[AbirAbbas]

The last real PC + Mac round (round 11, published dev build dev-20261027-644a38e6b403) passed every step except one: quitting codeaf right after a background task finished.

What was wrong

• /quit right after a task printed codeaf could not save this chat …, left the hold behind, and the other computer saw this one as offline. The drive's closing step cancelled a publish the relay had already accepted, then retried it with the old head and conflicted with itself (409).
• An ordinary quit gave up after about 12 s while saying up to 30s. The close and detach calls used the ordinary 10 s call deadline, and the save then succeeded after the failure had been printed.
• Each launch sealed an empty turn onto the chat, even one another computer held. A later take turned those turns, plus a late cost record, into a branch chat nobody typed (… 1 turn · <computer> on the other computer's Home).
• A removed computer's quit said it could not save, which can never be resolved there.

What changes

• Close joins the publish in flight, publishes the newest head, records it, and releases.
• Close and detach wait for the shared saving bound. A reply that never came says the save is still finishing, not that it failed.
• A launch never seals an empty turn, and a viewer never seals at all. On a take, accounting-only history (usage, cost, title or memory bookkeeping) is adopted rather than kept as edits. Real edits are still kept.
• A removed computer's quit saves locally and prints nothing.
• The chat manual's quit and continue pages say what happens now.

Proof
Hosted-path tests through the real engine host and native engine, one per behaviour. Each fails on 644a38e6b and passes here. A PC + Mac round on this branch follows before merge.

🤖 Generated with Claude Code