls the result. Flag any PET-specific factors that remain unsupported.
Finance queue tasks: compute total_cost = variable_cost + fixed_cost_allocated, margin = revenue - total_cost, and revenue_to_cost_ratio = revenue / total_cost. Flag rows below the revenue-to-cost threshold and rows marked as charge sensitive.
Step 5: Construct the basis audit
Every answer must include a basis_audit with four required keys: source_precedence, controlling_record_ids, exception_record_ids, and precedence_record_order.
Read references/basis-audit.md for the full specification, including the six precedence rules, how to select the correct one, and the construction rules for each audit field.
Step 6: Produce the JSON answer
Follow the answer template exactly:
- Date format:
YYYY-MM-DD. Usenullonly when the template allows it for date fields. - Currency: JSON numbers in USD, rounded to two decimal places. No currency symbols.
- Modifiers: Use
null(not empty string) for absent modifiers. - List ordering: Follow the ordering rule in the template for each list field. When not specified, default to ascending alphabetical or ascending ID order.
- Enums: Use exactly one allowed choice. Do not invent variants.
- Nested objects: Match every required key. Do not omit required keys even if their value is an empty list.
- No extra fields: The answer template defines the allowed surface. Do not add keys beyond what the template specifies.
Step 7: Validate before returning
Before returning, check:
- Every required top-level key from the template is present
- Every enum value is one of the allowed choices
- List fields follow their stated ordering rule
- Currency values are numbers (not strings) rounded to two decimals
- No extra fields beyond the template's allowance
- The
basis_audithas all four required keys with the correct precedence rule - Controlling and exception records are properly separated -- nothing appears in both lists
Return only the JSON object. Do not wrap it in markdown, add prose, or include commentary.