A reproducible AI tool-chain methodology for building survey / questionnaire research studies.
Status: scaffold (v0.1.0) · Dual-licensed (MIT for code, CC BY 4.0 for content) · Case study: recreational divers in Poland
Authors: Łukasz Minarowski, MD, PhD (ORCID) · Jakub Matuk, BEng (ORCID) · Maksymilian Lech — Department of Respiratory Physiopathology, Medical University of Białystok, Poland
DIVE-PL is a methods / framework project: it specifies and demonstrates an end-to-end, fully reproducible pipeline for designing, fielding and analysing questionnaire-based research using an integrated chain of AI tools, wrapped in an open-science layer that makes every step independently verifiable.
The contribution is the integration, not any single technique. Each component has prior art (see docs/related-work.md); to our knowledge the combination of all four components plus a full open-science reproducibility layer has not been published as a single survey-study methodology (literature scan, mid-2026).
- Investigators (authors) — design and orchestrate the study: research questions, protocol, item content, registration, interpretation. This is human work.
- GPT-5.5 (LLM-as-judge) — independently reviews survey items for clarity, bias, double-barrelled wording, and content-validity relevance; fixes loop back to the investigators.
- Claude / Cowork — builds and operates the approved instrument in REDCap via MCP / browser automation, under investigator direction.
- REDCap + AI-assisted dashboards (Python / React) — de-identified electronic data capture (consent / RODO) and aggregated analytics, published openly.
Human-in-the-loop at every step: AI tools assist, execute, and review; investigators decide and approve. AI tools are not co-authors.
See docs/architecture.mmd and METHODOLOGY.md.
| Channel | Purpose |
|---|---|
| GitHub (this repo) | methodology, code, instrument definitions, prompts |
| Zenodo | archival DOI per release (citable artifact) |
| OSF | project page + preregistration of the case study |
| GitHub Pages | live public site + analytics dashboard (docs/) |
The open-science stack is standard FAIR practice — not claimed as novel; it is what makes the integrated framework reproducible.
DIVE-PL/
README.md
METHODOLOGY.md # the framework, stage by stage
LICENSE # CC BY 4.0 (content)
LICENSE-CODE # MIT (code)
LICENSING.md # which license applies where
CITATION.cff # how to cite
AI_USAGE_DISCLOSURE.md # transparency: which AI did what
RELEASES.md # changelog / release notes
.zenodo.json # Zenodo deposit metadata
.gitignore / .gitattributes
docs/
index.html # GitHub Pages site (Overview + nav)
survey.html # Case study — the DIVE-PL diving survey (live)
methodology.html pipeline.html related-work.html
reproduce.html limitations.html cite.html
assets/ # style.css, favicon.svg
.nojekyll
case-study-survey.md # survey description (aim, sections, hypotheses)
related-work.md # component -> nearest precedent -> our difference
architecture.mmd # pipeline diagram (Mermaid)
open-science-checklist.md
thesis-outline.md # engineering-thesis (PL) IMRaD skeleton
CLINICAL_TRIALS.md # registration details
src/
README.md # code map (agent / reviewer / redcap / dashboard)
data/
README.md # data governance (RODO/GDPR) — NO PII in repo
The framework is demonstrated on a cross-sectional online survey of recreational divers in Poland: health status, physical fitness (VO2max via the University of Houston Non-Exercise Test), safety knowledge (DCS, barotrauma, safe-diving rules), risky behaviours, and self-reported adverse events. Aim, design, instrument sections, hypotheses (H1–H3), endpoint and analysis are described in docs/case-study-survey.md (live: Case study page).
The study is ethics-approved and registered (see docs/CLINICAL_TRIALS.md). No participant data are stored in this repository.
See CITATION.cff and DOI 10.5281/zenodo.20977890. Authors are listed under the title above.