Every day, Alexa estimates the average daily visitors and pageviews to every site over the past 3 months. The site with the highest combination of visitors and pageviews over the past 3 months is ranked #1. The site with the least is ranked somewhere around 30 million. If no one in our measurement panel visited a site over the past 3 months there is no rank at all for that site.
As per the above definition, your rank can decrease:
a) Because a given site receives less visits and/or page view.
b) Because other sites receive more visits and/or page views.
The rank is really expressed in terms of “global internet engagement”. That is, it is not contextualized to related or comparable internet sites, but rather to all internet sites.
Now one would no doubt prefer to see the graphic the other way round (an upward trend, and not a downward one), but the issue with the chart is that we cannot discern how much is due to “a”, and how much to “b”. If a bunch of other non-related sites have upped their visits and/or page views, that can cause quite a shift in the ranking.
Having said all that, I rather much like some of the information that @Coin-1 sums up on this thread: [Chart] Bitcointalk statistics on impression counts for ads. Specifically, the Unique IP and Unique logged-in charts, which place us back in 2017 metrics, but that represent endogenous data which is not influenced in the reading by the environment (such as happens on the Alexa chart).
I follow the post count (see https://public.tableau.com/shared/5G8SFTFN7?:display_count=yes&:origin=viz_share_link), but prefer the above information gathered by @Coin-1 if I had to choose the best first level objective metric.