Evals for Alaffia Health

Scanned and mixed-quality document handling

Extraction from large PDFs and scanned images, including poor scans, rotated pages, and handwriting-heavy records.

Normalization into clinical facts

Mapping heterogeneous record formats into the pre-built clinical fact structures used for decisioning.

Provenance to source page

Every extracted fact retains a traceable pointer back to the originating document location.

Incomplete or missing records

Behavior when the submitted record set lacks documents needed to support a determination.

Criteria application to the record

Matching documented clinical findings to the specific criteria elements required for a determination.

Plan policy precedence

Applying the plan's own policy language where it governs over general clinical criteria.

Review-type context

Prospective, concurrent, and retrospective review framings for the same underlying record.

Conflicting or ambiguous evidence

Handling records where documentation partially supports and partially contradicts the billed service.

Citation grounding

Each asserted clinical fact in the rationale links to a real location in the submitted record or policy.

No unsupported assertions

Rationale does not introduce clinical findings absent from the cited sources.

Rationale legibility for appeals

Explanations stated in terms a reviewing clinician or appeals body can follow and contest.

Uncertainty disclosure

Explicit signaling when evidence is thin rather than presenting a confident conclusion.

Route-to-human thresholds

Escalating claims where automated handling is not appropriate instead of auto-completing them.

Prioritization of review queues

Ordering claims so reviewer attention lands on the most consequential cases.

Reviewer follow-up questions

Responding to a clinician's clarifying questions against the same evidence base.

Sign-off as the completion gate

Claims reach completion only through licensed clinician validation, not agent output alone.

Plan-specific policy ingestion

Honoring the configured policy set rather than defaulting to generic industry rules.

Workflow and output shaping

Producing outputs in the structures the plan's claims process expects.

Configuration change adherence

Behavior updates consistently when a plan revises its criteria or thresholds.

Org-type variation

Differences across health plan, TPA, reinsurer, and government agency configurations.

Never used to deny care

Agent output stays a recommendation and does not present itself as a care denial.

Scope of clinical judgment

Declining to substitute for the licensed clinician's determination.

PHI handling discipline

Member and payer data stays within the configured environment and is not echoed beyond the workflow.

Learning boundary

Insights derive from the plan's internal workflows rather than leaking across tenants or to third-party models.