Fictional teaching case: no real library, product or deployed 1Cademy feature is described.

An AI drafts a brief (version 1). The only evidence supplied is a notice: "During July, the Central Reading Room stays open until 8 p.m. on Wednesdays." The brief concludes: "All branches stay open until 8 p.m. every evening." An editor is asked to approve the brief. Compare three arrangements:

1. Acknowledgment after release. The brief went out at noon. Clicking Approve at 12:05 records acknowledgment; it did not control the noon release.
2. Release gate without scrutiny. The brief cannot be sent until approval, and refusal or no response keeps it unpublished, but the editor sees only the summary, not the notice. The editor has real release authority but no established way to check the inference.
3. Inspectable release gate. The same gate shows the exact notice and version, allows revision or a hold, and honors refusal. The editor can see that the notice covers one room, one month and one weekday, not all branches and evenings. That is missing support, not proof that other branches close earlier, so the editor can ask for a narrower claim or hold release while support is checked.

Now suppose a strengthened version 2 replaces version 1 after approval. In this case approval covers only the version shown, so approving version 1 is not evidence that anyone examined version 2; other policies may treat minor edits differently, which is why a policy must say what approval covers. Approval of this version is not evidence of authority over the AI's next draft or another group's later reuse.

Release authority, informed scrutiny and independent review are different properties: a conflicted editor may hold a real veto, and an expert may spot the defect without any authority to stop release. Even the third arrangement does not guarantee a correct or independent decision. The arrangements differ in several ways at once, so they illustrate distinctions rather than measure effects.

Questions to consider (ungraded): What can the editor still change, and when? Which parts of the conclusion go beyond the notice? Does unsupported mean false? What would show that refusal really blocks release? What did version 2 change?