Vax or Fiction
A room game for talking through immunizations. One screen hosts, phones vote, and every answer reveal shows the public source behind it.
decision
A facilitator chooses immunization concepts for a group round. Players vote before the answer and source are revealed, giving the facilitator a concrete misconception to discuss.
baseline
A static quiz is the simpler alternative. Phone voting and concept rotation support group discussion, but no study here shows better learning than a static quiz.
review
A clinical reviewer still needs to verify the questions and source links before use. The facilitator can inspect the source at answer reveal rather than treating the AI-rated difficulty as a clinical judgment.
data
The portfolio uses fictional votes and public-source question content. It does not collect participant responses. Retention, access and deletion for a live multiplayer session are not evaluated here.
limits
The 150-question bank has AI-rated difficulty and still needs clinical review. Coverage checks do not establish factual accuracy, calibrated difficulty or learning gains.
Evaluation
Run 2026-10-30. 88 questions are core and 62 are alternates, across 13 deck modules. All 180 question appearances cite at least one source, 80 distinct URLs in total, mostly public CDC pages plus FDA, WHO and NLM. The questions have not had clinical review.
Rater agreement: Exact agreement | 127 of 150 (84.7%)
Rater agreement: Within one level | 150 of 150
Rater agreement: Linear-weighted Cohen's kappa | 0.81
Rater agreement: Records needing an adjudicator | 47
Final levels: Easy | 73
Final levels: Medium | 53
Final levels: Hard | 19
Final levels: Expert | 5
Tests: npm test (unit) | 1,331 tests, 1,331 pass
Tests: npm run lint | 0 errors, 29 warnings
Tests: npm run build | Passes (tsc -b and vite build)
- No clinical review. Questions cite public sources, but no clinician has checked the content, and guidance can change after a source's accessed date.
- Difficulty ratings come from two instances of the same AI model, so they are not independent judges, and they estimate difficulty, not correctness.
- Difficulty is not tuned from play. The game records no per-question answer data.