The sandbox
A scoped credential against the live exchange. Safety is a property of the credential itself, not a special demo build — which is why we can hand it to a stranger.
Most vendors answer an integration question with a slide. We issue you a scoped credential and a base URL, and you query the running exchange from your own terminal — in about sixty seconds, without talking to anyone.
You can inspect how the exchange works in as much detail as you like. What stays closed is the clinical content inside it.
A scoped credential against the live exchange. Safety is a property of the credential itself, not a special demo build — which is why we can hand it to a stranger.
Rendered from the same catalog the automated grader enforces. It is structurally incapable of describing a rule we do not check.
Hand a privacy officer the audit trail without handing them the console. There is no write path in the page — not a hidden one, none at all.
The exchange runs on open standards and open-source software — HL7 v2, FHIR R4, LOINC, HAPI. We are not pretending the plumbing is secret, and showing you exactly how it is assembled costs us nothing.
What is ours is the clinical vocabulary inside it and the record of every decision that built it. Those are not on this site anywhere, and they will not be.
The credential is scoped at issue. That is what makes handing one to a stranger a non-event.
Work email and organization. No phone number, no calendar link, no qualification form.
Issued on the page. Valid 30 days, rate-limited, scoped to a synthetic participant.
The same endpoints a production participant uses. Not a mock, not a recording.
A sandbox proves the interface works, not that we have run at your volume. We have never carried production clinical traffic — every message through this exchange to date is synthetic. If you need a reference site under load, we do not have one yet, and we would rather you learn that here than three meetings in.
Not because we are careful. Because of where the text comes from.
A hand-written specification drifts from the software the moment either changes, and you have no way to detect it from outside. This one renders from the same catalog the automated grader evaluates against, so a rule we do not enforce cannot appear here, and a rule we enforce cannot be missing.
You are not trusting our documentation discipline. You are reading the enforcement itself.
An empty review queue does not mean a finished system. It means an unsupervised one.
The remaining 3,853 are quarantined and visible in the queue below — not lost, not silently guessed at. Figures read live from the platform, as of 7 Sep 2026 05:47 UTC.
Machine-generated proposals a steward has not yet ruled on. This number going up is the system working, not falling behind.
The tiers, which tier may commit without a person, what happens to everything else, and how much is waiting right now. The whole mechanism.
The terms themselves and what they were bound to. That vocabulary is the part that took a year to build and the part a competitor would actually want. It is not on this site anywhere.
A binding may be written automatically only when the code exists and is ACTIVE in the loaded release and matches exactly. Everything else becomes a proposal that waits for a named person on a dated action, and a result whose code we cannot resolve is quarantined rather than guessed at — then resolved retroactively when a steward finally decides.
What a privacy officer asks for is rarely access. It is evidence.
READ-ONLY BY CONSTRUCTION — this view has no form, no action parameter and no write handler. It is not a permission that could be misconfigured; there is no code path here that changes anything.
The fourth row is a request that was denied. A log that only contains successes tells a reviewer nothing about whether the control works.
The same record exports for a single individual, which is the form 45 CFR 164.528 actually requires — built as a feature, not assembled by hand on request.
// Synthetic deployment. Participants are numbered, not named, on the public view.
Every vendor has one. Most make you find it out in month three.
We are not certified under SOC 2, HITRUST or FedRAMP, and there is no such thing as being “HIPAA certified.” We have not run US Core conformance testing against Inferno. We do not have a production cloud deployment. We publish no service-level commitments, because we have never measured against one. And we have never carried real patient data.
If any of those are prerequisites for you, we are not the right fit today. We would rather you close the tab now than discover it during procurement.
A credential takes a minute and commits you to nothing. If you would rather talk first, that works too — but the credential will tell you more.
Clinical Decision Support Notice — CareCompile is a non-diagnostic clinical decision support tool intended to augment, not replace, physician judgment. All AI-generated analyses are advisory only; clinical decisions remain the sole responsibility of the licensed treating clinician. CareCompile is not FDA-cleared or FDA-approved as a medical device. Every figure on this page is measured on synthetic data in a development environment.