Written 2026-10-21 ~05:52 UTC, before #7 ran.
Method (plan amendment 20:27): NTP (UDP 123) and NTS-KE (TCP 4460) dropped at the VM's tap; the VM restarted cleanly
with its RTC at 2026-10-25T00:20:00Z (= 02:20 CEST); it runs ~80 real minutes, through 03:00 CEST → 02:00 CET, to
~02:40 CET (01:40Z). Positive control for the clock: the box's own HTTP Date header (a different channel than the RTC
setting). demo-hp's clock is not touched.
Expected from the code replay (B6-failures/dst-replay): the box's 02:30 database copy runs ONCE, at the SECOND
(CET) 02:30 = 01:30Z box time, not at 02:30 CEST (00:30Z). Nothing runs twice. The dashboard shows local times in CET
after the change.
Side effect of the method, not a DST result: the box wakes „4 days later" than its last backup; a missed-night
catch-up may start at once.
