From Memory to Causal History: Evidence-Bounded Retrodiction for Persistent Agents

Persistent AI agents increasingly require more than the ability to store and retrieve prior information. They must distinguish between what was historically recorded, what evidence indicates an actor could access at a particular time, and what can be reconstructed retrospectively about the causal history that produced a present state. This paper introduces Evidence-Bounded Retrodiction (EBR), a protocol for persistent agents that separates three non-interchangeable evidence views: the contemporaneously attested epistemic record, later evidence supporting historical actor access, and retrospective causal evidence. EBR combines provenance-aware temporal memory, actor-access assertions, mechanism-gated candidate histories, contradiction-preserving retrieval, open-set uncertainty, and explicit abstention. It is designed to allow later evidence to improve causal reconstruction without rewriting the historical information state against which earlier reasoning and decisions should be evaluated. A deterministic synthetic prototype demonstrates why event time and database record time alone are insufficient proxies for historical epistemic state. The paper further proposes a falsifiable evaluation framework covering temporal leakage, actor-state fidelity, causal-path recovery, contradiction handling, historical-state invariance, and abstention. The work is presented as an architectural integration and research protocol for persistent AI agents rather than as a new general theory of causality, temporal databases, provenance, or abductive inference.