Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0159rPz1ZhFKsS53msqPYxtS
Part 0 — the rulings recorded, last night read (read-only), 2026-09-25 midday
Recorded: 09 §3 decisions 35 and 36; D3 in STATUS.md's decision section; CONTEXT.md.
Last night (24 → 25 September) on both demo boxes
The controllers were restarted by the 0.272.0 delivery at ~11:30 CEST, so their own logs and debug rings no
longer reach the night (5 000-line ring). Read instead from the hub's stored reports (P0-hub-update-leg.txt,
a copy of the hub DB with its -wal, deleted after) and the agents' journals (P0-wholebox-backup.txt):
| box | update leg | steps | whole-box backup |
|---|---|---|---|
| demo-felhom | 04:15:46–04:16:06 CEST, after the off-site copy | opengist 1.13 → 1.15 done in 20.1 s | ran 07:29:19–07:30:31 CEST to felhom-backup — after the leg had ended at 04:16; it had nothing to wait for |
| demo-hp | 04:18:41, 0.07 s | none (every app current) — reported as "steps": null (R-687's cosmetic item, seen live; fixed in v0.273.0) |
none — its last was the evening's 21:59 hand-triggered one (Part A4 of the previous night), not due |
R-687 item 4 (the full-system gate waiting for the leg) did not occur on either box again: on demo-felhom the backup came three hours after the leg; demo-hp had none due.
Found while reading, filed as R-689: demo-hp's scheduled restore test picks the golden template
local:backup/felhom-golden-0.236.0.tar.zst as "the newest settled archive" and fails with
HTTP 403 … missing privilege VM.Backup every 6 h (2026-09-24 10:36, 2026-09-25 04:57 and 10:57)
— P0-demo-hp-restoretest-golden.txt. Agent-side; the agent is untouched this session.
Which boxes run docmost (the brief said: demo-hp only)
TRUE. demo-felhom runs cloudflared, controller, filebrowser, opengist, traefik — no docmost. demo-hp runs
docmost 0.96.0 on postgres:16-alpine (P0-docmost-demo-hp.txt): pinned and installed refs equal, digests
recorded 2026-09-22; deployed 2026-08-31.
Its last copies, per tier: own unit (Tier 1) db-dumps/ 2026-09-25 02:15 UTC, manifest 09:31 UTC;
second drive (Tier 2) hdd_1/backups/secondary/docmost/recovery-unit mirrored 2026-09-25 01:30 UTC (the layout
marker) — directory mtime 2026-09-24 20:59; off-site — the night's off-site run covered 9 apps / 90 snapshots
at 04:15–04:18 CEST (the previous night's record); per-app snapshot times were not read.