Known defects in this version

None known that touches a guarantee this guide makes. Since the previous revision (2026.10.24), these were found and fixed:

In 2026.11.0:

• Daily snapshots had stopped on the hosted service. From 11 August to 30 October 2026, the daily snapshot of the disk was not taken, and nothing alerted: a policy had gone into error after a disk was replaced. The nightly database backups ran throughout. It was fixed, a snapshot was proved, and a missing snapshot now raises an alarm within a day.
• A busy database answered with a raw internal message. Under heavy load, an agent got the database driver's own error, which said neither that it was temporary nor when to retry. It now answers 503 «Database busy» with a Retry-After.
• Ordering at the end of a list wore out. Each issue created at the end of a long list made its position one character longer. Around 16,000 issues a project would have stopped accepting new ones. Positions now grow slowly, and long ones are rewritten.

In 2026.11.1:

• Three of the platform's own tables were writable by the application's database role, which only needed to read them or nothing at all. They were closed, and a test now reads the database's grants for every such table.
• Bulk acceptance with a read-only key answered «unexpected» for each item instead of refusing the request as read-only. Nothing was written. It now refuses the whole request.

A defect found later is listed here in the next revision.