v2 draft outline
Outline only. Bullets list what this page will eventually cover.
Module 4: Capstone, Build Your Factory
Sizing: 5 days (Week 4)
Outcome
- Build a software factory from scratch for your own project
- The factory is the deliverable; what you build through it is secondary
What you'll do
- Pitch your project and your factory plan
- Build the factory around your own project
- Push real work through it
- Demo the factory and what it caught
Hallmarks of a factory
- Quality gates
- Dependency automation
- Test generation
The bar scales
- Nontechnical learners show:
- The quality-check instinct: you run a check before shipping
- The gaps an agent found in your project
- Technical learners show:
- The full hallmarks, working and demonstrated
- Where they drew their factory vs. hacktory line, and why
Where OSV-Scanner runs: your choice
- In the good repo OSV-Scanner runs in pre-commit; in the Capstone you choose where it runs
- Options to weigh: pre-commit, the PR gate, a scheduled job, or more than one
- Cheap, fast checks run early; slow or expensive ones can wait
- Defend your placement: what it costs, what it catches, what you give up
- Practice repos: Emerald Grove Hotels (good) and Chaos Emerald Hotels (bad) are the references, not the project
Stretch: the gotcha scenario
- A scenario that breaks the machine if only deterministic checks or only AI checks are used
- Show that you needed both, and where each one fails
- Details to be designed (see TODO)
Gate
- No new gate from the course; your factory is the gate
- Your own work passes through every tier you built
Anti-patterns
- Building the product and skipping the factory
- Audit once and forget
- A factory nobody else can reproduce and audit
- Handing over first-pass output
- Scope sprawl
Demo
- Process, not just output: what the factory checks, what it caught, what it missed, what you would add
- Defend your work: explain each choice if questioned
From v1
- Carries over: pitch before building, final demo
- Changes: the factory, not the project, is what is judged
TODO
- TODO: what a defensible OSV-Scanner placement looks like, so facilitators can judge a choice they did not expect
- TODO: concrete rubric for the nontechnical and technical bars
- TODO: what "test generation" means here (agent-written tests, property tests, or both) and how it is judged
- TODO: design the gotcha scenario and decide whether the facilitator or the learner triggers it
- TODO: deploy target: the decisions log said learner's choice (AWS, Azure, Vercel or the Lab); confirm that still holds, and how much support each target gets
- TODO: whether learners without a project get a default (for example, a starter app)
- TODO: whether learners re-audit with the same check used in Modules 1 to 3