Reproduce the original learner story

1. Start with an empty browser sample. The landing page labels a fictional scenario, deterministic rules and no trained model or tracked behavioral history. No recommendations appear yet.
2. Choose Choose my starting point. Keep Make sense of data, 60 minutes, Foundational, with no completed courses. Choose Preview preferences. Inspect the before/after limits and Undo scope. Cancel first to observe no save, then preview again and Confirm change.
3. Expect this complete order: Read a small dataset → Tell the story behind a number → Ask a better learning question → Notice what the data leaves out. Open Why this: Read a small dataset: match 3/3, score 3 × 10 = 30, duration 30 ≤ 60, foundational, no prerequisites and all eligibility checks passing. Escape returns focus to its opener.
4. Select Exploration, inspect and confirm the policy preview. Expect ethics → story → questions → foundations. Why shows breadth × 10 + match; no model has been trained.
5. Choose Not for me: Read a small dataset. It immediately hides only that course, with local-save status. Restore Read a small dataset returns it if other limits qualify. Undo last change, preview and confirm, reverses just that last restoration; the snapshot is consumed. There is no history stack.
6. Edit preferences to 20 minutes, preview and confirm. No eligible courses appear. The idea-sketch course has zero data match and is not a fallback. Inspect catalog eligibility shows all eight courses and explicit exclusions. Edit my preferences to 60 minutes, preview and confirm, then For your goal recovers qualifying results; restore the hidden foundations course as needed.
7. Policy comparison is frozen, independent of current choices. Four eligible courses, three useful. Goal top-three useful 3/3 and coverage 3/3; exploration 2/3 and 2/3; both rejected 0/3. Inspect all eight judgments and the ineligible rejected outline.
8. Reset sample shows a clearing preview. Cancel preserves choices. Confirm reset returns to cold start. A reviewed Undo last change restores the prior valid choices, policy and hidden courses; confirmation consumes Undo. Reload removes Undo.
9. Refresh saved choices previews replacement of current tab work, then Confirm load applies the reviewed snapshot and clears Undo without offering restoration. A second tab changing saved content after preview causes confirmation to reject as stale, even at the same revision.
10. With malformed or unsupported saved data, inspect the warning: raw data is retained. Tab-only choices can continue. Only confirmed reset replaces corruption, including corruption introduced in another tab without a reload. An unavailable storage getter/read or failed write keeps current choices in this tab and warns of refresh loss.

This walkthrough demonstrates software behavior, not a human study or learning outcome. See Validation for executed checks and release gates.