Three published demo surveys, one per Design‑of‑Experiment (DOE) mode, each run live
through the real respondent flow (local widget on :9049) with screenshots and results.
All three use the same 8 items and ask "Which feature matters most/least to you?":
Price · Quality · Brand reputation · Design · Eco‑friendly · Warranty · Customer support · Delivery speed
| DOE mode | Document | Demo survey | Respondent link |
|---|---|---|---|
| Standard | 01-standard-doe.md | SV_yYzxH3QcNOdAYm0k | /s/2horBxGZZDeJ |
| Tournament | 02-tournament-doe.md | SV_WBp7VZbpnyoj4CnI | /s/gZblgzNo3bWZ |
| Top‑N Bandit | 03-top-n-bandit-doe.md | SV_wIO3EXLAKKRZrDjC | /s/wxD02ewKFVok |
Cross-cutting behaviour#
- Balanced first, then random. The Standard design fills each slot with the least-exposed item, so every item appears once before any item repeats. Screens looking "sequential" at the start is the balance working, not a missing shuffle.
- Per-respondent rotation. All three modes seed from the question plus the respondent's session id, so no two respondents get the same screens/order/positions. Stable within a session; preview uses a fixed id.
- Constraints. Item pairs that must never co-occur are authored in the question's Constraints panel. Honoured by Standard and Top-N Bandit; unavailable in Tournament (bracket groups).
Each recorded run below is therefore one respondent's rotation.
One‑line summary of what each run proved#
- Standard — every item shown exactly 3× across 6 screens (perfectly balanced).
- Tournament — 8 → 2 → 1 live; round‑1 winners advanced to the final (Brand reputation won).
- Top‑N Bandit — after the R model ran on 5 seeded responses, a fresh respondent saw the top items far more (Quality 6×, Price 5× vs Support 1×, Speed 1×, Eco 0×).
Screenshots live in img/.
Was this page helpful?