Did the AI lose the truth on the way to you?
A fluent summary can drop the one line that matters: the allergy, the dose ceiling, the two-person transfer. Paste the original and the summary, and the gate shows what was kept, weakened or left out. Both texts stay in your browser; nothing is sent or stored.
What do you have?
- ◆Penicillin allergy — prior anaphylaxis. Do not give amoxicillin or any penicillin-class antibiotic.
- ◆Warfarin 5 mg daily. Do not exceed 5 mg without an INR recheck.
- ◆Two-person assist for all transfers. Fall risk, prior hip fracture.
fidelity_contract:
protected:
- clause: penicillin_allergy rule: preserve_verbatim
- clause: warfarin_ceiling rule: never_soften · keep "5 mg" + "INR"
- clause: two_person_transfer rule: never_drop
on_violation: block_and_flagThe check is deterministic — edit the summary above and the gate re-scores live. Illustrative care plan; the mechanism is the point.
Check your own AI summary
Paste the original (a care plan, discharge note, medication list, a label) and the summary an AI wrote from it. This pulls the warnings and doses out of the original and shows which ones the summary kept, weakened, or left out. It runs in your browser. Nothing you paste is sent anywhere or saved. Do not treat a clean result as proof: it finds what is missing, it cannot tell you the summary is right. Read both.
Did a physician sign off before the AI acted? The hard intercept is designed to stop autonomous action without a human in the loop.
Did the AI keep the truth after it acted? The check that stops a summary from quietly losing the clause that keeps a patient safe.
Capability is racing. Governance is the product.
The frontier labs are optimizing raw capability. In medicine, the system that ships is the governed one — the model that earns a human sign-off and never loses the truth on the way back. That layer is what HarnessHealth builds.
Keep your family's papers together in the ComfortCard family file, on your device. In a crisis, call or text 988; in immediate danger, call 911.
HarnessHealth is the AI governance infrastructure layer of the SolvingHealth stack — physician-supervised AI deployment, draft-until-attested gating, audit-ready receipts.