DNative-Twin Replays a Committed Agent Decision and Finds Graph Structure Alone Cannot Explain Unobserved Tool State

arXiv 2609.03787·low signalXBlueskyLinkedInCopy link

A final agent output cannot show which evidence, tool state, rule, authorization or action path produced it. DNative-Twin records a committed agentic decision as a typed trajectory in a graph-native digital twin linking observed state, path followed and the authority behind the action, then re-executes the decision mechanism in isolation under declared conditions. The useful result is a stated limit: graph structure localizes represented changes but cannot determine the consequence of an unobserved tool state, and in a three-condition experiment with 300 injected instances, unresolved-divergence recall rose from 0 to 0.667 once replay-contract state was added and to 1.0 once verification results were too.