Platform
Four layers on every screen
The hardest problem in clinical AI is not accuracy. It is telling a clinician, at a glance, which of the things in front of them is a fact, which is a machine's suggestion, which is an independent check, and who decided what. VerityRT encodes that distinction in the interface itself.
Confirmed record
Signed diagnosis, prescription, approved anatomy and approved plan.
A solid rule in the darkest ink. These are the things a person put their name to.
AI work
Drafts, suggestions, uncertainty, and the version that produced them.
A dashed teal rule. Never mistakable for the record, however confident it is.
Independent checks
Deterministic rules and the safety agent's findings.
A rule in the severity colour, always paired with an icon and a word — never colour alone.
Decision history
Who accepted, changed, rejected or escalated each item, and why.
A hairline. Quiet, complete, and impossible to edit after the fact.
Workspaces
Built around the clinical decision sequence rather than a menu structure.
- Command
- The department view. The headline number is cases requiring attention, not total case count.
- Case Board
- Referral-to-treatment timeline with the four layers distinguished, missing inputs named, and what ran without anyone asking.
- Vision Room
- Axial, sagittal and coronal panels, a volume view, structure visibility, draft-versus-approved comparison and a spatial uncertainty heat map.
- Prescription Console
- The directive, the protocol and the record reconciled field by field, with a visible physician gate.
- Plan Lab
- Beam and arc geometry, an objective table where every AI proposal states its reason, and plan variants that name their own cost.
- Dose Room
- DVH, isodose, reported metrics with provenance and uncertainty, and the result of the independent recomputation.
- Guard Review
- Pass, warning and block findings, with cross-agent disagreements shown prominently and never bundled away.
- Approval Room
- The five gates, the roles each requires, the blockers that refuse a signature, and the immutable history.
- Submissions
- Clinical review and prior-authorisation packets, assembled automatically, with missing items named and human attestation required.
- Treat monitor
- Approved plan identity and hash, fraction history, deviation monitoring and a simulated safe hold. No beam control.
- Teaching Center
- Expert corrections captured with reason, materiality and downstream impact, feeding a candidate training set — never the live model.
Safety zones
Four separated environments. An AI service never holds credentials for the delivery interface.
- Research
- De-identified data, experiments and candidate models. Cannot connect to clinical export or delivery.
- Clinical draft
- Patient data and unapproved versions. May create drafts. Cannot release anything.
- Clinical approved
- Signed contours, prescription, plan and QA. Immutable except through a new controlled version.
- Delivery interface
- The minimum approved package handed onward. Strict allowlist, no experimental agent access, acknowledgements required.
A deterministic release service, not an agent