Case 7 — Direct engagement with the finished scholarly object

Research setting

The project had already passed through extensive AI-assisted analysis, drafting, checking and adversarial review.

A later stage required direct inspection of the completed scholarly object.

What inspection found

Residual problems included:

• visual semantics that could imply relationships not warranted by the data;
• figures with encodings inconsistent with the underlying number of clusters;
• repetitive or awkward prose;
• bibliographic and DOI metadata problems;
• institutional or identity contamination from other project material.

Several corrections were not cosmetic because they changed what a reader could reasonably infer from the finished document.

Reason-provenance lesson

Continuous participation in the process is not identical to direct mastery of the finished research object.

The final object needs to be read and inspected as an object in its own right.

Evidence available to public readers

Readers can inspect the categories of inspection issues reported here; the complete private working drafts, review records and corrected exports are not provided as independently auditable evidence.

What this case does not claim

In this project, the final direct inspection was human.

The case does not establish that only humans can perform final verification, nor that human inspection is intrinsically error-free. The broader framework remains actor-neutral.

The defensible lesson is empirical and local: direct human engagement with the finished object found problems that earlier AI-mediated governance had not removed.