Actor-access as a separate relation
Event time and database record time do not necessarily tell us when a particular person could know something. EBR tracks actor-access evidence separately.
Research · Persistent AI memory · Temporal provenance
Evidence-Bounded Retrodiction
How can an AI reconstruct why the present became true without quietly rewriting the past with information that was only discovered later?
The problem
Later knowledge should improve history—not overwrite it.
Suppose a decision was made at 9:58 and an investigation happened at noon. The investigation may uncover evidence that helps explain what really happened. But if we later judge the 9:58 decision, we should not silently give the decision-maker information that was not available then.
EBR separates those questions so a persistent agent can use later evidence for reconstruction while preserving the historical information state against which earlier reasoning is appraised.
View 1Contemporaneously attested stateWhat was documented as accessible to the actor at the historical decision time?
View 2Reconstructed historical accessWhat does later evidence support that the actor had actually seen or accessed by that time?
View 3Retrospective evidenceWhat is known by the later audit point about the causal history that produced the outcome?
Research contributionTime Mechanism Uncertainty
An auditable protocol, not a claim that chronology proves causation.
Causal bridges must be supported
Temporal adjacency or similarity can nominate a hypothesis, but a candidate history must still account for the mechanism connecting required causal steps.
The missing story stays possible
Enumerated hypotheses are not forced to sum to 100%. An explicit unmodeled-history reserve remains active, allowing ties, incomparability and abstention.
Why it matters
A persistent AI should be able to say both: “Knowing everything we know now, this is the strongest supported reconstruction,” and “Given what the actor could reasonably have known then, this is how the historical decision should be judged.”
Those are different questions. EBR is intended to keep them different.