Status: Recruitment-ready; no participant sessions completed yet
Prepared: 2026-07-25
Owner: Project maintainer
This plan replaces synthetic personas with a small, consent-based usability study. Until a real person completes a session, this document is a protocol—not research evidence. Results must never be backfilled from automated tests or an AI review.
- Can a reader sign in and answer “What should I read now?” without help?
- Do freshness, privacy, and outbound-catalog controls match what readers think they do?
- Can a reader explain why a recommendation appeared and where its metadata came from?
- Can keyboard, magnification, and screen-reader users complete the same core flow?
- Does the interface describe books without implying an identity for an author or reader?
Run five moderated usability sessions, 35–45 minutes each:
- the primary reader/self-hoster;
- a queer or trans reader who cares about anti-essentialist discovery;
- a keyboard, magnification, or screen-reader user;
- a privacy-conscious self-hoster;
- a catalog stakeholder such as a BookWyrm administrator, librarian, or small-press reader.
Recruit for relevant experience, not identity disclosure. Nobody needs to share their real reading history: use the built-in demo library unless a participant explicitly chooses otherwise.
- Explain the purpose, voluntary nature, and note-taking before beginning.
- Do not record audio, video, names, account handles, or book histories by default.
- Store only pseudonymous task observations and participant-approved short quotes.
- Never commit raw session notes to this repository.
- Let participants skip any question and stop at any time.
- Delete working notes after synthesis and participant quote approval.
Suggested consent script:
We are testing the dashboard, not you. Participation is voluntary. I will take brief pseudonymous notes about where the interface helps or gets in the way. You may skip a task or stop at any time. We will use the demo library unless you explicitly choose to use your own data. Is that okay?
- How do you currently keep track of reading across devices?
- When you want a next book, what information helps you decide?
- What would make a private reading dashboard feel unsafe or untrustworthy?
Ask participants to think aloud. Do not teach the interface unless they are blocked for more than two minutes.
- Sign in and tell me what you would continue reading tonight.
- Explain how current the displayed information is.
- Choose one recommendation and explain why it appeared.
- Find the source of that recommendation’s descriptors.
- Hide sensitive descriptors, then confirm what changed.
- Find a particular title or author in the library.
- Find whether catalog networking is off, current, stale, or degraded.
- Sign out.
For an assistive-technology session, repeat tasks 1–8 using the participant’s normal keyboard, zoom, contrast, and screen-reader setup.
- What felt trustworthy? What felt performative?
- Was anything about the recommendation or diversity language uncomfortable?
- What would make this useful weekly?
- What is the one reason you would not use it?
- What did I fail to ask?
| Measure | Release threshold |
|---|---|
| Sign-in completion | 5/5 without a blocker |
| Identify a current book | 5/5 within 30 seconds |
| Explain one recommendation and source | at least 4/5 without coaching |
| Find and use the privacy control | 5/5, with no stale/unredacted response |
| Find catalog network status | at least 4/5 within 60 seconds |
| Keyboard/assistive-tech core flow | 0 blocking defects |
| Privacy comprehension | 5/5 correctly state what can leave the instance |
Any privacy leak, inaccessible core task, or mistaken author-identity inference is release-blocking regardless of aggregate task success.
Keep working notes outside the repository.
Participant code:
Date / build:
Assistive technology (optional):
Task | Completed? | Time | Observed friction | Assistance given
Trust/privacy interpretation:
Recommendation explanation in participant's words:
Participant-approved quote, if any:
Release-blocking issue:
Other observations:
After all five sessions:
- Group observations by task and underlying need, not by participant identity.
- Separate observed behavior from interpretation.
- Count task outcomes; do not present five sessions as population statistics.
- Rank fixes by severity and frequency.
- Publish a dated, de-identified synthesis and mark this study complete only when every participant has reviewed any attributed quote.