After analyzing the global structure of merit networks in
[1], I was interested in taking a closer look of it.
Total nodes (users who have sent/received more than 1 merit): 21103
Total edges (merit transactions): 51890
Here, I used the latest available data, merit.txt, so accounts which do not have any merit transactions are not taken into account. Colors and relative positions are slightly different from the previous analysis as they are set automatically, but the qualitative structure is the same as before.
1. Merit rank distributionTo investigate how the current merit system is working, let us consider an imaginary rank up system solely based on the number of received Merit: Let us suppose all the accounts started with 0 merit on January 24, and define the imaginary rank only by the merit requirement ignoring Activity requirement, which I denote
"Merit rank" in this thread. The analysis of Merit rank distribution helps us to single out the role of merit system in the forum.
For accounts who sent/received more than 1 merit, the current merit rank breakdown is as follows:
| Merit rank | Merit | #Accounts |
| up to "Jr. Member" | 1--9 | 17163 |
| "Member" | 10--99 | 3860 |
| "Full Member" | 100--249 | 164 |
| "Sr. Member" | 250--499 | 31 |
| "Hero" | 500--999 | 6 |
| "Legendary" | 1000-- | 1 |
and their distributions in the merit network space are as follows:
up to "Jr. Member"
"Member"
"Full Member"
"Sr. Member"
"Hero" and "Legendary"
and lists of accounts in "Member" and higher merit ranks are provided in
#2 and
#3 below.
Up to "Full Member" rank, forum members are distributed in various sections in the merit network space and hence exhibit nontrivial structure of networks, above which rank the networks are very simple. I interpret this fact as the threshold of the merit for rank up working more or less for the following logic.
Roughly speaking, the Merit threshold is set as similar to the Activity threshold. Therefore, as people have discussed already, if a user puts high quality posts constantly and can obtain 1 merit per day on average, merit threshold is not obstacle for him/her to rank up. Such a user should be a fiducial qualified poster the current rank up system is implicitly assuming.
Given that the merit system was introduced about 3.5 months ago, the fiducial value of the merit score is around 100. Thus, the nontrivial network structures for "Member" and "Full Member" suggest that the current threshold is not too strong. On the other hand, at this moment "Sr. Member" and above exhibit the simple networks, which implies that the current threshold is not too weak.
Therefore, the merit network analysis tells us that the current threshold is working more or less as desired. It would be interesting to check if "Sr. Member" will indeed acquire more complex structure of merit networks around October, i.e. about 250 days after the introduction of merit system.
2. SatellitesAnother thing one may notice is that this visualization can identify mutual merit transactions very easily as they appear like satellites in the figure. We can find several examples in the "Full Member" network. Note that other merit rank accounts are not shown in this figure.
jooj and
wnj4 are connected by mutual transaction and well isolated from other accounts. Indeed, they have exchanged 50 merits and been already tagged as connected accounts together with
hoop, which is not shown in the "Full Member" network though, as hoop have obtained 1 merit so far. We can check that all these three accounts in the table in
[2].
Another example is a mutual merit transaction between
Coin++ and
CarlOrff, which is shown as a satellite of French community. One can check that their merit histories show that they have exchanged about 60 merits, and the mutual transactions are dominant in their histories [Note that their transactions to other accounts are not shown in the figure as those other accounts are not in "Full Member" merit rank]. They are listed in the table in
[2] as well.
However, of course, the mutual transaction does not immediately mean abuse of merits. While the merit network analysis enables us to identify the mutual transactions intuitively, I would like to emphasize its limitation and that it is still necessary to investigate on a case-by-case basis.
A pair
Paractor &
LastJedi near from yahoo62278 and me, and another pair
First1by &
BOMG respectively have mutual merit transactions. They have been exchanging merit frequently, but it was not immediately clear to me if they are abuse or "healthy" transactions. It may be the case that they have similar opinions and/or preferences and hence they evaluate their posts.
Hence, it would be conservative to conclude that the merit network analysis provides an efficient way to detect the mutual transactions intuitively as satellites, but it is still necessary to take a closer look of them.
References:
[1]
sncc, "Global structure of merit networks",
Bitcointalk, 3650124, May 08 (2018)[2]
sncc, "Merit stat & all transactions more than 40 Merits",
Bitcointalk, 3046077, March 01 (2018) 
