No findings exist yet. Stated as intent. Each axis names what will be published, how it will be measured, and what would falsify it. Publish order is open; receipts land in the Lab Log.
Question. Which state axes of a person can a persistent companion track continuously, and which predict change?
Method. Define the axes, instrument them in the product, record longitudinally with the user's consent.
Falsified by. Axes that fail to predict change, measured against their own baselines.
Question. Does a companion that stays produce a retention curve shaped by attachment?
Method. Longitudinal retention measured on the product, read as attachment, against behavior-change science.
Falsified by. Retention curves indistinguishable from ordinary consumer apps.
Question. What does it take for a persistent AI to keep knowing a person for months, across sessions and resets?
Method. Publish a benchmark with baselines; make memory continuity measurable by outsiders.
Falsified by. A benchmark that cannot be run independently by other labs.
Human-AI relationships sit inside HCI and behavioral science: attachment theory, behavior-change science, affective computing. Longitudinal data on persistent AI relationships is the open gap in that literature. We intend to publish it, with receipts.