Configure laboratory rules, policy and AI

Available to

Laboratory directors.

Everything that decides what the layer does is controlled configuration. It moves from Draft to Submitted, then Approved by the laboratory director, then Active, and a different person must approve what someone else submitted. Approved content cannot be edited: make a new version instead.

Open the + menu in the Laboratory app and choose Rules and AI configuration.

Rules and AI configuration

Screen	What it controls
Policies	One policy for each facility, and a default for the rest: AI mode, critical alerts on arrival, second-person verification, quality-control gating, strict reference intervals, time windows, allowed data regions, retention and the daily cost cap.
Rule sets	The rules that run on every result: plausibility limits, change-since-last-time checks and specimen age.
Reflex protocols	Follow-up tests suggested from a finding. A suggestion stays a recommendation until the clinician authorises it.
Unit conversions	Approved conversions between units. Without one, a result in another unit is held back.
AI models, Prompts, References, Routing	The models that may be used, the prompt templates, the approved reference notes and which model handles which use case.

AI mode

AI mode	What happens
Off: rules only	The default. Only the rules run.
Shadow: model runs, nothing is shown to reviewers	The model runs and is measured, but nobody who treats patients sees it.
Supervised: suggestions go to named reviewers	Suggestions appear on the review worklist for named reviewers to accept or reject.

Maintain a rule set

Open Rule sets and select the active set, or New for a first one.
Add a line for each rule: its Kind (plausibility, delta check or specimen age), the Test, the Severity and the limits.
Select Submit. A different person with the director role selects Approve, then Activate.
To change an active set, select New version. The earlier version stays on every case it produced. Retire ends a set.

A rule set and its lines. The status bar shows its approval stage.

A hosted model needs every approval on the form before it can be activated.

Good to know

An external model also needs a signed data-processing agreement, a no-training contract term, a security assessment, a transfer-impact review, validation evidence, a data region inside the policy's allowed regions and an HTTPS address. A model on the laboratory's own network is the only exception for a private address.
The built-in explainer is deterministic and runs on site: nothing leaves the system.
One policy without a facility is the default for every facility. A facility's own policy overrides it entirely. Changing a policy is a configuration event and is audited.

Was this helpful?

Report an issue with this guide