Data provenance
Full source of truth: research/data_provenance.md. Summarized here.
AI-generated stimuli
No human dataReal LLM stimuli come from a self-hosted OpenAI-compatible gateway, model name Yuu no Sekai (a self-hosted Qwen3-8B deployment -- not a frontier commercial model, and never described as one). Every stimulus file records its model, prompt version, temperature, draw index, and generation timestamp, and is tagged "source": "llm_generated". If the gateway is unreachable at generation time, the pipeline falls back to researcher-written deterministic fixtures, tagged "source": "researcher_fixture" instead -- the two are never conflated downstream (see /stimuli).
Synthetic participants
Simulated pilotAll behavioral numbers on this site before real human recruitment come from analysis/simulate.py, a rule-based agent population (rational, AI-trusting, AI-skeptical, price-sensitive, brand-insensitive, framing-sensitive, confidence-sensitive, and noisy archetypes). Every synthetic row is written with "data_source": "synthetic" and a fixed, recorded random seed. This data exists only to validate that randomization, persistence, and the analysis pipeline work end-to-end, and to check that a known injected effect is recovered correctly -- it is never merged into any table that could later hold real participant rows.
Real participants
No human dataNone have been recruited. When/if recruitment begins, this section will describe the recruitment platform, the consent version shown, IRB/ethics status (currently none), compensation, and the exact experiment-version commit hash each participant saw.
Storage & migration status
Pilot/dev persistence is SQLite at app/data/biasmarket.db (file-based, git-ignored). This is adequate for pipeline validation and small pilots, but is not durable Vercel-deployment storage -- Vercel’s filesystem is ephemeral. Before any real, paid deployment collects participant data, the persistence layer must migrate to a hosted Postgres instance (e.g. Neon) using the same schema. That migration is an open item in docs/HANDOFF.md, not yet started.
Every claim this project makes -- including the ones on this page -- has a row in research/claims_registry.md, with a status that is set by the pipeline that produced the underlying artifact, never hand-typed on the site.