The engine is the company. The device is the first thing it proved.

Building that device, we built the thing behind it: an independent adversarial-validation engine whose job is to find out whether a clinical decision is wrong. To trust our own methods, we turned the engine on ourselves first. We pre-registered a hypothesis across 286,510 ICU patients before touching the data, and when the primary result came back against us, we published the null. The device is the first thing the engine proved. Everything below is anchored to an artifact you can check.

1. The wedge: ICH Triage.

ICH Triage detects intracranial hemorrhage on head CT. It runs offline, on a single consumer-grade GPU, inside the institution. No patient data is transmitted off-site; inference happens on-premises and returns in under 200 milliseconds. It received an FDA Pre-Submission acceptance, control number Q253397, in January 2026, on the Class II Software-as-a-Medical-Device pathway.

The differentiator is deployment economics, not a benchmark. Most clinical AI assumes a cloud round-trip, a data-sharing agreement, and recurring per-study cost. ICH Triage assumes none of that. A hospital runs it on hardware it already understands, behind its own firewall, with nothing leaving the building. That removes the two things that usually stall a clinical-AI purchase: the security review over PHI egress, and the per-inference cloud bill.

Preliminary performance, reported on the product page: AUC-ROC 0.9592, sensitivity 92.5%, specificity 85.0%, NPV 98.5%. These are labeled preliminary on the site and will move only with the evidence behind them.

2. Real demand, before any sales effort.

We have not run a sales motion. The signal we have is pull, not pipeline: working ICU physicians have opened the demo and exhausted their trial allocations in single sessions. That is the behavior of people who have the problem the device addresses, using it until they hit the wall, the first time they see it. We are reporting the demand we can stand behind, and nothing we cannot.

3. The same engine generalizes.

ICH Triage is one output of the validation engine, not the whole of it. The engine compares a clinical decision against an established, peer-reviewed deterministic instrument and asks whether the decision holds, whether that decision came from a model or from a person. The deterministic instrument is the locked reference standard; the verdict is measured against that standard, not against the output being checked. Every evaluation is cryptographically signed with Ed25519 and carries a validation ID: input hash, scoring-tool version, output, divergence, timestamp. A reviewer who had no part in building the model can audit the receipt end-to-end.

What shows the engine is general is the range of instruments already wired into one comparison layer: 19 deterministic clinical scoring specialists, spanning renal function (KDIGO, Cockcroft-Gault), deterioration and ICU mortality (NEWS2, APACHE II), readmission and discharge (LACE, BRASS), comorbidity (Charlson), emergency triage (ESI), and psychiatric and population-health screening (C-SSRS, PHQ-9, USPSTF). Two imaging pipelines (chest X-ray, head CT) and an LLM-reasoned sepsis pathway are integrated and in pre-demo hardening. Adversarial pharmacovigilance, FAERS disproportionality, operated by RocSite Discovery as Caliper ↗, is live in production. These are not separate product lines asking to be funded; they are the same engine reaching new kinds of evidence.

4. The proof.

This is the part that does not depend on trusting us. Each claim below points to something you can open, hash, or look up.

The methodological foundation is a pre-registered falsification study of sepsis prediction models across 286,510 ICU patients (MIMIC-IV and eICU-CRD). The hypothesis was locked on OSF before any data access; the primary cohort returned a null against our own hypothesis; we published it anyway. The methodology was reviewed by an established sepsis informatics researcher in connection with the credentialed dataset access.

Running the same engine against the published literature surfaced a second, separate finding: in a pre-registered study of 201,905 ICU encounters, the mortality risk that standard tools assign to older adults diverged from the published figures by +66.3% for atrial fibrillation, +131.7% for diabetes, and +168.4% for acute myocardial infarction. That study is pre-registered with a locked OSF timestamp (2026-03-10) and is submission-pending, not yet published; we are listing it here at exactly that status.

5. Founder, and what comes after one founder.

The company is operated by Adam Dickens: 21 years at IBM, an IBM Certified Architect, with regulated-industry experience including Johnson & Johnson. That is the execution discipline behind a pre-registered study, an FDA pre-submission, and a signed-receipt validation engine built by a two-person company. More on the founder.

The honest risk in a two-person company is key-person dependency, and we are naming it rather than papering over it. The first use of funds addresses it directly: a clinical co-founder and a small clinical advisory board, so the medical judgment in the loop does not rest on a single person and the company can stand without any one individual.

6. The ask.

One milestone, with a finish line, not a stack of addressable markets. We are raising capital to take ICH Triage from accepted pre-submission to FDA clearance, land the first hospital design partners over roughly 18 months, and add a clinical co-founder. The money has one job. If ICH Triage clears and runs in its first partner hospitals, the engine underneath it has earned the right to its next problem on its own merits.