Skip to lesson content
Free course

AI Foundations / Module 7 / Module 7 capstone

Module 7 capstone

Capstone Rubric

Completion rubric. Use only supplied, public, fictional, or safely prepared synthetic material.

Rating: Not yet, Meets, or Exceeds. Completion requires Meets on every essential criterion. Exceeds is descriptive and does not compensate for Not yet elsewhere.

Essential criterionNot yetMeetsExceeds
Task and userVague task or missing user/successBounded deliverable, intended user, success, and exclusionsTests important edge cases without expanding scope
Input/data handlingUnclassified, unnecessary, or unauthorised dataEvery input classified, permitted, and minimised; prohibited data excludedSafer synthetic/reduced alternatives are documented
Tool choiceTool chosen by popularity or without fitAI/other-tool/human choice matches evidence, calculation, consequence, and authorityAlternatives and tradeoffs are tested clearly
Workflow and autonomyHidden stages or uncontrolled actionInputs, stages, outputs, owners, handoffs, autonomy, and confirmations are explicitFailure recovery and monitoring are integrated cleanly
Task specificationMissing outcome, sources, boundaries, format, or criteriaComplete usable specification with missing-information ruleSpecification is concise and handles meaningful variation
VerificationMaterial claims accepted without independent evidenceMaterial claims map to suitable sources, tests, calculations, or qualified reviewDisagreement and uncertainty are handled transparently
SafeguardsMaterial privacy, security, fairness, accessibility, or rights risk ignoredRelevant controls, owners, escalation, and stop conditions are presentControls are tested against realistic failure cases
Human authorityGeneric review or action before approvalNamed authorised person receives evidence, criteria, time, and ability to reject or stopDecision handoffs and records are especially clear
Test and revisionNo meaningful test or revisionSafe test exposes at least one defect; targeted change is retestedMultiple contrasting cases show improved robustness
ReflectionDescribes only success or confidenceExplains failure, evidence, changes, limits, and next improvementConnects limits to a realistic maintenance plan

Automatic revision conditions

The submission remains Not yet when any of these is present:

  • Real sensitive data is used without approved authority.
  • Material claims have no verification method.
  • A consequential action lacks prior meaningful human approval.
  • A known material risk has no owner, control, escalation, or stop condition.
  • The test record hides failure or does not show a revision.

Record your essential-criterion review

Rate every criterion. The capstone reaches the browser-only completion standard only when every rating is Meets or Exceeds.