B5 — the first night, no hands (2026-10-20 21:00 → 2026-10-21 05:36 UTC)
Expectations written before: 00-expected.txt. Evidence: the box's own log ring copied off every 10 min (logs/),
health and memory every minute (../B4-use/), the hub's log (a different channel), two data snapshots through the apps'
own front doors (01-…, 03-…).
| # | expected | observed | verdict |
|---|---|---|---|
| 1 | database copy ~00:30Z, all apps | 00:30:00–00:33:27Z: 4 databases (55 MB) + 8 volume dumps, recovery units for all 8 apps; event db_dump_completed |
PASS |
| 2 | second-drive copy ~01:30Z | 01:30:00–01:30:58Z: 8 apps + the share „Filmek" (1.8 GB) to the second place; 8 + 1 crossdrive_completed events; the system share „beolvasas" skipped with a WARN (W28) |
PASS |
| 3 | off-site ~02:15Z, one snapshot per class | 02:15–02:18:11Z: dump leg 2 m 22 s, 8 apps + 1 share folder + definitions, 44 snapshots, 3 m 6 s; retention skipped (no clean-up window). Second channel: the hub's own listing at 05:10Z sees 44 files, newest 02:18:00Z | PASS |
| 4 | whole-box backup 02:30–06:30Z | NOT RUN, by design: the agent's rule is „due 24 h after the last one" (localapi/server.go:1187); the last was the install's own at 12:09Z, so at 02:30Z it was not due. The household's first day is in a whole-box backup only on the SECOND night (~38 h after install). Then at 02:45:59Z demo-hp itself rebooted for its own kernel update (02-host-reboot.txt) and shut this VM down cleanly; the VM has no auto-start |
NOT A PRODUCT DEFECT on Tester4; see the two notes below |
| 5 | no freeze | health 200 every minute from 21:00 until the host's shutdown at 02:46Z; memory steady (used 3638 MB, load < 1) |
PASS |
| 6 | no mail on a good night | PENDING — mailbox read with the B6/B7 reads | — |
Data: the morning snapshot (05:36Z, after the restart) is IDENTICAL to the evening one: Paperless 47/47 and Immich 277/277 originals byte-identical to the manifest, Nextcloud 3030/3034 (the 4 known gaps of the day), Mealie 2, BookStack 5 pages + 5 images, Radicale 2 events, Vaultwarden 50 items + 2 attachments, the share's 4 files.
Restart: 05:24:02Z start → dashboard 200 at 05:24:41Z → all 8 apps running at 05:26:02Z (Nextcloud answering ~05:35). A first snapshot taken at 05:25 (one minute after start) read 404s and „directory not found" — the apps were still starting; it is overwritten by the 05:36 one and is not a finding.
Notes:
- The whole-box tier on day one. A box installed in the afternoon is not whole-box-backed-up on its first night (24 h rule). R-899 (closed) fixed this for a daytime PRESS; the install's own backup still moves the cadence. Filed as a row (P3, design question) — see the register.
- The harness host reboots itself. demo-hp is a Felhom APPLIANCE box on ring 0; its nightly kernel step reboots the
host and with it every VM on it. Tester 1 (VM 341, auto-start on) came back; the rehearsal VM (auto-start off) did not.
NOT-A-FINDING: a BYO host skips the host step by design (agent
internal/osupdate/leg.go:674, „a BYO host belongs to its owner",11§1); on an appliance every guest is Felhom's. A harness lesson: a rehearsal VM on demo-hp needs auto-start on.