Payer prior authorization / public-policy aware

The record must stand when the question is asked.

The argument

Payers do not need another black box that promises speed. They need decisions whose criteria, evidence, and handoffs can be inspected after the fact.

States are setting expectations for deterministic criteria, human review, and appeal trails in AI-assisted prior authorization. The audit is not an edge case. It is part of the product.

Vistra is a career artifact built from the payer's chair: a platform concept that treats the evidence trail as the work.

The demo

Three-minute walkthrough

Decision logic, not theater.

A concise walkthrough of evidence extraction, rules evaluation, and the human pend path is in production.

DEMO STATUS / COMING THIS WEEK

Video and interactive demonstration are not published yet. Demo coming soon

How it works

Separate the reading from the ruling.

The language model is constrained to intake and extraction with citations. A deterministic rules engine applies the real public CMS LCD criteria. When the record is incomplete or uncertain, the platform does not guess.

01 / INTAKE

Extract with citations

Clinical documents become traceable facts. The source remains attached to every assertion.

02 / DECIDE

Apply published criteria

Deterministic rules encode relevant public CMS LCD policy criteria and produce the decision state.

03 / ESCALATE

UNKNOWN means pend

Ambiguity routes to human review with the evidence trail intact. It never becomes an invented certainty.

Detailed architecture diagram: coming soon with the public payer-os repository.

The proof

Evidence before a score.

An evaluation against a labeled golden set will be published after the evaluator and its failure taxonomy are ready for scrutiny. Vistra will not present illustrative performance as a real result.

Evaluation status / Coming soonMetrics, sample definition, and a one-line failure taxonomy will appear here once reproducible results exist.

The build log

Progress should leave a record too.

Implementation notes and public reflections will be collected here as they are published. No retrospective certainty; just the decisions, limits, and revisions that shaped the work.

POST 01 LinkedIn build log coming soon