Review & Self Study — Defend the System Without Overselling It

Your final presentation should make it easy to distinguish what you built, what you tested, what failed, what responded, and what remains unknown.

Final review questions

1. What is the primary purpose of your system?
2. Which data and model assumptions are required?
3. Which model-quality evidence is actually held out?
4. How does a model score become a decision or action?
5. Which dependency failure did you test?
6. What fallback or safeguard responded?
7. Who has authority to interrupt or override the system?
8. Which claim in your original design became smaller after testing?
9. What did you not test?

A strong closing statement can include uncertainty:

The tested system classified the fixed evaluation cases, rejected the simulated invalid input, and entered fallback during the tested service failure. We did not test physical hardware, long-duration drift, adversarial inputs, or every network failure mode.

That is not weakness. It is a defensible boundary.

Self study

• Revisit one Microsoft lesson from Weeks 1–12 and identify how its model or technique would fit inside the Week 16 system map.
• Compare your capstone evidence with the responsible-AI questions from Week 12.
• Identify one new test that would most increase confidence in your current claim.
• Rewrite one sentence in your presentation to make its evidence boundary clearer.

Final rule

Evidence sets the boundary of the claim.