Skip to content

AI test planning and knowledge guidelines

AI-assisted planning combines change intelligence, requirement baselines, and authorised project context into editable test-plan drafts. It helps teams select and justify coverage; accountable project roles retain ownership of the final plan and approval.

New test plan drafts and project knowledge guidelines each open in a dialog. Project settings also use dialogs to add release contexts, mappings and change manifests.

Change intelligence and baselines

Change manifests and path mappings describe which product areas a change may affect. Requirement baselines make later differences and maintenance impact easier to review.

  • Keep mappings as narrow as possible and investigate unknown or ambiguous paths.
  • Compare new results with the correct versioned baseline.
  • Separate proven impact, plausible inference, and areas that cannot be assessed.
  • Update change intelligence when repository structure or domain ownership changes.

Edit and hand off test-plan drafts

  1. Create a draft in the AI test-planning area and select the intended domain scope.
  2. Review proposed tests, priorities, rationale, dependencies, and identified gaps.
  3. Edit and save a new draft version instead of overwriting earlier decisions.
  4. Export the reviewed version or hand selected items to the next planning workflow.
  5. After hand-off, verify which project objects were actually created or assigned.

A draft is not an approved test plan. Project data changes only through a confirmed domain action.

Prepare a test-plan draft

Open AI test plans with an authorised administrator or project-manager role. Describe the objective and scope, test types, environments, risks, assumptions and entry/exit criteria. Add the relevant sources and packages. Specific input makes it easier to assess the generated draft.

After generation, check for missing requirements, unrealistic assumptions and untested environments. Save the edited version and use Markdown export for discussion where needed. The draft does not replace configuring and approving the executable test plan.

Narrow the selection for a run

Start AI test selection from the test-runs area. Choose a plan and optionally a release context, change manifest or focus requirement. Visible relationships and previous results determine the recommendations; no AI provider is called. The existing AI access permission is still required.

The budget counts planned cases, not variants or minutes. It ranges from 1 to 200. Directly affected mandatory candidates can be recommended beyond the budget. At most 200 cases can actually be selected. Multiple planned copies of one library case count separately; each requirement counts only once in coverage.

Coverage follows the currently checked cases. Its scope is the focus requirement, otherwise the visible requirements of the release context, otherwise all visible project requirements. This includes requirements without a case in the selected plan. The display distinguishes matching cases not selected from requirements without a visible candidate. Coverage means an existing relationship, not successful execution, complete variant coverage or approval.

With more than 2000 visible candidates, only a limited set is analyzed. The limit is shown explicitly and no complete percentage is displayed. A requirement without a known candidate remains unresolved in that situation. Narrow the scope instead of treating a limited analysis as a complete project assessment.

Hand only the reviewed selection to run start, then check the plan, environment and scope again. If a selected case has meanwhile been deleted or become unreadable, the handoff is rejected. Cases are neither silently dropped nor replaced by all cases. Open the selection again. Calculation alone creates no test run.

Read selection through MCP and REST

ztm_test_selection_read returns the same report, including rationale, limits and requirement coverage. testPlanId is required. Optional arguments are releaseContextId, changeManifestId, requirementId, budget (1–200, default 20) and selectedPlannedCaseIds. Omitting the ID list uses the default selection; an empty list explicitly selects nothing. IDs must be unique and currently visible. The REST operation is POST /api/v1/operations/ztm_test_selection_read.

A read-only project key is sufficient when the user has the necessary object and AI access permissions. The operation creates no run and requires no write key. Only visible requirements, cases and results are returned; a new request observes revoked permissions.

A limited budget can create deliberate coverage gaps. Record those gaps in the domain approval instead of treating the selection as complete evidence.

Maintain knowledge guidelines

Global and project-level knowledge guidelines provide approved terminology, constraints, and working rules. Project guidelines supplement global rules; they do not replace permissions, data zones, or scenario approval.

  • Write short, testable guidelines with an explicit scope.
  • Assign them only to relevant scenarios and approved provider profiles.
  • Version and review guidelines after process, product, or compliance changes.
  • Do not store secrets or unnecessary personal data in a guideline.