Nacovo Atlas · pre-submission policy validation

Check payer rules
before the claim goes out.

Atlas is a pre-submission rules check for medical claims. It converts explicit payer-policy requirements into deterministic, page-cited logic and returns PASS / REVIEW / BLOCK with the reason attached — while the claim can still be reviewed or corrected.

Same input → same verdict Material findings page-cited Reviewable by a human coder
verdict · reproducible audit trace
claim837P · outpatient MRI · facility NPI 1063███412
policyoxford-imaging-013B.62 v2026.1
BLOCK would be denied on submission — stop it here
ruleIMG.ACCRED.MISSING
clause§4.2 — facility must hold ACR / IAC accreditation for advanced imaging ↳ p.7
evidenceaccreditation roster: facility not listed
fixattach accreditation ID, or route to accredited site → re-run
dispositionBLOCK coverage 1 block1 review3 pass 2 not evaluated ⚠
trace sha256:9f2c…a71b · deterministic
Evaluates encoded policy requirements against claim-level data
Payer medical policy NCCI PTP edits LCD / NCD coverage Prior-auth criteria Payer pharmacy PA 837P claims 835 / ERA remittance NPPES + accreditation rosters
The policy execution gap

Policy is written for people. Atlas makes explicit requirements executable.

EHRs, coding tools, clearinghouses, and RCM platforms are essential parts of the revenue cycle. Their primary roles are documentation, code assignment, transaction routing, and workflow management. Atlas is designed to complement that stack with deterministic payer-policy evaluation before submission. AI can help structure policy and triage ambiguous cases; approved rules — not a model response — drive the claim verdict.

Existing revenue-cycle stack

Documentation, workflow & transactions

  • EHRs support clinical documentation and patient records
  • Coding systems support code assignment and validation
  • Clearinghouses support claim transport and connectivity
  • RCM platforms coordinate billing workflows and follow-up
// Atlas works alongside these systems
The Atlas engine

Policy as executable logic

  • Approved rules run deterministically — same governed inputs, same result
  • Material findings retain the clause, rule id, evidence, and source page
  • Known policy conflicts can be held before submission for correction
  • Ambiguous or missing evidence routes to human review rather than a guessed decision
// deterministic · reproducible · audit-grade
How it works

Encode, review, execute — with audit context attached.

Why now: current models can assist with reading dense, unstructured policy documents and drafting candidate rules, making policy normalization more practical. Human-approved, versioned rules remain the basis of claim-level execution.

01 / ENCODE

Payer PDF → reviewable logic

AI drafts a candidate rule from the source policy. It stays a candidate — unable to touch a verdict — until a reviewer with coding / RCM knowledge approves it Active, clause by clause, each tied to a page.

candidate → active gate · page-cited · versioned
02 / EXECUTE

Deterministic verdict

The engine runs the encoded policy against the 837 before submission and returns PASS / REVIEW / BLOCK. The verdict path uses explicit, deterministic logic.

same input → same output · zero drift
03 / PROVE

Audit trace, on demand

Material findings retain their audit chain: clause, rule id, evidence, source page, and a stable fingerprint that supports reproducible re-runs.

defensible · exportable · reconcilable to 835
From source policy to a verdict you can proveOxford imaging 013B.62 · v2026.1
Payer policy PDF
the source document, on file
AI reads
& drafts
Candidate rule
page-cited draft — cannot touch a verdict
◇ decision boundary CPC reviews
& signs
Active rule
signed · versioned · immutable
IMG.ACCRED.MISSING§4.2 · p.7 · verdict BLOCK
engine executes
on the 837
Verdict + trace
PASS / REVIEW / BLOCK · audit chain
◂ AI assists — drafting only deterministic — the engine decides ▸
Each active rule is represented as reviewable, page-cited data approved through a human gate. The same governed inputs and rule version produce the same verdict.
Honest by construction

It separates “did we check it?” from “did it pass?” — and shows you both.

Atlas keeps two questions separate: evaluation status — could the encoded rule run with the available data? — and rule result — what did it decide? Unevaluated rules do not become a PASS. Missing evidence, policy gaps, and unverified sources are identified for follow-up instead of being inferred. Policy normalization and keeping versions current remain important operational work, so coverage is stated explicitly rather than assumed.

evaluation_statuscan Atlas run this rule at all?
EVALUATEDthe rule ran to a result — a PASS requires this state
NOT_APPLICABLErule doesn't apply to this claim — not the same as passing
POLICY_GAPno executable policy or rule exists yet for this case
INSUFFICIENT_DATAclaim or evidence too thin to complete the check
SOURCE_UNVERIFIEDcan't confirm the policy source is still current and complete
EXECUTION_ERRORparser, rule execution, or data processing failed

Material findings answer six questions

Not just a label like “Medical Necessity Not Met.”

  1. What failed — the submitted value vs. the expected condition
  2. Why it failed — the exact reason, in plain words
  3. Which policy & rule caused it — cited to version and page
  4. What evidence is missing or was used
  5. Who should act next — coder, biller, or supervisor
  6. What actions are safe to take
// rule_result (PASS · WARNING · REVIEW · BLOCK) is produced when evaluation_status = EVALUATED. An incomplete check remains visibly incomplete and does not imply coverage of risks outside the encoded policy.
A testable system property

Same claim + same inputs + same rule versions = same result.

This equation is the acceptance test Atlas is built to pass. Within the encoded policy boundary, a verdict is a function of the claim, supplied evidence, and active rule versions. Evaluation records preserve the inputs and policy provenance behind each run, supporting audit review, version comparison, and reconciliation against the 835. Deterministic execution is the product foundation.

try it — the hero verdict above

↻ Re-run 1,000× → verify the stable fingerprint

The same claim, evidence, and rule versions produce the same sha256 trace. Change an input — such as accreditation evidence or the policy version — and Atlas creates a new evaluation with the change tied to the relevant rule.

// a reproducible record for internal review, payer follow-up, and audit
Designed to complement the stack

A distinct role in the claim pathway.

Healthcare platforms overlap and capabilities vary by product. This view describes their typical primary function and where Atlas is intended to sit: after claim creation and before clearinghouse submission, applying approved payer-policy logic at claim level.

LayerTypical primary functionHow Atlas complements it
EHRClinical documentation and patient recordsEvaluates available claim evidence against encoded criteria
CodingCode assignment and coding validationApplies payer-specific coverage and policy requirements
ClearinghouseClaim transport, format checks, and payer connectivityAdds a policy gate before release to the transaction layer
RCM platformBilling workflow, follow-up, and denial operationsRoutes PASS, REVIEW, and BLOCK results into existing work queues
AtlasDeterministic payer-policy executionProduces a versioned result with rule-level audit context
Who it's for

One deterministic core. Three teams with a shared record.

For the CFO

Revenue you can defend on paper

Potentially preventable denials identified before submission, with claim-level evidence that can be measured against the 835.

“Show me why” → the applicable rule and source.
For RCM leaders

Shift work toward prevention

Atlas gates the 837 before submission and routes known conflicts or missing evidence to a focused worklist with the relevant correction context.

Move appropriate work earlier in the claim cycle.
For compliance & legal

Decision context stays attached

A version-controlled, page-cited, reproducible trace for each verdict. When a payer or regulator asks how a determination was made, the answer isn't a model — it's the record.

Deterministic by design. Auditable by default.
Security & posture

Built to earn the trust the decision requires.

A bounded, deterministic decision path is easier to inspect and test. Atlas is designed to keep model inference outside claim-level execution and to minimize the data used for each evaluation.

Deterministic decision path
Claim-level execution uses approved rules without a model inference call in the verdict path.
Tenant isolation
Production deployments are designed to keep customer data and configuration separated.
Versioned history
Policy changes and evaluations retain dated, reviewable version context.
Explicit evaluation gaps
Checks Atlas could not complete are labeled rather than reported as passing.
Partners & embed

A Decision Engine you can call — not just a standalone UI.

Atlas exposes a stable input/output contract so a clearinghouse, EHR, or RCM stack can receive the claim disposition, applicable findings, evaluation gaps, suggested owner, and a reproducible trace id in one response. Existing products can keep their workflows while adding policy execution at the submission boundary.

POST /api/evaluateall findings · one call
# one call → disposition + applicable findings
{
  "claim_disposition": "BLOCK",
  "eligible_for_release": false,
  "findings": [
    {
      "rule_id": "IMG.ACCRED.MISSING",
      "evaluation_status": "EVALUATED",
      "rule_result": "BLOCK",
      "owner": "biller",
      "policy_version": "2026.1"
    }
  ],
  // honest about what it could NOT evaluate
  "not_evaluated": [
    {
      "rule_id": "MED.NEC.STEP",
      "evaluation_status": "INSUFFICIENT_DATA"
    }
  ],
  "policy_provenance": "oxford-013B.62 §4.2 p.7",
  "trace_id": "sha256:9f2c…a71b"
}

Stop denials before you make them.

See the engine render a real payer policy into a deterministic, page-cited verdict — live, on synthetic claims, with the full audit trace exposed.