Cedar Ridge Intake Reconciliation

Cedar Ridge Intake Reconciliation

Core Workflow

1. Read the user's prompt and every file under input/payloads/, especially answer_template.json.
2. Read environment_access.md when present and use only its base URL and listed endpoints for network access. Substitute any <TASK_ENV_BASE_URL> placeholder with that base URL. Credentials are normally absent unless the file says otherwise.
3. Identify the target object from the prompt/template: roster, referral batch, transfer batch, program code, or explicit patient/referral/transfer IDs.
4. Query the portal for the complete target cohort before reasoning. Prefer POST /query for compact joins, and use detail endpoints such as /patients/{id}, /referrals/{id}, /transfers/{id}, /chart/{patient_id}, /icd/{code}, /documents, and /pharmacies to confirm ambiguous fields.
5. Build the JSON strictly from the template: required top-level keys, controlled values, nullability, ordering rules, and summary/count keys. Do not emit prose outside JSON when the task asks for JSON only.
6. Derive every ID, date, count, and row from the current prompt/template/environment. Never reuse literal values from examples or prior tasks.

Useful portal tables exposed through read-only SQL include patients, intake_rosters, coverage, pbm, patient_pharmacy, pharmacies, lifestyle, clinical_history, referrals, icd_codes, documents, transfer_requests, facility_capacity, program_candidates, and chart_artifacts.

General Validation

• Treat answer_template.json as the output contract even when endpoint data uses different field names.
• Sort each list exactly as the template specifies. Common orders are ascending patient_id, referral_id, transfer_id, or group_id; priority lists use the template's priority definition.
• Treat reason-code, issue-code, and blocker-code arrays as unordered sets unless the template gives an ordering.
• Recompute summary counts from the completed item rows, not from separate assumptions.
• Include empty arrays and zero-valued count buckets when the template requires them.
• Use null only where the template permits it.

Access Verification

For new-patient roster verification, join the roster to patient identity, coverage, PBM, preferred pharmacy, lifestyle, and clinical history records.

• Use the roster row for requested_service_date and service_line.
• Mark insurance valid only when a coverage record is active for the requested service date and includes the requested service line. Flag expired, pending, missing, or excluded-line cases with the template's matching reason codes.
• Mark prescription benefits valid only when PBM data is present, active, approved, and covered. Rejected, pending, inactive, not-found, review, missing, payer/policy mismatch, or specialty-policy conflicts should map to the closest PBM reason code allowed by the template.
• Use the patient's first preferred pharmacy by preference_rank; map its pharmacy network_status to in-network/out-of-network/unknown.
• Score lifestyle risk from smoking, alcohol, exercise, and sleep. Current smoking, heavy alcohol, absent exercise, or very low sleep increase risk; multiple adverse factors usually become high.
• Score overall risk from lifestyle risk plus clinical burden such as high medication count, recent hospitalization, risk flags, and major access blockers.
• Registration status should follow blocker severity: hard coverage/service-line conflicts reject, remediable administrative or benefit issues hold or clinical-review, high clinical risk routes to clinical review, and clean low-risk rows approve.

Referral Readiness And Activation

For referral batch audits, query the target batch and reconcile each referral with ICD metadata, patient/chart information, documents, authorization fields, appointment fields, duplicate patterns, and shared insurance IDs.

• Use /icd/{code} or icd_codes to compare referral service_line with ICD service_family, expected chapter, narrative, and laterality.
• Clinical code discrepancies include wrong service family/chapter, mismatch between referral reason or diagnosis text and ICD meaning, and laterality conflicts.
• Records and imaging blockers come from records_received and imaging_received.
• Authorization blockers apply when authorization is required and the status is denied, pending, not submitted, or otherwise not cleared.
• Existing scheduled appointments before clearance should be flagged with the template's appointment/hold reason.
• Duplicates usually share patient, insurance, referring contact, reason, or a "duplicate" note. Keep the earliest or primary referral unless the task gives another rule.
• Shared insurance anomalies are shared insurance IDs across distinct patients; shared insurance on duplicate referrals for the same patient is usually a legitimate duplicate.
• Readiness precedence: no issues is ready; records/imaging/auth blockers are blocked; clinical-code, duplicate, or scheduled-before-clearance issues are under review unless combined blockers make the template call for blocked; pure administrative insurance/contact issues are administrative follow-up.
• Map issues to action or correspondence codes using the template vocabulary. Urgent clinical discrepancies are highest priority, clinical/admin blockers usually follow, and pure administrative verification is lowest.

Dialysis Transfer Packet Review

For transfer batches, gather transfer requests, packet documents, patient records, and facility capacity for the requested start date and modality.

• A required packet item is complete only when the document exists for the transfer/patient and is finalized/final. Draft, absent, or non-final documents count as missing.
• For stale checks, compare received_date with the requested start date. Common freshness windows are 30 days for current dialysis labs/infectious-screening/PPD-CXR style items and 365 days for history and physical; use task/template wording if it gives a different limit.
• Capacity is available when the sum of facility_capacity.open_chairs for the requested date and modality is greater than zero.
• Feasibility combines packet readiness and capacity: ready on requested start only when the packet is complete/fresh and capacity is available; otherwise choose the template value matching packet-not-ready and/or capacity-unavailable.
• Intake decision should accept only fully ready transfers with capacity. Missing or stale clinical packet items usually require clinical review; capacity-only issues route to scheduling or hold; administrative-only missing items route to intake coordination.

Chronic-Care Program Panels

For program candidate panels, load /programs/{program_code}/candidates, then join each candidate to patient, chart artifacts, and clinical history.

• Eligibility depends on the program target condition and active clinical history. For diabetes/hypertension panels, require both active diabetes and hypertension evidence unless the task defines another target.
• Consent declined usually rejects. Missing consent or inactive/missing chart data usually holds unless combined with wrong target condition, which rejects.
• Missing chart artifacts come from absent or stale chart_artifacts rows for the artifact types required by the template. Treat status != current as needing refresh.
• High-touch monitoring is appropriate for recent hospitalization, recent ED/risk flags, very low adherence, or other high-risk clinical markers. CKD commonly routes to biweekly monitoring. Stable eligible candidates can use monthly or standard monitoring.
• Outreach should follow the candidate or patient preferred channel when usable; fall back only when the preferred channel has no usable phone/email/portal information and the template permits another channel.
• Initial monitoring packages should match disposition: high-touch for high-risk enrollment, standard for stable enrollment, deferred for holds needing consent or chart updates, and not applicable for rejects.

Output Discipline

Before finalizing, parse the JSON locally if possible and check:

• All required keys from the template are present.
• All controlled values are from the template.
• Every required target row is included exactly once.
• Summary counts equal the row-level data.
• No task-local answer constants or example rows have been copied from prior work.