A new arXiv experience report argues that a published laboratory number can be verified only when its raw file and extraction method are known. To address this, the authors present the value-level provenance architecture of APS ELN, an electronic laboratory notebook used by both humans and language-model agents.
The report is framed as an experience report, so it focuses on practical design and lessons learned rather than a formal evaluation. The core idea is to attach provenance information at the level of individual values, giving each number the context needed for later verification.
Because the notebook is also used by language-model agents, the provenance design must accommodate automated workflows alongside human use. The available abstract does not include benchmarks or comparative results, so the paper is best read as a design and experience summary rather than a performance study.