Prior auth · denials · appeals · revenue recovery

Know whether the packet proves the policy.

LogicPearl checks healthcare cases against the payer's actual policy and returns the answer your team can defend: ready, blocked, or missing evidence, with the source-linked proof and replayable receipt attached.

See the healthcare demo Run a shadow-mode pilot
Source-linked evidence Deterministic policy rules Replayable audit trail
PA-2214 Prior auth · lumbar radiofrequency ablation
Blocked
Scanning policy evidence 68%
Missing proof
MRI within 90 days Physician attestation

Most recent MRI is 141 days old. The policy allows 90. Everything else checks out: 6 of 7 criteria are documented in the packet.

Next action
Request missing evidence
records request drafted ✓
Checked against Payer medical policy §2.b · v2026.04
24

recorded determinations on one packet and one policy

20 / 4

frontier-model verdict split on identical evidence

99.9%

measured BCBSMA clause coverage, with misses visible

1 receipt

rule, source excerpt, evidence spans, and artifact version

The category

Healthcare does not have an AI answer problem. It has a proof problem.

The tools that read messy records best are the ones you can least afford to let decide. LogicPearl lets AI propose evidence while policy rules make the repeatable call.

We do not decide care. We decide whether the submitted packet satisfies the policy in front of it, and we show the exact rule, source text, evidence span, and artifact version behind the answer.

Messy intake

Faxes, scans, copied-forward chart language. Cases stall because nobody can quickly prove what is missing.

Rule-authoring bottleneck

Turning one policy PDF into configured rules takes analysts weeks. The backlog of policies never shrinks.

Clinician hours burned

Your most expensive people page through packets hunting for the one sentence that decides the case while the clock runs.

Model variance

The same packet gets different answers depending on which model you asked, or which run. Model choice is not a clinical policy.

Audit archaeology

Decisions that cannot replay turn every audit into archaeology. Reconstructing why is a project, not a lookup.

Silent policy updates

A changed policy PDF becomes an unverified rule change. Behavior shifts and no one gets an impact report.

"Same packet. Same exact policy. The models do not agree, and neither do their reasons."

We gave frontier models the exact policy text and the same patient packet. The verdicts split 20 to 4, but the sharper finding was inside the agreements: models that reached the same answer reached it for different reasons, and rerunning one model rewrote its own rationale. In healthcare, the reason is the decision. AI can summarize. It cannot be the policy of record.

The experiment · 24 determinations
GPT-5.5 ✓✓✓
Opus 4.8 ✓✓✓
GPT-4.1 ✓✓✓
GPT-4o ✓✓✓
4.1-mini ✓✓✓
4o-mini ✓✓✓
Sonnet 4.6 ✕✕✕
Haiku 4.5 ✓✓✕
20 approve 4 deny 3 runs per model group

Same vendor, opposite verdicts: Opus approved every run; Sonnet denied every run. Haiku flipped on its own third try, and even matching verdicts cited different criteria.

Model variance

AI belongs in the evidence step. Policy belongs in the decision step.

LogicPearl uses models to help read the packet, then evaluates the result against compiled payer policy rules. The readiness decision is repeatable, source-linked, and attached to a receipt.

The workbench, live

A reviewer can see the case and the policy at the same time.

The prior-auth workbench runs in your browser. Evidence boxes are drawn directly on the scanned, handwritten intake form, and every policy criterion links to the packet location that satisfies it.

The workbench highlighting a handwritten physical-therapy course on a scanned progress note, matched to the failed-conservative-care criterion with OCR evidence chips
  • Evidence spans anchored to the scanned page, down to a handwritten note that 12 physical-therapy visits plateaued
  • Every criterion shows its packet locations and the verbatim policy excerpt that requires it
  • OCR findings carry QA state, so a reviewer can agree, flag, or reject each match
  • The disposition ships with a source-linked receipt, not just an answer
Open the workbench synthetic patient · real published policies
How it works

The model can propose. The policy rules decide.

LogicPearl reads the packet with the latest OCR and document-understanding tech, then makes the deterministic readiness decision. The model proposes evidence; the policy rules decide. Your workbench receives the result.

your systems

Case data

Faxes, scans, chart notes, claim lines: the packet as it actually arrives.

LogicPearl

Evidence extraction

State-of-the-art OCR and clinical NLP read the packet: faxes, scans, even handwriting. Every finding gets an evidence state.

LogicPearl

Policy artifact

The source policy, compiled into versioned, checkable criteria.

LogicPearl

Readiness result

Ready, blocked, or missing evidence, decided by policy rules, not model mood.

your workbench

Next action

Status, blocker, rule, and evidence land as fields your workqueue already understands.

compliance

Audit packet

Every decision replayable later, against the same artifact version.

The readiness result

Every case answers the same five questions.

Status, satisfied and missing requirements, the blocking rule, what would make it ready, and the next action, each tied to the source policy text. Misses and open items become reviewer work queues, never silent omissions.

PA-2214 Readiness result
policy_artifact_v2026.04 Blocked
Requirements
✓ Chronic low back pain ≥ 6 months satisfied: packet p.8 ¶2 · policy §1.a
✓ Conservative therapy trialed ≥ 6 weeks satisfied: packet p.11 ¶4 · policy §1.b
✓ No exclusion criteria matched satisfied: prior fusion at level documented absent
✕ Advanced imaging within 90 days of request missing: most recent MRI dated 141 days before request · policy §2.b
? Symptom laterality consistent across chart needs review: p.9 notes left-sided pain, p.14 notes right-sided
Counterfactual

What would make this ready: an MRI dated within 90 days of the request, and reviewer sign-off on the laterality conflict.

Next action

Request missing evidence

Records request drafted for imaging; laterality conflict routed to the nurse-review queue.

Source policy

§2.b · imaging recency · p.4 of source PDF
quote hash sha256:9f3c...a41e

Audit & versioning

Every decision can be replayed later against the same artifact version.

A decision receipt binds the packet, the rule, the evidence spans, and the exact policy version into one replayable record.

Decision receipt
Replayable
Artifact hashsha256:c81d...f209
Policy sourcePublished payer medical policy · §2.b, p.4 · verbatim quote hash bound
Ruleimaging_recency.v3
Evidence14 packet spans, each anchored to page and paragraph
Reviewer historyNo overrides · nurse-review sign-off 2026-04-18
Semantic diffv2026.03 → v2026.04: imaging window tightened, 180 → 90 days
Replay same packet + same artifact version → same answer

Why an answer changed is a question you can answer.

Run the same case against the April source set and it routes to review. No covering policy existed yet. Run it against June, where a new policy took effect on the first, and it decides. Same case facts; the source set changed. That is an explanation an auditor accepts, and a model can't give.

When a payer updates a policy PDF, the change lands as a semantic diff between artifact versions, not a silent behavior shift. You see which open cases just changed, before they surprise you.

"You can't defend a denial with a reason you never gave."

Move the goalposts mid-appeal and the process starts over: new disclosure, new clock, new exposure. And the stakes compound: industry-wide, most appealed denials are overturned, and overturn rates feed CMS Star Ratings, which drive bonus payments and enrollment. LogicPearl keeps the original rule, evidence, and source bound to the decision, so the reason you gave is the reason you defend.

Measured, not promised

The engine arrives with its coverage already counted.

The numbers below are for Blue Cross Blue Shield of Massachusetts policies. We can run the same coverage accounting for any payer policy set before a pilot starts.

468
Blue Cross Blue Shield of Massachusetts policies compiled from 528 published policy PDFs
99.9%
measured clause coverage. The misses stay visible, never silent
~2,900
reviewable criteria, each tied to verbatim source policy text
~118k
policy table rows interpreted into checkable rules

Measured on the Blue Cross Blue Shield of Massachusetts policy corpus. The same coverage accounting can run against any payer policy set before a pilot starts.

Where it pays for itself

One decision layer. Four places it goes to work.

Each returns the same thing: ready, blocked, or missing evidence, with the proof attached.

Prior authorization

Evidence readiness before the clock starts

Know which requests are ready to decide, which are blocked, and exactly what proof is missing, with the records request already drafted.

See the use case →
Denials & appeals

A denial you can defend or dissect

Issuing denials? One source-bound rationale that survives the appeal. Fighting one? See exactly when it was decided, under which policy version, why, and what evidence would flip it.

See the use case →
Revenue recovery

Know which claims are worth working

Ready, blocked, and do-not-work signals on every account before an analyst opens it, with dollars ranked by what can actually be recovered.

See the use case →
Policy change impact

A policy updates. Which open cases just changed?

Semantic diffs between policy versions, replayed against your open inventory. A changed PDF stops being a silent rule change.

See the use case →
The shadow-mode pilot

This is a readout, not a rollout.

Evaluate LogicPearl beside your current process before changing production.

A shadow-mode pilot runs on a bounded set of historical cases. Production stays unchanged. The output is a case-level readout your clinical, policy, and operations teams can review together.

How the pilot works Talk about a pilot
What you get back
  • Historical case comparison
  • Agreement / disagreement report
  • Missing-evidence patterns
  • Compiled policy artifact
  • Audit packet samples
  • Production integration plan
Fits your systems

Your workbench stays in place.

Keep your current platform. If you already run extraction or AI tooling, its output plugs straight in. LogicPearl sits at the decision boundary and returns structured JSON, browser-runtime output, or workqueue fields your systems already understand.

UM platforms Claims workbenches Care management CRM & case systems EHR-adjacent queues

Built to sit beside Epic, GuidingCare, Waystar, Salesforce, or the claims workbench you already run, as the governed decision layer underneath, not a replacement for any of them. It can run in your environment with no hosted SaaS dependency and no phoning home. You own the artifact, trace, and readout.

Start with one workflow. Prove it in shadow mode.

A bounded case batch, a case-by-case readout, and a decision you can defend either way.

See the healthcare demoTalk about a pilot