Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/sputhenofficial/claimjumper/llms.txt

Use this file to discover all available pages before exploring further.

ClaimJumper uses GPT-5.6 Sol to decide the correct resolution path for each denied service line on a remittance. That decision is called a lane and it drives everything that comes next: what artifact (if any) the tool produces, whether the billing team can approve and export it, and what manual steps they must perform. Every line is assigned exactly one lane before the pipeline moves on to prioritization and drafting.

The five lanes

What it meansThe denial is clinically or procedurally disputable through the payer’s formal reconsideration process. Common triggers include a medical-necessity determination the provider believes is incorrect, a timely-filing denial where clearinghouse acknowledgment evidence exists, or a coverage decision tied to a specific LCD.What triggers this laneGPT-5.6 Sol reads the Group Code, CARC, and RARC together. Lanes such as the following are typical appeal candidates:
  • CO / CARC 50 / RARC N115 — medical necessity denied against LCD criteria
  • CO / CARC 29 — timely-filing denial with available submission evidence
What ClaimJumper producesA cited appeal draft — a letter body requesting reconsideration that includes the exact group code, CARC, and RARC. Every factual statement in the body is backed by a citation pointing to the parsed remittance data or the code legend.What the billing team does nextValidate payer-specific requirements and supporting records, then follow the payer’s appeal process. The draft is a starting point for human review — not a ready-to-file document.Approvable and exportable? ✅ Yes — eligible for one-click approval and inclusion in the downloaded citation bundle.
What it meansThe denial was caused by missing or incorrect information in the original claim submission. The line should be corrected and resubmitted rather than formally appealed. Attempting an appeal on a correctable claim wastes a reconsideration opportunity.What triggers this laneMissing-information denials are the canonical trigger. For example:
  • CO / CARC 16 / RARC M51 — claim/service lacks information needed for adjudication; the RARC names the specific missing element
The triage prompt explicitly instructs GPT-5.6 Sol: “Missing-information denials (e.g. CARC 16) are corrected claims, NOT appeals. The paired RARC names the missing element.”What ClaimJumper producesA cited corrected-claim worklist — a list of the exact fields to correct, with no appeal language. Citations point to the parsed claim and code legend so the biller can verify each item against the source remittance data.What the billing team does nextVerify each correction against the original records, then prepare the replacement claim in the billing workflow. Do not submit the worklist directly to the payer.Approvable and exportable? ✅ Yes — eligible for one-click approval and inclusion in the downloaded citation bundle.
What it meansThe service line carries a Patient Responsibility (PR) group code, meaning the amount has been calculated as patient cost-share (deductible, coinsurance, or copay) rather than a provider write-off or a disputable denial.What triggers this lanePR group codes route here by default unless a RARC contradicts the patient-responsibility assignment. The billing team then follows their organization’s patient-billing policy.What ClaimJumper producesDisposition only — no generated letter. The queue shows the lane and the triage rationale, but no draft artifact is created.What the billing team does nextVerify patient responsibility and follow the organization’s patient-billing policy for issuing a statement or applying it to the patient’s account.Approvable and exportable? ❌ No — patient-bill lines cannot be approved or exported.
A CO (contractual obligation) group code can never be routed to patient_bill. This rule is enforced in code by enforceInvariants() and cannot be overridden by the model or through configuration. See Safety Invariants for details.
What it meansThe denial represents an amount the provider cannot recover and has no practical basis to dispute — for example, a contractual adjustment already reflected in the allowed amount.What triggers this laneGPT-5.6 Sol determines that neither an appeal nor a corrected-claim resubmission is viable and the amount must be adjusted off the account. Recovery probability is typically very low.What ClaimJumper producesDisposition only — no generated letter. The queue shows the lane and the triage rationale.What the billing team does nextObtain the required internal approval and follow the organization’s adjustment policy before posting the write-off.Approvable and exportable? ❌ No — write-off lines cannot be approved or exported.
CARC 197 denials can never be automatically written off. Because CARC 197 is a payer-responsibility denial, abandoning that revenue requires a deliberate human decision. The invariant escalates these lines to human_review. See Safety Invariants.
What it meansOne or more safety invariants fired after the model’s initial lane decision, or the model’s routing confidence fell below the automation threshold. The line cannot be handled automatically and needs a human to decide the next action.What triggers this laneThe three code-enforced invariants each escalate to human_review:
  1. CO group code with a model-suggested patient_bill lane
  2. CARC 197 with a model-suggested write_off lane
  3. Model confidence below 0.6
What ClaimJumper producesA deterministic human-review brief — plain text listing the claim ID, patient, service line, rationale, evidence needed, and the exact override reasons that caused the escalation. This brief is not model-generated; it is assembled directly from the validated pipeline data.What the billing team does nextRead the override reasons, then decide the correct next action using clinical, compliance, and payer-specific judgment.Approvable and exportable? ❌ No — human-review lines cannot be approved or exported under any circumstances.

Lane decision model

GPT-5.6 Sol acts as a healthcare reimbursement specialist for triage. It receives a structured AdjustmentLine object — group code, CARC, RARC, billed amount, description, notice date, and any notes — and produces a structured TriageModelOutput object. Critically, the model reads Group Code + CARC + RARC together, never a CARC in isolation. The output schema contains six fields:
FieldTypeDescription
laneenumOne of the five routing lanes
root_causestringThe specific denial category driving the decision
rationalestringExplanation citing the Group Code / CARC / RARC that drove the decision
evidence_neededstring[]What the biller must attach for the recommended lane to succeed
confidencenumber (0–1)How confident the model is in its lane assignment
recovery_probabilitynumber (0–1)Probability of successfully recovering the billed amount via this lane
After the model returns its output, enforceInvariants() runs synchronously and may override the lane to human_review. The model’s original lane and the override reasons are both preserved in the DenialTriage result so the billing team can see exactly what happened.
Code-enforced invariants can override the model’s lane decision after triage. These rules protect billing teams from incorrect automation and are covered by regression tests. See Safety Invariants for details.

Build docs developers (and LLMs) love