About the Human–AI Trust Lab

What this is

An open research platform studying appropriate reliance on AI recommendations: an experiment engine, a statistics pipeline, and this site, built to the standard of a workshop or conference submission. The manuscript and methodology pages are the canonical description of the design; this page covers project logistics.

Technology

Next.js/TypeScript for the experiment and site; PostgreSQL for persistent participant data (Neon, via Drizzle) with an in-memory fallback for local development; Python (pandas, statsmodels, scikit-learn, matplotlib) for analysis and figures; tectonic for the LaTeX manuscript; Vitest and Playwright for testing; GitHub Actions for CI.

Privacy & deletion

We collect the minimum necessary: your responses, confidence ratings, timestamps, and condition assignment. We do not collect names, emails, or IP addresses. Two demographic questions are optional. Every session receives a deletion key; using it (from the debrief screen, or by contacting us with the key) permanently removes the participant, their sessions, and all trial rows via a cascading delete, exercised by an automated test on every change to the platform.

Contact & source

Repository: github.com/Gariyuuu/human-ai-trust-lab. Issues and questions about the study can be filed there.

Status

Engineering complete for experiment v1: engine, simulator, analysis, tests, and manuscript pipeline. No human-subject data have been collected; live collection requires an IRB (or equivalent) review before launch.