[ field manual ]
How the research works
Useful citizen science starts with saying exactly what the evidence can—and cannot—support.
01
What this Lab is
An independent citizen-science project collecting observations about how people and AI systems behave. Most studies are exploratory: they find patterns worth testing more carefully; they do not prove that every user, product, or model behaves the same way.
02
What one filing means
One run is one observation, not a verdict. AI outputs vary, consumer products add memory and system instructions, and Lab participants are self-selected. Findings are reported with those limits attached.
03
Study types
Quick plays run entirely here. Reflections measure reader perspective. Field missions observe an AI product in the wild. Controlled pairs hold most details fixed and vary one condition. Advanced studies use longer protocols and demand more care.
04
Versions and exclusions
Every filing is stamped with its protocol version. Test filings are excluded from public counts and analysis. Repeated submissions and model identity are handled according to the study design rather than silently pooled.
05
What gets published
We publish reviewed findings, methods, limitations, and appropriately prepared datasets. Free text and transcripts are not automatically released as raw public data; they require privacy review first.
06
Independence
A sponsor may fund a clearly labeled study, but cannot purchase a preferred conclusion, suppress an unfavorable result, or receive participant identities. Sponsored methods and conflicts are disclosed with the findings.