Human Oversight and Reproducibility in Autonomous Materials Research: Authority, Auditability and Failure Containment

Autonomous materials research changes who or what selects experiments, but it does not remove scientific accountability. This review defines human oversight as an engineering control system with explicit authority boundaries, audit trails, intervention rules, independent replication and failure containment. Safety-focused self-driving-laboratory literature shows that increasing automation creates new requirements for safeguards and human-machine responsibility. For photopolymer development, the problem is amplified by coupled chemistry, light exposure, thermal behaviour, robotic handling and post-processing. The framework separates authorisation to propose, execute, modify and release; requires every autonomous decision to remain reconstructable; and defines conditions under which the system must stop rather than continue learning. Reproducibility is treated as a property of the entire closed loop, not merely repeated execution of the same software. ER-410.