Gap Report

Purpose

Give the coordinator an explicit, actionable missing-evidence list and the draft packet, after criteria-extract has run.

When to Use

• As the last skill in the chain on every run - including when gaps is empty, to confirm the all-clear explicitly.
• Standalone asks once a criteria_result exists: "what's outstanding", "what do we still owe the plan", "assemble the packet for review".

When NOT to Use

• criteria-extract has not run (there is no criteria_result), or the user asks whether the patient meets the criteria or why the plan would push back - use criteria-extract instead.
• The user wants the evidence found rather than the gap named - use evidence-match instead.
• The user wants the justification written - use necessity-draft instead.
• The user wants a general Word report or document - use docx instead.
• The payer already denied the claim - use the denials-appeals-response plugin instead.
• The user wants the packet sent, filed or uploaded to a payer portal - that action does not exist in this package and the request is declined.

Inputs

criteria_result.gaps (id, text, evidence_key, reason, clause citation); criteria_result.criteria_met; criteria_result.determination; necessity_narrative; packet; escalations.

Workflow

1. Read criteria_result.gaps, criteria_result.determination and packet from the contract produced by the engine (tool: python3 base/scripts/criteria_engine.py --question <contract.json> --policy config/payer-policy-library.json --out <contract.out.json>, run by criteria-extract).
2. For every entry in gaps, write one table row: criterion id, policy clause, evidence_key, reason, and the evidence needed, restated from the criterion text in the policy library.
3. If gaps is empty, state explicitly that all required criteria are evidenced and cited - never leave the all-clear implicit.
4. If determination is out_of_scope, report that no checklist could be assembled and carry the payer-relations escalation forward instead of an empty all-clear.
5. List met criteria with their citations, then every escalation word for word.
6. Present the packet block exactly as the engine wrote it - status draft, submitted false.

Output Format

A markdown table, one row per gap: | # | Criterion | Clause | evidence_key | Reason | Evidence needed | or the line All required criteria are evidenced and cited (<n> of <n>).; then Met: bullets with citations; then Escalations: bullets; then the packet JSON exactly as on the contract. The first line always states Not ready - <open_gaps> open gap(s) or No open gaps; this is never an approval.

Example

Not ready - 1 open gap. This is a draft for review, not an approval or denial.
| # | Criterion | Clause | evidence_key | Reason | Evidence needed |
| 1 | conservative_therapy_6wk | RAD-7.2 | conservative_therapy_6wk | no_evidence | Documented >=6 weeks conservative therapy (PT and/or NSAIDs) with no improvement, cited with note, date and quote |
Met: no_red_flags, exam_findings - Spine clinic intake note, 2026-09-12
packet: {"status": "draft", "submitted": false, "determination": "packet_assembled_gaps", "clause": "RAD-7.2", "criteria_met_count": 2, "open_gaps": 1}

Grounding requirements

Every listed gap matches an entry in criteria_result.gaps verbatim - nothing is added or summarised out. The "evidence needed" column restates the criterion text from the policy library; it never names an invented document, deadline, payer instruction or readiness percentage. Counts (open_gaps, criteria_met_count) are read from the engine, never recounted by the model.

Guardrails

• Action level L1 (draft only): assembles a packet for human review; L2 submit/file/upload actions are not available in this package.
• Never omit a gap to make the packet look cleaner.
• Never soften a gap list into "basically ready" - the specific evidence_key and clause are always named.
• Never report a gap as resolved because someone says the document will arrive later.
• Never change packet.status from draft or packet.submitted from false (prior-auth-rules.md #5).
• Never present the packet as an approval, denial or clearance, whatever the determination value (prior-auth-rules.md #2).

Error Handling

• If criteria_result is missing: stop and hand back to criteria-extract; do not build a gap list by reading the policy yourself.
• If determination is out_of_scope: report it as such with the payer-relations escalation (prior-auth-rules.md #7), not as an empty all-clear.
• If a remaining gap needs a document: route it to the prior-auth coordinator role in config/escalation-contacts.json; if that contact is still [TO CONFIGURE: ...], say so rather than naming anyone.
• If the user asks to upload or submit: decline, and keep the packet as a draft.