Human factors for AI operators. Studying AI agents as situated operators: perception, tools, claims, evidence, human belief, recovery.
Glossary
Inspectable definitions.
The recurring terms used on this site, each defined in one sentence, with a link to the evidence.
A final claim is credible only when the obligation, action trace, post-state and inspectable evidence are aligned.
The fit between a human operator and an AI agent in a real work loop — not only model performance.
A JSONL format for reduced field cases: claim, action, evidence, oracle, boundary. Signal for post-training and API teams.
Understanding what AI changes in real work — gestures, responsibilities, cooperation, agency — before prescribing a transformation.