Keeping the space open: the reason and method of SE4RA, and its common ground with disclosure, provenance, and assurance in AI-enabled science

This position paper in the GCPA-SIDCER Science, Policy, and Ethics in Conversation Series sets out the reason and method of SE4RA, a CoARA Boost Cascade Funding project that pilots the instruments of the CoARA-ERIP working group as institutional self-evaluation in research assessment across four scales, from the individual researcher to the institution. It argues that research assessment reform cannot be imposed on an institution from outside and must begin from within, from what the institution values, and it connects this method to the debate on disclosure, provenance, and assurance in AI-enabled science: a metric and a fluent machine output are two forms of the same premature closure of the space in which judgement, reflection, and conscience reside, and assessment is the layer that sets the cost of keeping that space open. The paper situates the argument within the United Nations International Decade of Sciences for Sustainable Development (2024-2033) and the Global Digital Compact, and in the capacity-building work of CAI3R-science within the AICDN network supported by UN-ODET. AI, provenance, and assurance statement This publication develops from the work of the CoARA Working Group on Ethics and Research Integrity Policy for Responsible Research Assessment in Data and Artificial Intelligence (CoARA-ERIP) and its SE4RA pilot, and from conversations in the EOSC-Future/RDA Artificial Intelligence and Data Visitation Working Group, the RDA Sharing Rewards and Credit Interest Group, and the FORCE11 Scholarly Communication Institute course of 2026. Handwritten notes, drafting in Word, and presentations and audience discussions provided the ideation and the conceptual and structural background. Large language models, including Microsoft 365 Copilot, ChatGPT-5.5, DeepSeek Expert, and Claude Opus 4.8, also contributed, under the author's direction, for structuring, drafting, and language editing. All content, claims, citations, and conclusions were set and verified by the named author, who takes full responsibility for the work. AI was not used to generate citations, data, or normative conclusions. The full structured statement is deposited with the record as a machine-readable object.