# ConnectOnion 1.9.0a43 (preview)

## Install this preview

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

```bash
python -m pip install --upgrade 'connectonion==1.9.0a43'
```

## A failed pass needs attention

Home now shows failed investigation counts and short reasons for refused model
access, timeouts or model usage limits. A 44px button opens the maintenance
record and moves keyboard focus to it. If no memory pages were written, Home
says so. Successful memory cards remain visible alongside partial failures.
The [design review](../design-evidence/rem-blocked-init/REVIEW.md) and
[release image set](assets/v1.9.0a43/manifest.yml) use invented contacts.

A public a42 installed-wheel init on real Gmail and Outlook data took 11m 12s
for a five-day sample: eleven mapped People and two projects, but zero written
pages. Four investigations timed out and one received a model authentication
403. A separate minimal Claude request also received 403. This reporting fix
does not restore model access or change prompts. Runner preflight and stopping
queued work after refused access remain in [#2302](https://github.com/openonion/connectonion/issues/2302).

## Init includes the nightly upkeep offer

The overview now describes investigating mapped pages and approving nightly
upkeep from init. The public a42 real trial confirmed its root-specific
launchd schedule was installed; it was stopped after testing. This scheduling
behavior already shipped in a41. `--yes` approves the displayed summary;
`--no-start`, estimate-only, or declining approval leave upkeep off. Init does
not repeat the first foreground sync. `co rem start` can enable upkeep later.

## Verification and limits

The reader browser suite passed 59 tests, followed by an additional passing
partial-success regression. Reader/store tests passed 53 tests and CLI
acceptance passed 109 tests. Private real-data Home screenshots at 390px and
1440px showed the five failures with no horizontal overflow; no private
contacts were published.

The full historical model investigation remains unverified. A complete
metadata-only scan previously took 14m 04s; that is not a writing benchmark.
The failed five-day writing sample is not an estimate of full-run duration.
Durable historical retries remain in [#2298](https://github.com/openonion/connectonion/issues/2298),
and nightly stalls on rewritten sessions remain in
[#2301](https://github.com/openonion/connectonion/issues/2301). REM remains a
preview with these unresolved product requirements.