Build your first case
Follow the editor from sign-in through a small argument and export.
Edit on GitHubRebuild one branch of the Fair Recruitment AI teaching case. You are making a practice argument, not recording an audit of a real system. Keep the example's missing monitoring evidence in mind: a diagram can look complete while a claim remains unsupported.
Walkthrough
-
Go to
/login. Enter your email or username and password and choose Login, or choose Google or GitHub if that provider is configured. Complete sign-in to reach/dashboard. A Google connection may also support Drive features later, but it is not required for this activity.
-
On
/dashboard, choose Create new case. In Create New Assurance Case, enter Fair Recruitment AI as Name and a short practice description. Choose Submit. You arrive at/case/<id>. This creates a blank working case with a top-level goal namedG1; there is no template selector in this dialog. Import File on the dashboard is a separate way to make a case from JSON.
-
Use the pencil labelled Edit element on
G1. Its initial description is a prompt to describe your top-level assurance goal. Replace it with "The AI recruitment system makes fair and non-discriminatory hiring recommendations." Choose Update Goal to save the element edit. You now have a claim to test, not proof of it.
-
Use Edit element on
G1again and add two entries in Context: fairness is considered against the UK Equality Act 2010 protected characteristics, and this practice system screens applications for entry-level software engineering roles. Save with Update Goal. Context explains what the goal means and where it applies; it is not evidence that the goal is met.
-
On
G1, select the + button labelled Add child element, then choose Add Strategy from the popover. In Add Strategy, describe an approach based on preventing discrimination in the data and model. Choose Add. A strategy tells the reader how you intend to support the goal. The editor generates the element name, such asS1.
-
On the strategy, select its + (Add child element) button, choose Add Property Claim, and describe a testable statement: "The training dataset has been audited for representation across relevant groups." Choose Add. The editor generates a
Pname. In the element's edit dialog, review Assertion status. Asserted is the default; Needs support can mark a claim that still requires evidence. The status states your position, not an automated verdict.
-
On the property claim, select its + (Add child element) button and choose Add Evidence. Describe the audit report used in the teaching example and, if you have a suitable practice artefact, add its URL or reference in Evidence Link(s) (Optional). Choose Add. Do not invent a real report URL. A description alone tells a reviewer what evidence you expect; the underlying artefact must still be available and assessed.
-
Use the pencil labelled Edit element on the property claim to inspect its description and assertion status, then change wording if it overstates what the example evidence supports. Save with Update Property. The toolbar's Undo and Redo controls can reverse and reapply supported editor operations. Use each once and check the diagram again. The editor generates element names such as
G1,S1,P1andE1; add and edit dialogs have no name field. Imported names must follow the prefix rule. The toolbar also offers Reset Identifiers for people with edit access.
-
Use Export in the case toolbar. Export Case offers raw JSON, image, report and Backup to Google Drive sections. Choose a format to keep a review copy. The element edit and add dialogs save their own changes when you submit them; there is no separate whole-case Save button in this workflow. Check the exported result against the diagram before sharing it.
Sharing a working case gives named people or teams permission to collaborate. Publishing makes a public snapshot on Discover. You will practise both in Module 4, after you have considered how to justify and review claims in Module 3.
Check your work
Trace G1 through the strategy and property claim to the evidence. Say aloud what each connection contributes. Ask whether the evidence you recorded supports only data representation, or the whole fairness goal. The remaining gap is a useful prompt for the next module.