Beyond Persistence: A Causal Framework for Developmental Continuity in Long-Lived AI Agents

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Persistent-agent research increasingly addresses durable memory, authorized lineage, runtime migration, and behavioral stability. These capabilities do not establish that prior experience remains causally active in what an agent becomes next. We distinguish persistence, causal inheritance, and developmental continuity. Developmental continuity obtains when formative history changes an inherited response-and-update disposition F that subsequently shapes novel behavior or further development. F is a functional causal abstraction; Q denotes an architecture-specific candidate realization, and S a preregistered observable developmental signature. We separate causal genealogy from authority genealogy and propose PLI-2, a falsification-oriented protocol based on different formative genealogies, controlled intervention on the explicit-policy channel, history occlusion during critical testing, held-out prospective signatures, and intervention on candidate mediators. IIPSEON is treated as an internal instrumented testbed rather than as evidence of architecture-general validity. This paper is conceptual and methodological: it specifies causal estimands, identification assumptions, falsifiers, and portable experimental protocols, but does not report a completed empirical demonstration of developmental continuity.