Three places, and none of them is a number we invented. Measurement is computed rather than performed by hand, so the repetitive arithmetic is gone from every single reading. Work is ordered by severity rather than by arrival, so the traces that need attention are the ones a reader opens first. And the state of a study — pooled, acknowledged, reviewed, signed — is visible on the study rather than in somebody’s memory, which is what removes the chasing. We publish no minutes-saved figure because this build has not been evaluated in a service, and a performance claim about a clinical system is the last place to invent anything.

The deterministic engine does the measuring: intervals, axis, amplitudes and the findings its rule set supports, computed from the stored samples under a stated algorithm version. A language model is used afterwards, and only to describe findings the engine has already established, in prose a person can read quickly. It is never asked what an ECG shows. If you see a number in a report, it came from the engine, and the interpretation records which version of it.

The interpretation is withheld. You get the explicit statement that the recording was inadequate for automated interpretation, which tells you to repeat the acquisition or read it manually. You do not get a low-confidence measurement to mentally discount — that failure mode is exactly what the quality gate exists to prevent.

Yes, and disagreements are recorded rather than overwritten. Review, override and sign-off are three separate acts with three separate permissions. An override preserves the automated finding alongside the clinician’s, with the reason and the author, and it stays in the record.

No. There is no self-service sign-up, because a system that will hold patient recordings should not hand out workspaces to whoever fills in a form. You request access, a person reviews the request, and an approval issues a single-use link that creates your organisation and lets you choose your own password. Until that link is accepted, nothing exists: no organisation, no account and no allowance.

One unit per published interpretation. A recording that fails quality validation costs nothing because nothing was interpreted, and re-analysing a study that has already been counted costs nothing. Viewing, reviewing, overriding, signing, exporting and transmitting are never counted — including after an allowance is spent. Prices are on the pricing page, read from the same configuration the checkout uses, and they do not renew automatically.

No. It is decision support for a trained clinician. This build has not been validated against a clinical dataset, it holds no regulatory clearance anywhere, and no diagnostic accuracy figure exists for it. It must not be used to diagnose a patient or to decide their treatment, and the software says so on its own screens rather than only on this page.

No, and it is built so that it cannot. The engine measures and grades; a person reviews, may override, and signs. Nothing leaves the system as a clinical document without a named professional’s signature, and the signature is bound by digest to the exact interpretation it covers.

Recordings arrive by REST or CSV upload with their acquisition parameters, and signed studies are transmitted over FHIR R4 or HL7 v2.5 (MLLP) with a delivery log for every attempt. No font, icon, stylesheet, image or analytics script is loaded from a third-party origin anywhere in the application, so a hospital install inside its own network has no outbound dependency for clinical work.

The deployment declares its AI posture — no external AI at all, minimised data, approved external AI, or an on-premise model — and the policy is enforced at the point where a payload would be built: identifiers are refused when the policy forbids them. The default is minimised data. The health probe reports which posture is actually active rather than which one was intended.

A paid period is a single purchase for a fixed term and does not renew by itself, and invoices and refunds are not generated. No receipt or reminder is ever sent, and nothing about a patient is emailed, texted or paged to anybody — no alert about a finding leaves the system by any channel. There is one exception, and it concerns equipment rather than people: where this deployment has a mail relay or an SMS gateway configured, it can send an operational notice that an ECG machine has failed, is misconfigured, or has stopped sending, because that fault produces no symptom on any worklist and someone has to be told. On a host with no transport configured, that notice goes to the deployment’s own log, and the software reports it as delivered to nobody rather than leaving you to discover it.