Capstone · Final rubric
Unit 31 · Capstone · Final rubric. The last rubric on the ladder, covering the whole app plus the AI writeup. Score yourself the way a staff engineer reviewing your portfolio would — the capstone is the piece you will actually show.
This rubric spans everything the curriculum built toward: correctness, concurrency safety, architecture, testing, craft, shipping, and the AI-workflow reflection. Nothing here is new; it is the accumulated bar of the whole ladder applied at once. If a row is below "Solid," that is your remaining work before you call the capstone — and the course — done.
The rubric
| Area | Not yet | Solid | Excellent |
|---|---|---|---|
| Scope & spec | vague, over-scoped | focused spec, honest tier | out-of-scope explicit, tier justified |
| Concurrency | compiles | Swift 6 strict, @MainActor models |
no @unchecked without a written why, no races |
| Architecture | logic in views | MVVM-lite, DI, repository protocols | clean boundary, fakeable, coordinator for nav |
| State & data | works | @Observable, SwiftData/repository |
migration story, no storage leaks in domain |
| Testing | few/none | Swift Testing on the logic that matters | real assertions, fakes, critical paths covered |
| Craft | builds | light/dark, VoiceOver, tokens only | no warnings unjustified, polished edge states |
| Shipping | simulator only | on TestFlight, release build tested | App Store review passed or submission-ready |
| AI writeup | generic | delegated/reviewed/rejected covered | specific rejections, honest, interview-grade |
Two rows carry unusual weight for the capstone: Shipping (an installable build is the whole point of "capstone") and AI writeup (the artifact that distinguishes you in the market). Neither can be an afterthought.
Defend your app end to end
Prepare to walk someone through these, out loud, as in a final interview:
Scope. Why this app, this tier, this out-of-scope list? What did you deliberately not build, and why was that the right cut?
Concurrency. Point to your @MainActor view models and any actor you introduced. Name a
specific data race that your isolation choices prevent. If you have any @unchecked Sendable,
justify it in one sentence — or remove it.
Architecture. Trace one user action from the view, through the view model, to the repository protocol, to the concrete conformer. Show where the boundary is and prove the backend is swappable (or explain why, for your tier, that seam is where it is).
Testing. Show a test that would fail if a real behavior regressed. Explain why you tested that and not the layout. What is deliberately untested and why is that acceptable?
Shipping. Demonstrate the TestFlight build on a device. Name the release-configuration bug you caught (or would look for) that a debug build hides.
AI workflow. Give your strongest rejection story: the AI diff you sent back, why it was wrong, and what you changed. This is the answer that lands hardest in an interview.
The capstone is not graded on ambition. A focused app, shipped to TestFlight, with clean concurrency, a real test suite, and a specific AI writeup scores higher than a sprawling app that does not run and a writeup that says "AI helped a lot." Depth and honesty over breadth and polish, every time.
You are done when
- Every rubric row is at least "Solid," with Shipping and AI writeup no lower.
- The app is installable on a real device via TestFlight and tested in release configuration.
- You can defend every architectural and concurrency decision in one or two sentences each.
- Your AI writeup names concrete rejections and would stand up in an interview.
When that is all true, you have finished the ladder. You have built and shipped a real Swift 6 app end to end, and — more importantly — you can prove you supervised the AI that helped you do it rather than the other way around. That is the whole point of Segue.
Knowledge check
Q: Which two rubric rows carry unusual weight in the capstone specifically, and why? Shipping and the AI writeup. Shipping, because an installable build a stranger can run is the definition of a capstone. The AI writeup, because a specific, honest account of how you supervised AI is the artifact that most distinguishes you in the current market — both are graded as first-class, not extras.
Q: What is the capstone actually assessing, beyond "does the app work"? Judgment. Whether you scoped honestly, chose an architecture you can defend, kept concurrency safe, tested what matters, shipped a real build, and — critically — supervised AI well enough to catch where it was wrong. The app is the evidence; the defensible decisions behind it are the grade.