A clinical attestation layer for the non-extractive health ecosystem

Every claim has an owner.
A licensed physician signs last.

A letter is not one decision. It is a list of claims, and each one belongs to exactly one party. The family attests to the facts they know. The checks run against the written rule and are recorded, with the version that ran. A licensed, specialty-matched physician signs, asks or declines each medical judgment, and names the facts it relied on. Whoever owns a claim carries its risk, and the receipt shows who that is.

See who owns which claim on an invented case, in your browser, nothing sent.

Physicians: would you sign it? Try the five-case review drill, about 5 minutes, nothing sent.

TodayA pre-go-live pilot on synthetic cases. The review loop, the specialty gate and the receipts run.

NextThe first live path: a letter of medical necessity for one partner’s members, once a BAA is signed and reviewer coverage is bound.

SAMPLE CASE · SYNTHETICCare assessment · Internal Medicine
AI output

Patient meets criteria for skilled home nursing 3×/week. Fall risk 4/5. Recommends PT eval and medication management.

Flag: the documented needs (supervision, meal prep, reminders) read as custodial / companion care, a different benefit category.
The physician’s decision · tap one
Decision receiptSample · not anchored
DecisionREVISE
ReviewerMD · independent · specialty-matched
SHA-256 · made in your browsercomputed when a decision is made
Madeon your device
Verify a real demo anchor on the public ledger →

Same flat fee whichever you tap. A live decision is anchored to hashcare.com and countersigned into Bitcoin daily; this sample is not.

VALIDATIONNPI-verified reviewers NEUTRALITYFlat fee — same for yes and no PRIVACYHash-only receipts · no PHI FINALITYCountersigned into Bitcoin daily
The whole thing, animated

ClinicalSwipe in 15 seconds

Text version:
  1. The AI called it skilled nursing. The notes say meals, reminders and supervision: companion care.
  2. A review is one card. Swipe it. Right to approve, left to reject, up to revise. Only in your specialty.
  3. Yes or no, it becomes a receipt. A fingerprint, no patient data, on a ledger anyone can check.
  4. Try three sample cards. Then apply. Synthetic cases today. Your NPI is checked against NPPES.
  5. ClinicalSwipe. The accountability layer for healthcare AI. clinicalswipe.com
Honest state
Where this actually is, today

ClinicalSwipe is a working system in a pre-go-live pilot, and we would rather you read that here than discover it later. The review loop, the specialty gate, the receipts and the public ledger anchoring all run today. What is not yet in force: no BAA is signed, so every case runs on synthetic or de-identified data; and reviewer malpractice, AI-liability and cyber coverage is not yet bound, and it must be bound before the first real case. The Medical Director signed the qualifying-condition criteria on 11 September 2026 (the protocol). The first live path is a letter of medical necessity for one partner’s members — the smallest piece of what a signature is for, not the business.

“Specialty-matched” is enforced in code and it fails closed: a reviewer is shown only cases inside their own specialty, and an output with no matching reviewer is held, never quietly signed by whoever is available. The honest consequence is that our live coverage is narrow — family medicine and primary care today. If your output sits outside that, the gate will hold it rather than pretend.

We publish this because the product is accountability. A service that overstates its own readiness has already failed the thing it sells. What a company gets today →

Where the risk lives

Three owners, in this order. Nobody signs over a question.

01

The family attests to the facts

Who has authority to speak for the person, which daily activities need hands-on help, for how long, how many hours, and who is paid. They answer in their own words and attest that it is true as far as they know. They own those facts, and nothing about medicine. Change an answer and they attest again.

02

The checks run and are recorded

Each check reads a written rule, such as the chronic-illness definition in 26 U.S.C. §7702B(c) or the payee rule in §213(d)(11), and records pass, question or block with the version that ran. A wrong check is the platform’s error. ClinicalSwipe never attests to anything on its own authority.

03

The physician signs last, claim by claim

An NPI-verified physician in the right specialty signs, asks or declines each judgment and names the facts it rests on. Sign stays off while any fact it relies on is blocked or in question. A decline or a question needs a reason, and the family is told the part, the reason, the rule and what they can do next.

Run the ownership rules on a clean case and a troubled one →

This is the whole job

A review is one card. Swipe it.

Read the AI's draft, catch what's wrong or missing, and decide. Drag the card — right to approve, left to reject, up to revise — or tap. Three sample cases, synthetic.

ClinicalSwipe
3 in queue
Every decision is signed under your NPI and audit-trailed. You're attesting documents, not treating patients.

These cards show the flag. The review drill hides it: five cases, your call, then the reference answer and its source.
The proof layer

Every decision ends in a receipt.

Signed is good; provable is better. Each determination — yes or no — is fingerprinted and anchored to a public ledger, countersigned into Bitcoin daily. An auditor verifies it in one click, without trusting ClinicalSwipe. Read the whole ledger →

Attested decision receipt

Demo specimen
DeterminationQUALIFY
ReviewerMD · independent · specialty-matched
SHA-2563978a3160bbe1d04f55b47893fdec48bb03977159573b208633da4cced99f78f
Anchored2026-08-20 20:17:36 UTC

A real ledger anchor from the 2026-08-20 demo — synthetic case, real anchor; the fingerprint contains no name, no diagnosis, no PHI.

What a signature can prove

A signature proves who signed. It does not prove anyone looked.

A draft signed without a single change is not a feature: on its own, it is no evidence anyone looked. The evidence is the difference between what the AI drafted and what the physician signed. Drafting is allowed. Judgment needs a name.

Today: every decision carries its reason

Sign, revise or decline, the physician writes why, and the decision is recorded under their NPI. An approval needs a source document in the case packet. A revise is a decision in its own right, fingerprinted and anchored like the others. Check a decision yourself →

Next: the delta, line by line

ClinicalSwipe will keep what the physician kept, struck and added against the AI draft, as labeled lines on the record. That difference is the evidence a person was in the loop, and the dataset this field does not have yet. Try keep-or-strike on one draft →

The open test

orthoharness is an open-source (Apache-2.0) benchmark harness for orthopaedic AI. It turns a draft and the signed note into labeled kept, struck and added lines, and records who signed: an answer with no named signer is reported as a draft. Any vendor can run it. github.com/blainomd/orthoharness →

California drew the same line

AB 1979, signed 30 September 2026 (Chapter 854, Statutes of 2026), bars a clinic or physician’s office from using AI to independently perform a clinical function the law requires a licensed person to perform. It carves out documentation and communication that involve no professional judgment. Read the chaptered bill →

One question to put to any AI vendor: what share of your drafts does a clinician change before signing? A promised outcome is marketing; a measured error is evidence. Six questions for any AI vendor →

Why it holds up

The signature isn't the point. The accountability is.

A named, licensed human

Every determination is owned by a real, specialty-matched physician — not an algorithm, and not a stranger to the field.

The honest "no" is real

A physician who can decline is what makes the "yes" mean something. The decline is a feature, not a failure.

No stake in the answer

The review fee belongs to the physician who decided — flat, the same whether they sign or decline — and we take no percentage of any claim, sale, or savings. Economics that keep the no honest. Today: the physician fee is set in writing before any live case. Next: the software is paid on a benefit that can be measured, and not before.

Audit-trailed from step one

Signed, timestamped, anchored, and retained — a record built to survive review, not just to look official.

Who the signature serves

Most of healthcare is an exchange of information. The signature is the part that needs a person.

The patient

The best available choice of care, decided quickly, because the reading was done before the card arrived — not a seven-minute visit and a surgery date they don’t understand.

The physician

Reimbursed for the decision, not for the visit it used to take to justify one. Read the draft, weigh the risk of a yes, decide. Trust is earned one case at a time; that is what the drill is for.

The family caregiver

The person giving care at home is part of the record. Where a program requires a clinician’s signature, it goes to one; everything else stays with the family. The family caregiver program →

The AI company

A named, licensed human behind the output, with a receipt. A condition of entry now, not a nice-to-have. What a company gets today →

Why the demand is becoming structural

A named accountable physician is turning into a condition of entry.

Two things happened this year that turned that from a design opinion into a written requirement.

CMS made it an admission ticket

CMS’s ACCESS Model (live since 5 July 2026, a ten-year run) requires participating organizations to designate a physician Clinical Director for clinical oversight and compliance — a condition of participating, not a nice-to-have.

The most serious agentic program bought a checker, not more reading

ARPA-H’s four-year ADVOCATE program separates the acting agent from a second, independent supervisory agent, with an outside laboratory evaluating both. Making clinical AI trustworthy at national scale did not call for a bigger review committee.

So the human moved, and it matters where

In that architecture the named human owns escalation and accountability, not accuracy — the same split this product is built on: cheap deterministic checks catch what’s checkable; a licensed human owns judgement calls.

And the limit we will not sell past

A human in the loop buys accountability, not accuracy. The evidence doesn’t support “reviewed, therefore more accurate,” and we won’t claim it.

Sources: CMS and ARPA-H program pages, read 10 September 2026. Nothing in ADVOCATE is FDA-authorized — authorization is that program’s stated goal. Full derivation: solvinghealth.com/model.

Where a determination is worth money

If a clinician is treating you for something, the care you already pay for may be a medical expense.

Exercise therapy, nutrition, sleep, recovery, aging at home — and the care a family gives a parent. One independent physician's determination, written honestly, decides whether it counts. An honest no costs the same flat review fee as a yes.

First partner Kinesis Integrated→ Members: see how the review works for your pathway.
Three ways in

Physicians, companies, partners.

For physicians

Put your license to work.

Review AI-generated clinical outputs in your own specialty, on your own time. NPI-verified, clear scope, no shifts and no minimums.

Apply with your NPI →
For companies

Your agents act. Ours attest.

Give your AI outputs a licensed, accountable physician signature and an audit-ready, ledger-anchored record — without building a clinician network yourself. Or try the sandbox first.

Start a conversation →
For partners & brands

HSA/FSA-eligible — for customers who qualify.

Coaches, gyms, wellness brands: an independent physician reviews each customer's case and signs a Letter of Medical Necessity — or declines honestly. The fee is flat per determination, paid by the partner and never the customer, and set in writing before any live case; receipts for everyone.

See the partner motion →