# ConnectOnion 1.9.0a15 (preview)

This preview fixes one first-run trust break in REM. The map announced a
project count, but the later session-material pass could silently create more
project pages before the first run finished. On an isolated private run of
1.9.0a14, the map listed 4 projects while the final notebook contained 13:
9 were created by that later pass. No private names or source text are in this
release note.

`co rem init` now prepares session evidence only for projects already in its
map. Unmatched session folders remain private candidates in
`.state/projects/index.json`; they do not change the first-run project count or
join its model-writing queue. `co rem projects` and the later daily discovery
path keep their existing behavior. This is the first slice of [#2079](https://github.com/openonion/connectonion/issues/2079),
not a claim that all project detection, privacy or page-quality issues are fixed.

## Evidence

An invented notebook with one mapped project and one later-discovered folder
shows the old two-row Projects directory and the new one-row directory at
[desktop and phone widths](../design-evidence/rem-init-project-map-a15/REVIEW.md).
The changed path has a regression test that checks the first-run selection,
unchanged page count and retained candidate. Relevant first-run, project and
CLI tests pass. The screenshots use invented data; no private reader screenshot
is published.

An installed 1.9.0a15 wheel was also run against an isolated private five-day
notebook. It exited successfully: the map listed 3 projects, the final notebook
still had 3 project files, one selected project page was accepted, and material
extraction created 0 additional pages. It retained 122 unmapped session-folder
candidates, including older folders. These are aggregate counts, not a review
of whether every candidate represents a real project. The owner's page also
completed and the init result reported no source errors.

## Remaining work

The map can still miss a legitimate workspace-attributed project. The private
candidate list has no reader review flow or useful ranking, and the daily discovery policy is
separate. REM is still a preview: a bounded overnight pass, verified morning
changes, owner correction, useful retrieval and durable recall all need their
full product acceptance checks before a stable 1.9.0 release.