VERSIO-AI (Version Reporting for Scientific Investigation of AI Capability) is a 13-item reporting checklist for papers that evaluate off-the-shelf large-language-model capability on applied tasks. It is the reporting-standard companion to the empirical audit Frontier Lag: A Bibliometric Audit of Capability Misrepresentation in Academic AI Evaluation (Gringras and Salahshoor, 2026). Existing AI-reporting standards (CONSORT-AI, SPIRIT-AI, TRIPOD-AI, TRIPOD-LLM, DECIDE-AI, STARD-AI) cover clinical trials, diagnostic studies, and prognostic models built on top of AI. They do not cover the modal capability-evaluation paper: an off-the-shelf empirical probe of a named commercial LLM on an applied task. VERSIO-AI fills that gap, with thirteen items grouped into four blocks: model identification, tier and comparator context, elicitation, and interpretation. v1.2 is a candidate specification submitted for community revision. Each version of the specification receives a fresh Zenodo DOI under the same concept-DOI, so a concept-DOI citation always resolves to the latest version while a version-DOI remains fixed to the version cited.
David Gringras (Thu,) studied this question.