Answers are grounded in this paper and cite their sources.
Public comment draft, open until 1 December 2026. A specification in development for checking, from outside, that an AI system's activity records are complete for the scope the operator declared and have not been altered. It focuses on omission: counters, reconciliation against records the operator does not control, liveness, clock independence and sampling. It sits beside existing receipt and logging formats such as SCITT and does not replace them. It is not an implementation-ready test suite, and it issues no certificates.Drafting history: circulated in draft as v1.1 (3 October 2026), v1.1.1 and v1.1.2 (31 October to 1 November 2026); requirements C1 to C19 first appeared in 10.5281/zenodo.23546436. Demonstration code and frozen test vectors are planned before 15 November 2026.Comments: comments@sts-forensics.com (subject "O4M Record Tests"). Comments and the editor's replies will be published in a comment log. SHA-256 (source text, .md; this is the reference hash): 8758174e927e971391443ce9142f9c78bafad21ad52f63613d2443a422869d3d (64 characters). SHA-256 (reading copy, .pdf): c692969b547d780d51975104b4f8a77983ab67806b85c749325f3bb17b1f37da (64 characters, ends da). AI assistance: drafted and edited with Claude (Anthropic); the editor reviewed every requirement and takes responsibility.
No takes yet. Share an insight, caveat, or question.
Gareth Wong (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: