From symptoms
to a safe, signed
plan.
Cognition Boxes forces clinical AI through seven transparent, non-skippable steps — with deterministic safety checks, a mandatory clinician gate and a tamper-evident audit trail.
- No login
- No patient data stored
- Deterministic safety core
Structured history
Tap symptoms from 31 specialties, tick each red flag present or absent, and the boxes do the rest — red flags escalate the safety verdict, negatives are documented, and guideline-referenced workup lands in the plan.
Try it now →Your complete workflow
From raw intake to a signed plan and a patient leaflet — every step visible, every step sealed.
Analyze
Vitals, labs, notes, ECG and imaging text and medications are ingested. Units are converted (µmol/L ↔ mg/dL, mmol/L ↔ mg/dL), negations understood (“no fever”), implausible values flagged as input drift.
Deterministic
safety core
Doses, interactions, renal limits, allergies and cumulative QT are checked by rules you can read — never by a model guessing. Same input, same verdict, every time.
Boxes that
insert themselves
Missing creatinine? A Data Completion box appears. Notes say “afebrile” but the thermometer says 39.2 °C? Contradiction Resolution locks the verdict to HALT. Shock physiology? Enhanced Safety applies sick-day rules.
Dashed boxes are inserted at run time by the Adaptive Cognition Pathway Engine.
Zero AI
autonomy
A named clinician accepts, modifies or rejects every plan with a reason code. Only then are FHIR R4 orders produced. Overrides are tracked to learn where humans and machine disagree.
{ "resourceType": "MedicationRequest",
"status": "active",
"medication": "Doxycycline",
"dosage": "100 mg IV every 12 h",
"note": "authorised by Dr A. Clinician" }Tamper-evident
by design
Every box seals a SHA-256 hash of its output into a linked, HMAC-sealed chain. Change one character after the fact and verification fails — the gate refuses to sign an altered run.
db65f4…a25097…7f266c…a0a6a3…1ac373…0a38cf…e11e27…1c5b87…✕ Entry 5 altered after sealing — chain broken.
678 symptoms.
3,849 red flags.
Build a history in taps across 31 specialties. Every symptom carries its own red-flag checklist; present flags escalate the verdict, absent ones become pertinent negatives, and 1,533 guideline-referenced rules suggest the workup.
Take a peek inside
Three synthetic cases, three different verdicts. Open any of them in the live console.
Everything governed clinical AI needs
7 + 3 boxes
Seven fixed steps and three adaptive inserts — Data Completion, Contradiction Resolution, Enhanced Safety.
678 symptoms · 3,849 red flags
Across 31 specialties, from gastroenterology to neonatology, each symptom with its own checklist.
1,533 workup rules
Guideline-referenced investigations and management.
612 generic drugs
Formulary notes for medicines outside the interaction engine.
Validated scores
qSOFA, CURB-65, Glasgow-Blatchford, CHA₂DS₂-VA, CKD-EPI 2021, Cockcroft-Gault, KDIGO staging.
Cumulative QT
Baseline QTc plus every QT-prolonging drug, current and planned.
FHIR R4 orders
Only clinician-approved actions become MedicationRequest and ServiceRequest resources.
Model-agnostic
Any LLM or vision model can plug in and inherit the governance.
Privacy by design
Stateless server; runs stay in the clinician’s browser.
One architecture. Every stakeholder.
Hospitals
Standardised, auditable decision support with reduced medico-legal exposure.
Clinicians
Transparent reasoning you can inspect, override and sign — in minutes.
Health-AI vendors
Route your model through the boxes to inherit safety, oversight and audit.
Regulators & auditors
A complete, verifiable record of every input, rule, verdict and human decision.
“The hard problems in clinical AI aren’t the models. They’re oversight and safe deployment at the workflow level.”
Got questions?
Not yet. This is a research demonstrator of a patent-pending governance architecture. It is designed to align with the EU AI Act and to follow the FDA SaMD pathway, but it has not been assessed or cleared by any regulator and must not be used for real patient care.
Your clinical AI.
Governed. In minutes.
Open the live console →Partnerships, pilots and investment: doctoredgeinnovations@gmail.com