---
description: A project page cited a reply from another project. We traced the source, narrowed attribution, and made Unknown visible when evidence ran out.
tags: [REM, Memory, Design Journal]
---

# 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](/evidence/v1.9.0a22) records the inspected
pages and states, and [issue #2195](https://github.com/openonion/connectonion/issues/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.