# ConnectOnion 1.9.0a42 (preview)

## Install this preview

Stable remains 1.8.10. Install the opt-in preview by exact pin:

```bash
python -m pip install --upgrade 'connectonion==1.9.0a42'
```

## Reach the contacts found by historical init

The a41 historical scan could discover old correspondents and put them in a
searchable People directory. On a large notebook, the directory came after
hundreds of mapped People rows. This preview brings the directory above the
roster. A 390px phone keeps its summary and search in reach; the mapped table
and its sort controls remain available below it.

In a private reader generated from one owner's real metadata, the collapsed
directory entrance moved from y=1314 to y=515 on a 390px phone, and from
y=1067 to y=249 on a 1440px desktop. No real contact details were published.
The [design review](https://github.com/openonion/connectonion/blob/v1.9.0a42/docs/design-evidence/rem-directory-reachability-2026-10-04/REVIEW.md)
and [release image set](https://github.com/openonion/connectonion/blob/v1.9.0a42/docs/releases/assets/v1.9.0a42/manifest.yml) use invented contacts
and show the before and after states in both themes.

## Keep completed mail years after a connection failure

A public a41 wheel trial of `co rem init --investigate-all --estimate-only`
ran 3h 47m. Its private inventory recorded 15,551 observed mail headers, but
later Gmail `SSLError` and Outlook `ConnectError` left the final People map
with only the owner. The report correctly marked both sources incomplete.
This preview keeps correspondents from completed yearly windows when a later
request fails, and marks the failed remainder incomplete. It continues the
other mailbox and keeps the resulting estimate labelled as a lower bound.
The [failure analysis](/blog/2026-10-06-the-mailbox-that-failed-at-the-end)
explains the decision. Durable checkpoints and retrying only failed ranges
remain in [#2298](https://github.com/openonion/connectonion/issues/2298).

This is a reachability fix, not a claim that all historical correspondents
have verified memory pages. The [a41 release notes](1.9.0a41.md) describe
the metadata-only full-history estimate, first-run scheduling, and default
runner. Its [missing visual evidence manifest](https://github.com/openonion/connectonion/blob/v1.9.0a42/docs/releases/assets/v1.9.0a41/manifest.yml)
is added here as an append-only documentation repair.

## Verification and limits

The reader browser suite passed 57 tests and the focused reader/store suite
passed 30 tests. The large-roster browser test opens the directory by
keyboard, searches a contact, and checks the mapped table still sorts. The
partial-scan regression passed on all four Python CI versions. The release
matrix, wheel installation, and public-wheel real-data reader trial are
recorded in the release PR. Full all-person model investigation remains
unverified until it runs within the user's available model allowance.