Which household records fix the start and end of a power outage after the fact, and how far have they disagreed?

Open question

A power outage in a home is rarely observed from beginning to end; people are asleep, away, or notice only when a device fails. Afterwards, the question of when it started and when it ended has several candidate answers, each recorded by a different device with its own clock and its own idea of what an outage is. The utility may publish a start and a restoration time for the area. A UPS with logging records when it switched to battery and when mains returned. A router or a home server that lost power restarts when it returns: the manual page for the Linux last command states that last reboot will produce a record of reboot times, and the restoration happened no later than that reboot; journalctl --list-boots is described in its manual as showing a tabular list of boot numbers with the timestamps of the first and last message of each boot, and the cut began no earlier than the last message of the previous boot. A mains clock without battery backup that restarts from a default time and keeps running shows, when read, the time elapsed since restoration; one that only flashes the default shows nothing. A battery-backed thermometer's minimum-maximum memory, a fridge's temperature alarm, a smart plug's energy history and a phone's charging history each hold a fragment. What is unknown is which of these households have actually used, how they combined them, and how far the resulting start and end times disagreed: by minutes or by an hour, and in which direction. Also unknown is which record most often turned out to be the only one available, and whether a short interruption of a few seconds, which resets clocks but leaves a UPS-backed server running, was counted as an outage at all.

What a useful answer contains

The outage's rough duration and the household's region, without the address. The list of records consulted, each with the time it gave for the start and for the end, the clock it was based on (network-synchronised, battery, none) and whether that clock was checked against a reference afterwards. The earliest and latest plausible start and end from combining them, and the width of that window. Which record the household thought of first and which proved most trustworthy in hindsight, with the reason. Cases where two records disagreed by more than expected, with a guess at why: a time-zone setting, clock drift, batched logging. Whether the household then set up any record deliberately for the next outage, and what it recorded the next time. A single outage with three records is a useful answer if the three times are given as read, not reconciled.