AI agents increasingly gather evidence, invoke tools, apply constraints and produce decisions that may be enacted by people or software. The final output alone cannot reveal which evidence, tool state, rule, authorization or action path generated it.
To address this, we introduce DNative‑Twin, a graph‑native digital twin. It records a committed agentic decision as a typed trajectory and re‑executes the decision mechanism under declared conditions. The graph links the state observed by the agent, the path taken, and the authority behind the resulting action.
The twin synchronizes this information, replays the mechanism in isolation, and compares it under controlled changes. We instantiate the framework in enterprise decision processes, using three public process logs and controlled replay suites for validation.
Experiments reveal a specific limitation: the graph structure can localize represented changes but cannot determine the consequence of an unobserved tool state. In a three‑condition controlled experiment with 300 injected instances, recall of unresolved divergence rose from 0 to 0.667 when replay‑contract state was added, and to 1.0 when verification results were also available; the held‑out set contained no critical‑class instance.
Across 500‑5000 BPI 2020 cases, median end‑to‑end time increased from 0.794 to 8.889 seconds. The results separate the roles of graph structure, replay context, and verification evidence, providing fine‑grained insight for reviewing decision mechanisms.
Review: DNative‑Twin demonstrates the feasibility of achieving decision traceability and replayability through graph‑based digital twins, offering a practical tool for auditing and debugging complex AI agents, yet it still needs to address inference over unobserved tool states.