Spec: the placement test
Builder: Devin. Reviewers: Devin Review and Claude. Bob merges.
What it does
A learner answers about 15 minutes of questions and is told which level to start at (100, 200 or 300) and which gear may suit them (plain, standard or technical). They can always choose differently.
Rules
- Static page in
apps/placement/, served from GitHub Pages. No account, no server, no analytics. - Answers never leave the browser and are not stored after the tab closes. Say so on the first screen, the same promise the Gate Check makes.
- Items come only from module quizzes: every
quiz.jsonitem marked"placement": true. Never copy item text into code; the build reads the module files. - Adaptive: start with level-200 items; two right in a row moves up a level, two wrong moves down; stop after at most 20 items or 15 minutes. Level 500 is not placed by test.
- The result screen shows the level, a one-line reason ("You knew how models learn, but not how to check their output"), the suggested gear, and a link to the first module of that level.
- Accessibility: WCAG 2.2 AA, full keyboard use, works with VoiceOver and TalkBack, respects reduced motion, readable at 200% zoom.
- Works offline after first load.
Done when
- Unit tests cover the routing rule (up, down, stop) and the result for each level.
- With only level-100 items available it still places sensibly (everyone starts at 100) and says the higher levels are not built yet.
python3 tools/check_modules.pyand the app's tests pass in CI.