Prove your humans are real.

The standard

What counts as real human review

CMS's WISeR model requires the review of a human clinician on every non-affirmation — it cannot be performed solely by technology, and inaccurate determinations are audited, quality-scored, and terminable. The AMA's 2026 policy requires meaningful human oversight: physician-led, transparent, auditable. Both name the requirement. Neither publishes the test for whether a review actually happened. Here it is — five things, all provable. Measure any vendor against them, including us.

1. Licensed and matched. A clinician with relevant expertise — matched to the case's state and specialty — makes the call on every adverse determination, and the license is verified against CMS NPPES at the moment of signing. Not a model, and not a model with a signature bolted on afterward: "a physician reviewed this" is a fact you can prove, not a claim you make.
2. Independent. The reviewer is organizationally independent of the developer of the AI under review, and compensation never depends on the direction of the verdict. A vendor grading its own homework is governance, not review.
3. Engaged — measured, not assumed. Double-review calibration, a published rubric, disclosed inter-reviewer agreement — because the failure mode of "human review" is the rubber stamp, and the only defense is a number an auditor can inspect.
4. Receipted. Every determination leaves a tamper-evident record — what was reviewed, the verdict, the reviewer's credential class, the timestamp — generated by an API call the decision can't skip. The artifact you hand an auditor, a payer, or a court, and it holds without trusting you or us.
5. Durable. Receipts are anchored into a public registry countersigned into Bitcoin daily and retained for the full audit period — the proof survives any dispute, any vendor, any of us.

The test that settles it: if the human can be removed "whenever a health system chooses," the human was a cost, not a control — and a removable human is not real human review. The Review Rail is built so the human can't be quietly subtracted. The receipt is the product. These five are v1.0 of a published, hash-anchored standard — The Five Tests of a Real Review — free for anyone to adopt, meet, or exceed, including our competitors.

The two-week pilot

1. Week 0: pick 50 real determinations (de-identified as your counsel requires). We staff state-licensed reviewers and calibrate them on a shared rubric.
2. Weeks 1–2: every artifact double-reviewed independently; disagreements adjudicated by a senior physician; every verdict receipted at signing.
3. Delivery: verdicts, inter-reviewer agreement statistics, a measured detection rate on seeded known-bad cases (the number a rubber-stamp rail can't produce), and a receipt per artifact — the exact packet a procurement questionnaire, a regulator, or a reporter asks for. Fixed price, no commitment beyond it.

The pilot is scored against a published, hash-anchored standard: The Five Tests of a Real Review — licensed and matched, independent, engaged (measured), receipted, durable. Free for anyone to adopt, including your other vendors; we meet it test by test.