[tor-relays] Previous Guard not getting Guard flag back

[starlight.2017q4 at binnacle.cx]

What might not be directly obvious is the calculation bases
on 12-hour intervals where on each interval the previous
uptime is down-weighted to 95% of the next-most-recent.  OnionOO
uptime intervals are four hours, so each set of three OO
intervals are averaged as a single 12-hour interval and then
weighted.  This only approximates logic of rephist.c which
keeps track of uptime with compact records indicating the
start and end of relay-is-up span.


At 00:34 3/21/2018 -0500, starlight.2017q4 at binnacle.cx wrote:
>>While I understand that my relay lost the guard flag because of a weekend
>>of downtime, I would expect that it would get it back after a while of
>>stable again? Anyone able to shed some light on when it will get the flag
>>back?
>>https://metrics.torproject.org/rs.html#details/924B24AFA7F075D059E8EEB284CC400B33D3D036
>
>
>Guard flag calculation is somewhat involved (due to several
>get-out-jail promotions/bypasses), but the essential
>part is weighted uptime in excess of the median uptime of guard
>candidates, or 98% whichever is lower.  Presently the authorities
>have it somewhere around 96% (they do not publish the value).
>Your relay is at about 91% and will be back as a Guard in less
>than five days when it will hit 97%.  The attached XLS shows
>it roughly, left side today and right side as-of 3/26.  Uptime
>data came from the "1_month" section of
>
>   https://onionoo.torproject.org/uptime?search=NSDFreedom
>
>Built this sheet for myself recently and simply stuffed in data
>for your relay.  Enjoy.
>
>P.S. If anyone spots flaws in the approach, please comment.