From Tool State to Scientific Action: Auditing Decision Provenance in Biomedical Research Agents

A scientific agent can give the same verdict while reading and citing different evidence. BioAgent-DF tests this separation using paired reference and candidate values saved at the same scientific computation site. It restores or transplants those values at their original coordinates, then runs the agent on healthy hardware. Across eight fixed-request-seed SciFact batches, Qwen3.5-4B shows repeatable citation changes in three capture--claim combinations, and Gemma 3 4B IT shows repeatable read and citation changes in four. These seven model--capture--claim combinations cover two claims. Full restoration recovers the reference evidence sets while verdicts remain unchanged. Verdict changes observed under the original prompt did not recur under the explicit-example prompt. Complementary tool and optimizer experiments distinguish numerical-state recovery, list recovery, and restoration of an agent's visible history. These results show that an audit should track both the evidence an agent uses and its final answer.