Part 3 — the live proofs on scratch 9202 (2026-11-03), controller test image of c4b6844
The image 0.350.0-rc.1-sec1102fault was local only: docker save/load, never pushed, removed after. Driver: proof1102.py. Events went to a sink in the guest, not a hub.
| Row | What was staged | What happened | File |
|---|---|---|---|
| R-1347 | an update with its undo copy cut off (finished-marker removed while it verifies; a held volume does not stop the undo, try1-…) |
hold update_failed; the household's restore: „the update hold … is CLEARED", „its update-held flag is forgotten"; six kill -9: crash_loop … STOPPING, hold unhealthy_stop: the watcher watches again |
stdout-r1347.txt, stdout-r1347b.txt |
| R-1348 | a restore cut at „the restore copy is recorded": docker kill, the volume held by a created-only container, the controller started |
„could NOT be put back", hold restore_mixed / interrupted_put_back, event backup_run_failures with the R-1340 sentence in the sink. It showed a wrong run name (offsite-reconstitute): fixed, 3764445 |
stdout-r1348.txt, r1348-after.txt |
| R-1339 | a guest reboot, before and after | 9 false lines → 0 | stdout-boot-before.txt, stdout-boot-after.txt |
| R-1351 | reboots: plain; the sync recreating; also the file manager held down so both paths run in one second | 0 Conflict and 0 first-boot WARN on all six, 3 per image. The race did not fire on the old image either, so these boots do not tell the images apart. The proof is the test with the lock removed | stdout-boot-*.txt |
| Traefik v3.7.17 | three boots | started, dashboard 200, file manager healthy | stdout-boot-after.txt |
Left on 9202: one hold in its settings, radicale (restore_mixed), from R-1348's proof. A brief's fence forbids CC to clear a restore hold; the app itself is removed. To clear it: pct exec 9202 -- docker exec felhom-controller felhom-controller --clear-restore-hold radicale. Until then a walk of Radicale on 9202 is refused its start.