Runbooks: the scripts/ move (PR #749) and the retired tools (PR #748) - ops_lock and shutdown_mode run as `python -m caper.<module>` from /srv/caper. Operator-lock gains a section on checkouts from before the move, and on the one deploy that crosses it (lock taken the old way, released the new way). - start-server.sh is at scripts/deploy/; the root stop-server.sh is gone, and the dev page names the two commands it ran. - Script paths under /srv/scripts/{diagnostics,maintenance}/; the backup cron line uses `-m caper.backup_sqlite`. - check_volume_reclamation.py, rebuild_version_chains.py and repair_head_chain_fields.py are retired; the reclamation answer (no) and how to re-check it are kept on How-backups-work, and the I9 entry is dropped. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01C3qw3gotMvPgGSJp5Tb2xe
The reference builds now have a verified copy off AWS Taken 2 Sep 2026: all ten tarballs downloaded and checked against the md5sums published beside them in the bucket. That was the last single copy of anything this site depends on, so it comes off the uncovered list on both pages.
Add a page describing how backups actually work The off-site page covers the copies a person takes by hand. Nothing described the system: which mechanism protects which store, the policy decisions and the reasoning behind them, how to restore and what the restore looks like while it is running, and how to tell any of it is working. Also corrects the metadata dump size on the off-site page: it is ~39 MiB now, not ~51, because fs.files fell from 1,065,009 rows to 808,532 when the GridFS residue was swept.
Correct the API section against the real endpoints, and index the wiki
Off-site backups: the procedure for the copies that leave AWS