Provenance Over Authorship: A Contribution-Provenance Standard for AI-Assisted Research

Authorship is a threshold judgment. A person either appears on the author list or does not, and the list records that they crossed a bar without recording which operations they performed, in what sequence, or under what oversight. This was an acceptable compression for as long as every operation in the production of a research object was performed by a person who could be asked about it. That condition no longer holds. This article argues that the response to machine-assisted research should not be a new authorship rule but a different kind of record. It names the present failure Attribution Collapse: the compression of many distinguishable operations into a single binary signal, which discards exactly the information that becomes necessary once operations can be performed by agents of different kinds. Current disclosure requirements do not repair the collapse. A statement that a language model was used for editing is a second binary appended to the first. The article proposes Contribution Provenance as the replacement construct: a structured record of which operations were performed, by which class of agent, under what oversight, with what verification, and with which identified human accountable for each. It develops Provenance Depth as a five-level scale for grading how much of the production chain a record preserves, and specifies PR-9, a nine-field provenance record designed to be completed per operation rather than per person, and to serialise to a machine-readable form deposited alongside the research object. Two design commitments are defended at length. The accountable-party field can never be null and can never name a machine, which follows from the position that accountability is not divisible and does not transfer. And the record is of operations rather than persons, which is what allows it to describe a production chain that mixes agent classes without adjudicating who deserves credit. The proposal is offered as a standard for adoption, adaptation, and contest. The article states what it would cost to adopt, who would have to move first, and the three conditions under which it would fail. It does not claim that authorship should be abolished, that existing taxonomies are wrong, or that the specification presented here is final. Paper 2 of 10 in The Answerability Series. Manuscript ID PROV-AUTH-2026-02. Includes 4 figures, 18 tables, the PR-9 specification, a blank worksheet, a JSON schema, and a completed PR-9 record for the article itself — all released under CC BY 4.0.