Fictional case, synthetic data, staff review, the insurer decides.
Coverage radar
payer × procedure · what is encoded, what is verified, what stays honestly unknown · → case clearance demo
as of · engine 0.1.0 corpus 9a85ee206a34
FIXTURE-GRADE · every value on this page is encoded from published primary sources and is pending credentialed-reviewer validation · not a coverage determination · Procedure codes appear in the full product for customers.
The variance exhibit: transesophageal echocardiography
Same procedure, verified from each payer's own published documents. Whether prior authorization is even required depends on the plan. In this August 27, 2026 snapshot, UnitedHealthcare had removed the requirement effective January 1, 2026. No scheduler can carry this in their head. That is the product.
The radar
Each cell answers two separate questions. Top chip: is prior authorization required at all, a fact that moves in both directions and is tracked with its own source and recheck date. Bottom chip: are the governing criteria encoded in the engine for that plan's own policy edition. A cell with no verified answer says UNKNOWN. It is never filled from a neighboring plan, another line of business, or a different edition of the same vendor's guidelines. Hover any cell for its source document and dates.
How this map is built
Rules decide, never a model. An evaluation runs the same encoded criteria on the same facts and produces the same answer, byte for byte, with a sha-256 receipt proving it. Every encoded cell cites its source document and the date it was read. This August 27, 2026 snapshot includes refresh-ledger rows and recheck dates, including the TAVR national coverage reconsideration recorded as open at that time.
The national layer compounds. Medicare NCDs and regulation are displayed for every plan as the cited floor for Medicare Advantage. The floor is shown and cited; it is never used to fill a commercial cell.
Unknown is an answer. Where a payer publishes nothing, delegates to a reviewer that publishes nothing, or runs a guideline edition we have not read, the cell says so. An honest blank beats a confident guess, because the guess fails exactly once: at the moment a professional checks it.