TLDR: A reputation algorithm should not be made public if the reputation inputs could be easily exploited.
The team has recently removed the active badge, as V explained: “spammy accounts are aggressively gaming the system to get the badge.”
This decision, I believe, was the right one.
Publicizing the algorithm will heavily influence users action and therefore their behaviour. A public reputation algorithm is like a written constitution, what’s outlined sets the norms and rules within the community. Users will engage in behaviour that positively influences their score and avoid behaviour that harms it.
Publicly displaying the score as a badge naturally motivates most users to pursue it, given the high value placed on reputation, which is often considered one's most valuable asset. This pursuit is fuelled by the need for social validation and the influence of memetic desires.
The logic is the following: if the score is publicly displayed as a badge, then users will try to get it; and if the algorithm is public, then users will learn how it functions to acquire the badge.
That phenomenon was evident with Warpcast. Making the algorithm and score public, embedding the badge in applications, and linking it to social and economic benefits, along with speculation about future opportunities, led to the exploitation of the active badge, similar to airdrop farming.
The algorithm was exploited because the inputs weren’t robust
A reputation system is exploitable when its inputs are non-robust, meaning cost-free, easily fabricated, and can be generated at will.
As an example of robust inputs, consider a chess platform’s reputation algorithm that determines a player’s rank based on the number and outcomes of matches. These inputs are considered robust because they can only be produced by participating in a game, which requires an investment of time and gameplay. Additionally, these inputs do not change player behaviour, as rank can only be influenced by playing chess itself.
Below is the active badge algorithm, comprised entirely of non-robust inputs.
Among these non-robust inputs, 'Complete profile' and 'A connected Ethereum address' still provide some value by enhancing user experience and aligning with product strategy. However, other inputs like '400 followers,' 'inbound reactions,' and 'inbound replies' can degrade content quality and user experience. Free actions such as liking, recasting, replying, and following became tools for manipulation, leading to spam and detrimental changes in community behaviour.
Many users began creating and engaging with content solely to obtain the active badge, content that likely wouldn’t exist if the reputation system wasn’t public.
To exploit the '400 followers' criterion, 'follow for follow' (F4F) emerged as an extreme application of the rule of reciprocity. To facilitate this, eight F4F channels were created, enabling users to advertise their participation in the F4F business and gain followers.
The most successful of these channels is the one with the ID 'f4f’. It has 5.5K followers and is ranked #18 among the top 25 channels by cast volume, even outperforming major channels like Warpcast, as illustrated by the cast from @jtgi.
To exploit the “reaction and replies inbounds” inputs, users again employed the rule of reciprocity, leading to the emergence of like-for-like(L4L), recast-for-recast (R4R), similar to F4F.
Below are some examples of casts to illustrate this:
Using Neynar, I found that 36.08% of users active in the'f4f' channel earned the active badge, exceeding other relevant channels such as 'founders' (26.18%) and 'warpcast' (31.21%).
I'll list for fun the top 25 channels and their active user percentages, noting that this data might not fully illustrate badge acquisition due to cross-channel activity. Future analyses will focus on correlating active badge holders in 'f4f' with their power badge status.
I predict that many of F4F members that had the active badge won’t have the power badge.
