ConnectOnionConnectOnion 1.9.1b1 (preview)
Install this preview
Stable remains 1.9.0. This is the first 1.9.1 beta. Install it by exact pin:
python -m pip install --upgrade 'connectonion==1.9.1b1'
What 1.9.1a2 measured
A full 180-day first run on the owner's real Gmail and Outlook, public 1.9.1a2 wheel, Codex with gpt-6-luna:
| 1.9.0 (90 days) | 1.9.1a2 (180 days) | |
|---|---|---|
| Wall time | 83 min | 169 min |
| Input / output tokens | 134M / 2.3M | 420M (90% cached) / 5.5M |
| API-equivalent cost | US$3.75 | US$10.66 |
| ChatGPT Pro weekly window used | 1 point | 4 points |
| People written | 68 of 70 | 84 of 84 |
| Organisations | 18 of 18 | 27 of 27 |
| Projects | 6 of 17 | 14 of 20 |
| Refused pages | 12 | 7 |
Every remaining project refusal was one citation. This beta fixes those, and shortens the run.
One citation drops its lines, not the page
After its repair turns, a project page whose only remaining problem is a citation that cannot be traced drops the lines resting on it, as person pages already did. Sources are counted in prose on both sides, so a source cited only inside a diagram is removed instead of refusing the page. A skill page edited in place keeps one Run evidence section.
Decisions and principles from the first run
When init has written people, project or organisation pages, it ends by drawing
decisions, then principles, from them (co rem abstract). A failure is one line
with the retry command; the pages stay.
Documents on the page
A page lists the attachments it cites under Documents, and each opens from the page; a cited attachment's source popup opens the file too (#2315).
Faster
Init investigates 16 pages at a time instead of 10. Mail is fetched through one pool per mailbox, 20 for Gmail and 8 for Outlook, instead of one shared 10 that left half the pages in flight waiting on mail.
Better final rounds
The synthesis round keeps every thread the earlier rounds recorded and asks for a mail search before writing that a thread's outcome is not established.
Verification and limits
The co rem unit suite passed, with a regression for each change that failed on 1.9.1a2 first. A full first run on this beta is the next measurement and decides whether it becomes 1.9.1.