How it works

From policy PDF to a decision you can replay.

LogicPearl is the healthcare decision-proof layer: it shows whether a case is ready, blocked, or missing evidence, and leaves a versioned audit trail for every result. Here is the whole pipeline, no black boxes.

The pipeline

From policy text to reviewable, source-bound criteria.

Five stages. LogicPearl compiles the policy, reads the packet, and decides. Your systems supply the case and receive the result.

input

Policy PDFs

Published payer medical policies, as issued: tables, exceptions, nested criteria and all.

LogicPearl

Compiled policy artifacts

Each policy becomes versioned, hashed, checkable criteria. Every criterion carries its verbatim source excerpt, page anchor, and quote hash.

LogicPearl

Packet evidence extraction

Built-in OCR and clinical NLP read the packet: faxes, scans, handwriting. Findings arrive with assertion states: affirmed, negated, historical, hypothetical, family. Existing extractors can feed this layer too.

LogicPearl

Deterministic evaluation

Only affirmed, present-patient evidence can satisfy a criterion. Weak or missing support routes to review, never to a silent decision.

output

Readiness result + audit packet

Ready, blocked, or missing evidence, with the blocking rule, counterfactual, next action, and a replayable decision receipt.

Anatomy of a decision

The chart is messy. The criterion is not.

Clinical language is handled explicitly before a rule ever fires. Three examples of what the evidence layer resolves:

normalized

One service, many names

"Intracept," "BVN RFA," and "basivertebral nerve ablation" all resolve to the same requested service. The criterion matches regardless of which name the chart used.

assertion-aware

Family history stays family history

A family member's arthritis is recorded as family context. It cannot satisfy a criterion about the patient. Summarizers routinely get this wrong.

negation-tracked

Absence is evidence too

"Without below-knee radiation" is a documented absence, not weak support. Negation is tracked explicitly, so exclusion criteria behave like the policy says.

Measured coverage

Coverage is counted before your first case ever runs.

The value isn't that uncertainty disappeared. It's that uncertainty is visible before it becomes a silent reviewer problem. The uncovered clauses are preserved as review work, and policies without full support stay out of auto-run.

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.

Why AI alone is not enough

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

Across 8 model groups and 24 recorded determinations on one packet and one policy: 20 approve, 4 deny. Models help extract evidence. Policy rules make the readiness decision.

An AI checklist A LogicPearl artifact
Source excerpt per criterion Often omitted or summarized Every criterion tied to source PDF text
Clause coverage accounting Usually unmeasured 99.9% measured on BCBSMA; misses preserved as review work
Nested AND/OR & exceptions Can flatten or invert Preserved in the decision logic
Table semantics Frequently flattened ~118k rows interpreted into checkable rules
Terminology provenance Implicit / hidden Explicit mapping status per criterion
Repeatable replay Can vary on the same packet Same packet and policy version → same result
Reviewer state Answer only Promotion gated, sign-off tracked
Packet-to-policy evidence link Not guaranteed Click-through packet and policy excerpts
Versioning & replay

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

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.

Integration

It returns fields, not a platform.

LogicPearl sits at the decision boundary of whatever you already run. Nothing to migrate, no UI to adopt:

Structured JSON

Status, blocker, rule ID, evidence spans, counterfactual: one call, one schema.

Browser / edge runtime

The same artifact runs in the browser or at the edge. The demos on this site run it live, with no backend.

Workqueue fields

Ready/blocked status and next action land as fields your existing workbench displays without new UI.

See the pipeline decide a real case.

The prior-auth workbench runs the whole thing, live, in your browser.

See the healthcare demoRun it on your cases