The project page that borrowed a reply

The first project page looked persuasive. It had a purpose, a current-sounding finding, and a numbered source. Opening that source was the important part: the reply had come from a coding session that had been working on a different project. A short follow-up did not name either project, but our collector had assigned it to the folder we were investigating. A citation made the mistake look more certain.

The next trial exposed a second version of the problem. Once we withheld those ambiguous follow-ups, the page reached for an old repository note and described it as what was happening now. That note was real, and its source dialog opened, but its date and meaning did not support a current status. Source access alone had not answered whose work the note described, or when it happened.

We considered keeping every folder-matched message and asking the writer to disambiguate it, but the cited wrong-project reply showed that a confident writer could still turn the match into a false finding. We chose the stricter boundary and accept that it can leave a useful project page with no current finding.

The page now keeps that uncertainty visible. An unnamed follow-up after a session has used another project folder is not assigned to this project. If no session inputs remain, the current insight and open threads say Unknown, while the page can still explain the project's purpose from retained repository sources. The reader shows the input count beside that limit, and the numbered sources open the retained originals rather than a generated inventory.

In the final isolated seven-day trial, 55 ambiguous follow-ups were withheld; zero of 893 archived workspace inputs were assigned to the sampled project. All three cited repository sources on that page opened. That is a narrower, less exciting page than the first draft. It is also a page whose reader can tell what we know, what we do not, and where each claim came from. One sampled project and one sampled Skill page do not establish quality for every page. The independent AI founder/UI review records the inspected pages and states, and issue #2195 tracks the remaining attribution and source-accuracy work.

We should revisit the boundary when session attribution can identify the project of an unnamed follow-up from retained context and then pass a fresh source-level review. Until then, an explicit Unknown is the better answer.