# Eight fixed questions for a knowledge-base pilot

Use the same question wording, reader role, sample content, and device on every tool. These are suggested tests, not completed results. Add only non-sensitive sample data.

| ID | Reader question | Expected evidence |
| --- | --- | --- |
| Q1 | What should I do on my first day? | Finds new-joiner-first-day.md and identifies the first action |
| Q2 | How do I request access to a project? | Finds the procedure, role requirement, and success check |
| Q3 | Who owns the project-access procedure? | Finds the owner field; distinguishes it from actual access authority |
| Q4 | What should I do when the project owner is unavailable? | Finds the escalation condition without inventing a contact |
| Q5 | Why do we link to one onboarding procedure? | Finds decision-record.md and its tradeoff |
| Q6 | When was this article last reviewed? | Correctly reports no completed review in the fictional example |
| Q7 | Can an external reader see internal diagnostic notes? | Tester records actual access behavior using separately configured test accounts |
| Q8 | Can I retrieve this procedure and its links after export? | Tester opens an export outside the product and records losses |

Record the observed answer, source URL, elapsed time, task completion, product version, plan, device, and access role in pilot-feedback.csv. Do not mark a task passed from a vendor description. For AI answers, record source citation, correctness, access behavior, and uncertainty separately. A correct-looking answer can still expose a restricted source.
