R-435 — the weekly clean-up windows of 2026-10-26, read (2026-10-26 evening; the brief dated it 2026-10-27)
The box logs are gone, said plainly. The windows ran 2026-10-26 03:15–03:18 UTC. Every test box's controller
restarted twice that day (the 09:07 and 13:27 UTC floors), and the hub pod at 13:08 (hub 0.172.0), so neither the box's
window lines nor the hub's DEBUG „explained by clean-up windows" line exist any more. The check is reproduced from two
other channels: the hub's own records (the database copied with its -wal from the running pod) and the mailbox.
| Box | Window | Copies before → after | Box result | Reports the hub stored | Drop alarm |
|---|---|---|---|---|---|
| Tester 1 | 15: 03:15:52 → closed 03:16:06 | 78 → 57 (21 removed, cap 39) | pruned | 74 (01:31:56) → 57 (03:16:56) | none |
| demo-felhom | 14: 03:15:48 → closed 03:16:00 | 63 → 48 (15 removed, cap 31) | pruned | 58 (01:31:53) → 48 (03:16:53) | none |
| demo-hp | 16: 03:17:52 | 359 → 359 | guard-refused (R-1294) | 345 → 359 (no fall) | — |
- The positive control: the hub DID receive the lower count, 50 s after each window closed — so its checker had a fall to judge, and each fall (17, 10) is within what its window removed (21, 15). Without the window credit the D7 rule mails an error for even one missing copy.
- No drop alarm: no
offsite_snapshots_*/offsite_window_dropevent or mail row for either box (hub-windows-and-alarms.txt), and no such mail in the operator mailbox for 2026-10-26; the same mailbox search does find that night's demo-hp refusal mail (03:17:56Z), so the search sees off-site mails. - Files:
hub-windows-and-alarms.txt,report-count-replay.txt. The database copy was deleted after the read.