desc_bw_obs_last_avg is pretty spikey (recently)

I was looking at our average observed bandwidth graph for the last 7 days for gabelmoo, which looks like:

Now, what's going on in that spike on 2024-09-02? The peak happens with the bw file published at 2024-09-02 08:39:17 and the sharp drop is from the next one published at 2024-09-02 09:39:09. Is our average observed bandwidth really dropping from 19317815 to 13297058 just within 1h?

To check that I've used the respective 2024-09-02-08-39-27-bandwidth-622E9F5085B3FE4B37D5EA60BCB8ADE135B08F690AC2D13EE3465FDF7270528F.xz 2024-09-02-09-39-15-bandwidth-27D5FD229161533AEF0D913BEF40D72FD34E54E91275EE4A44954E3AA7480D27.xz directly with a bwfile_entries.py (you need to adjust the file path to your setup).

It prints out the time of the published bandwidth file, the amount of eligible relays (that is, those who have desc_bw_obs_last in the bw line), the sum of all desc_bw_obs_last values and then the average:

2024-09-02 08:39:17
8852
110137684733
12442124.348508812

2024-09-02 09:39:09
8852
110224437117
12451924.663014008

So, there is no big drop in average observed bandwidth visible anymore. Rather, a slight increase (from 12442124 bytes to 12451924 bytes), which does not look unreasonable on first glance. I wonder why our values grabbed from the DB look so much different. We do

SELECT DISTINCT
    published,
    avg(desc_bw_obs_last) as desc_bw_obs_last_avg
FROM bw_line
WHERE published BETWEEN '${__from:date:iso}' AND '${__to:date:iso}'
AND scanner_country IN( ${scanner_country})
GROUP BY published, scanner_country

but that looks reasonable as well... @juga