The version number is part of the contribution.

A final file hides the decisive work

A final file can look authoritative while concealing the sequence that produced it. Two assets can share a title, filename and polished appearance yet depend on different source material, model releases, human edits or permission states. If an attribution record points only to “the document” or “the model,” it leaves an important question unanswered: which one?

That is more than release administration. A contribution claim must identify the state of the work to which the claim applies. Otherwise a reviewer may credit an input that was later removed, miss an edit that changed the conclusion, or apply today’s policy to yesterday’s output.

Provenance needs identifiable states

The W3C PROV model offers a useful foundation. It describes provenance through entities, activities and agents, including relationships such as use, generation and derivation. In practice, that structure becomes much more informative when a changed asset is treated as a newly identifiable state rather than silently overwriting the prior one.

The Coalition for Content Provenance and Authenticity takes a related approach for digital media. C2PA 2.4 can record ingredients, actions and relationships such as parent-of, component-of and input-to; its implementation guidance recommends recording actions in chronological order. Those records can help establish what changed and how versions relate. They do not, by themselves, decide authorship, ownership, license scope or authorization. Those remain separate determinations that require their own evidence and review.

The version is part of the contribution

For AI-assisted work, a useful contribution record should identify the version of each consequential input: the source material, model or tool, retrieval context, prompt or instruction set, policy and permission state, human revision, and released output. A durable identifier or checksum is stronger than a mutable filename because it lets a reviewer test whether the cited object is the object that was actually used.

This does not mean that every autosave deserves equal weight. The objective is to preserve decision-relevant transitions: the version that entered the workflow, the transformation that changed the work, the reviewer action that authorized release, and the exact artifact that reached an audience. Attribution is then attached to observable contributions across that chain, not retrofitted onto the last file in the folder.

Test final-state review against a version-aware trace

I would evaluate this with synthetic or consented workflows that contain controlled revisions. One review group would receive only the final asset and a conventional attribution statement. Another would receive a version-aware trace linking the same sources, transformations, human decisions and permission changes.

Measure whether reviewers can reconstruct the exact inputs, identify who or what made consequential changes, detect a stale or revoked source, and determine which artifact was approved. Also measure time, error rate, unnecessary blocks and reviewer burden. Add negative controls so that a longer record is not mistaken for a better one.

This is a proposed evaluation, not a report of new results. Its purpose is to test a governance claim: if a version can change the evidence, permission or meaning of a contribution, then version identity belongs inside the attribution record. If the version is missing, part of the contribution record is missing.