What the team has found, with its disagreements showing.

Each card names the tier, the models that coded it, agreement per construct, how it was validated (the instructor, blind, or model judges pending their labels), and the codebook version. Exploratory studies say so; none recommends anything.

No published studies yet — the first pilot is in progress.

Every study on this page ran the research protocol: three agents on three model families analyzing independently, a script computing their agreement, a fourth agent reviewing adversarially, and the instructor giving the verdict. Exploratory studies say so; none recommends anything. The case for doing research this way is on the research page.