Pipeline

A human-led, AI-assisted loop: investigators design and own the study; AI tools execute and review under their direction; experts approve every step.

1 · Investigators (authors) — study design & orchestration

2 · GPT-5.5 (LLM-as-judge) — independent item review; fixes loop back to investigators

3 · Claude / Cowork — builds & operates the approved instrument in REDCap (MCP / browser automation)

4 · REDCap — consent / RODO + de-identified capture

5 · AI-assisted dashboards (Python / React) — aggregated analytics

Human-in-the-loop at every step · Open-science layer: GitHub · Zenodo DOI · OSF · GitHub Pages

Reviewed loop. Investigators draft items; GPT-5.5 returns structured flags (bias, clarity, validity); investigators revise; Claude / Cowork then builds the approved instrument in REDCap. Nothing goes live without investigator approval.

Governance. The investigators own study design and orchestration. AI tools assist, execute, and review under their direction; they augment and are verified — they do not replace expert judgment, and are not co-authors.

Full research architecture

End-to-end view from governance and the evidence gap, through investigator-led instrument design (with AI-assisted drafting and GPT-5.5 review), REDCap capture, scoring, analysis, the planned paper series, and the open-science layer.

DIVE-PL research architecture: governance, evidence gap, questionnaire design, instrument domains, REDCap capture, processing and scoring, statistical analysis, scientific outputs, and the open-science layer.

Source (Mermaid): docs/architecture.mmd. Investigators own design & orchestration; Claude / Cowork builds and operates the REDCap instrument; GPT-5.5 reviews items; human-in-the-loop oversees every stage.