For EWAI 0.2.8 · Install or update EWAI
Worked examples
These compact scenarios demonstrate how to apply EWAI proportionately. They’re illustrative decision patterns, not specifications to copy unchanged or exhaustive phase checklists. A stage omitted from an example isn’t waived; follow the current delivery’s required gates.
For a step-by-step exercise with observable results, try your first session and first delivery.
1. Start a new internal application
Situation: An operations team needs a small case-triage tool.
Approach:
- The Product Owner supplies the current process, users, desired response-time outcome, data classification, and non-goals.
- Guided Discovery engages product, operator, accessibility, and security perspectives as the sections change.
- If the organisation provides an applicable internal-application Blueprint, the team reviews it, accepts its required security and testing modules, and declines an irrelevant public-service module. Otherwise, Discovery continues without an Organisation Blueprint.
- Review shows the exact standards and project personas to be created.
- A named owner approves Discovery.
- The first intent delivers one journey: view and assign an untriaged case.
- Manual QA uses safe test records and checks keyboard access, permissions, exception handling, and operational visibility.
Key lesson: Start with one useful journey and the evidence needed to operate it, not a generated backlog for the entire imagined product.
2. Onboard a poorly documented legacy system
Situation: A team inherits a service with sparse documentation and several years of history.
Approach:
- The owner describes why the service exists, its users, known failures, and what must not be inferred from current code.
- After briefing and source permissions are agreed, the owner accepts the offer of Archaeology. It maps repositories, entry points, integrations, data, tests, and deployment.
- The inferred purpose conflicts with one current batch process. The discrepancy is presented before deep analysis.
- The owner confirms that the batch is a temporary workaround, not intended behaviour.
- Operations and data personas join the relevant deep passes; their output remains hypothesis until supported.
- Reviewed findings become canonical SPECS, while the workaround becomes a bounded migration intent.
Key lesson: Repository behaviour is strong evidence of what exists, not automatic authority for what should continue.
3. Deliver a user-visible feature
Situation: Users need to export a filtered report.
Approach:
- Intent records the user outcome, supported filters, data permissions, accessibility, performance expectation, and export retention.
- The repository index identifies the query, permission, UI, and audit-log blast radius.
- The plan creates one vertical slice through UI, service, authorization, export generation, and evidence.
- Build approval names that scope.
- Automated tests cover permissions, filters, empty results, volume, and formula-safe output.
- Standards review checks data handling and existing patterns.
- Manual QA exercises an authorised and unauthorised user, keyboard flow, a large report, and the downloaded result.
Key lesson: User-visible acceptance and data-boundary testing belong in the same slice as implementation.
4. Make a low-risk technical-debt change
Situation: A duplicated parser should be consolidated without changing behaviour.
Approach:
- The intent states the maintainability outcome and an explicit no-user-visible-change constraint.
- Index and tests identify callers and observable contracts.
- Reconcile confirms both copies currently behave the same—or records the differences.
- The plan uses characterization tests as the first failing or safety evidence.
- Build changes the smallest shared boundary.
- Manual QA is proportionate: a focused maintainer walkthrough plus one representative user journey.
Key lesson: Low risk can reduce ceremony, but it does not justify inventing equivalence or skipping impact evidence.
5. Apply a mandatory organisation Blueprint
Situation: A regulated project must use the organisation’s reviewed baseline.
Approach:
- Guided Setup shows publisher, version, compatibility, required modules, dependencies, digest, and destinations.
- Required modules cannot be deselected. Optional modules are chosen from actual project applicability.
- A compliance persona surfaces questions, but a qualified owner confirms applicable obligations.
- Existing destination conflicts stop application.
- The team reconciles the existing standard rather than deleting it.
- Named approval materialises organisation standards, project personas, receipt, and pipeline pin.
Key lesson: Mandatory organisational input still requires visible project consequences and accountable adoption.
6. Use project and premium personas together
Situation: A payments project has a local settlement-operations persona and access to managed specialist personas.
Approach:
- Check-in reports entitlement and that the premium library is installed; no download occurs.
- In user-outcome sections, a product lens leads.
- In data and assurance sections, the relevant project payments persona and a matching premium specialist are deliberately considered.
- The UI displays both names, tiers, concerns, and engagement reasons.
- The project persona points to existing local operating evidence; it doesn’t invent new facts. The premium persona challenges specialist gaps.
- A real payments owner resolves the decision.
Key lesson: Mixed tiers improve perspective coverage, but relevance, evidence, and human authority remain separate.
7. Deliver a security-sensitive capability
Situation: A public feature processes sensitive personal data.
Approach:
- Discovery records classification, purpose, access model, retention, jurisdictions, threat context, and operational ownership.
- Candidate compliance findings remain triage until qualified owners confirm them.
- Security and privacy standards become explicit Build constraints.
- Plan includes abuse cases, authorization boundaries, logging restrictions, dependency risk, incident visibility, and deletion behaviour.
- An independent security review is used when configured; if unavailable, the gap is recorded rather than marked passed.
- Manual QA uses non-production test data and verifies denial, recovery, and audit behaviour as well as the happy path.
Key lesson: More AI review does not replace data governance, qualified interpretation, or operational readiness.
8. Roll a Blueprint update across projects
Situation: An organisation publishes a new major Blueprint version with a changed mandatory API standard.
Approach:
- The publisher releases an immutable version with compatibility, affected modules, migration guidance, and deprecation dates.
- Catalogue owners identify projects pinned to the earlier version.
- Each project compares its receipt and materialised standards with the candidate release.
- One project adopts immediately, one records a time-bound exception, and one remains unaffected because the module does not apply.
- Each decision is approved and evidenced locally.
- The old source remains available long enough to interpret historical receipts.
Key lesson: A shared release creates review work; it does not create permission to overwrite every consuming project.
Adapt an example safely
For your own project, replace:
- users and outcomes with direct evidence;
- risk and data classifications with confirmed context;
- personas with the installed relevant catalogue;
- Blueprint and module choices with reviewed local packages;
- tests and Manual QA with observable acceptance criteria;
- approvers with people who have real authority.
Retain the boundaries: personas advise, SPECS holds durable truth, reusable inputs require review, Build requires explicit scope approval, and Manual QA remains human.