METADREAMER here, you might know me as the designer of the legendary OG $YFI Hoodie. As a certified SourceCred shill, thought it would be interesting to see how well it would work on the YFI forums so I ran it and got these scores:
TLDR on SourceCred: its a tool to measure the value of contributions in open communities using a modified version of the PageRank to give cred to contributions when they are referenced or depended on by other contributions with cred. More concretely for these forums, users earn cred when they get likes and engagement on their posts. You can read more about how SourceCred works here.
Was curious to get everyone’s thoughts on the rankings and cred scores it assigned and if they are accurate in any way and if there’s value in this data at all. If people generally think the rankings are representative of people who bring value to the community, it might be a cool way to distribute small amounts of funding and reward contributors in a decentralized way.
In a traditional system of just “counting metrics” to distribute rewards, yes it is ripe for abuse and easy to game. However with SourceCred and the nature of PageRank, it mitigates the potential to game the system a lot. Gaming can never be prevented entirely IMO, the important part is being able to detect and moderate bad actors, which is possible because all payouts would be longer term and cred updates retroactively.
The biggest way this could get gamed is people creating Sybil accounts to like their own posts. This can be mitigated by using the Trust levels feature in discourse so that new users enter in an “untrusted” state where they can receive cred but not mint cred for others by liking posts. Once a user has engaged in a meaningful and valuable way, they could be assigned a low trust level to mint small amounts of cred, and then users that are high trust in the community can mint larger amounts of cred for their interactions. This makes it fairly Sybil resistant.
The Maker forums has been using SourceCred for the last many months, and surprisingly after enabling it, the behavior of participants in the forums didn’t really change much. SourceCred itself has distributed over $700k to contributors over the last year or so and also hasn’t had problems with being gamed. The best way to “game” the system Is to write high quality, thoughtful posts and replies, which is a good outcome IMO. Spamming lots of threads won’t earn you cred because no one would be liking the content and you aren’t bringing real value to the conversation.
For that reason, I think it would be worthwhile to at least experiment with it on a small scale / trial run to see how it goes here, even if it’s $1000/mo or something.
I work as an applied game theorist and we use modified pageranks to evaluate the quality of scientific contributions (that is, has consequences for example the assignment of governmental funds). And it works really well. I know a little bit about those measures because I have worked on deriving one in some concrete contexts as well. I do not think that they are very gameable: for example, if two people decide liking each others posts, it does not matter at all for their measure if they do not receive likes from those that receive likes from many different parts of the network. Mathematically, we are speaking about the properties of the dominant eigenvector of a sub-stochastic matrix, which are well-understood.
Another question is if you really want that people spend effort in writing posts just because they expect some reward. I believe that not everything in life should be defined through an explicit contract. In fact, an explicit contract (receiving monetary compensation if you do this and that) is known to crowd out intrinsic motivations. That is, humans perform actually better if not everyhting is specified upfront.
Whatever, we are looking definitively at the relevant class of measures…
I think the likes from sybil account do not matter at all “unless” these accounts make relevant contributions. So, the person would have to divide “valuable” posts among many accounts. Since the measure itself should be linear in this respect, no benefit would occur. I can think of some weirder behavior where one person takes both sides of an argument with valuable posts on two different accounts in order to increase the total number of quality posts. But I find it farfetched and one could argue that there is added value in each of these posts.
That’s correct, there’s no such thing as “negative cred”, however ones relative share of the total cred can go down if they get diluted by new cred being minting by others if the person is no longer making contributions.
The other way cred scores could go down is with weight changes (e.g. you decide to lower the weight of Trust Level 1 posters), but that would effect everyones scores unilaterally rather than one specific persons cred going down because of their interactions / posts
As an analogy, think of these top contributors as the top contributors to wikipedia. The motivation factor for these top contributors is not monetary. Im sure most of them can make more money elsewhere for their time. I think some form of ‘recognition’ or ‘prestige’ might be a better motivating factor. Infact if we make the rewards monetary we might start attracting the wrong groups of people.
May be bluekirby can create special NFT awards for the top contributors and they get special swag at yearn(future) parties like special vip passes etc ;)…you know things money cant buy lol
So there is nothing in this system to discourage sarcastic posts such as the one I made? Overall, the cred system looks interesting (I checked my cred level for instance), but we are all posting and trying to contribute without any expectation of payment.
I think i still have 2 main issues. But that’s just my opinion
First i don’t really like this. Because it can very quickly become more about popularity than quality. Let’s see what is happening everywhere. More likes , doesn’t mean quality. But sometime it’s just because it’s cool and people follow people. Even if i understand that , in a certain way, you can limit that, it will still exist. Let’s imagine someone "popular’ here, not necessarily because of the quality of his post but just because he is active and people know him. So he will like others posts, create topics… SO then people will also like his content because they know him and you prefer to like people you know.
Then someone coming from nowhere will post something amazing and super useful. But he won’t probably receive the same amount of “cred” .
What it also means it that people, like me, who read and visit this chat everyday and do not often participate , won’t appear in your ranking. But are part of the community
I also have the feeling that some people are trying to get everything. And we are loosing the spirit of yEarn. We created the "The Yearn Team ". Some have a very nice salary. Some really deserved it , and they should probably have more, but i have to say that for some , i don’t really understand. But anyway, i guess people voted and agreed , so that’s ok.
But i guess that those people are probably the top one on your current ranking. So should they earn everywhere ? do you think we can “exclude” some people ?
Can we find an other way to reward people involved ? Like contests or things like that ?
I personally write articles for my blog about crypto, speak about yEarn, created some Meme … But just for fun and help the community, I’m not asking for rewards (but would be super happy to have one). Could also translate articles in other languages … What I mean here, it that if the goal of this ranking is to reward involved people, maybe we can also add other entries. People can be involved without participating here (and for many reasons…).
Last question: Where will you find the rewards ? from treasury ?
We could also give t-shirt or goodies instead of money !
The only information taken into account in the calculation is the likes one receives (and from whom). The calculation cannot check who owns the accounts or which information is given in these posts.
Think of three individuals (A,B,C). If A likes 2 posts of B and C likes 1 one post of B, the “I get liked”-matrix is
A B C
where rows indicate the likes you receive. The 1/3 in the second row means that overall, there is a 1/3 probability that C likes the post of B. You see, it is very reduced form, but at the same time powerful because it captures the essential underlying network structure. This is the original idea of Google, only replace likes by “weblinks”.
This is what I mean that not everything should be incentivized. Actually, if you see that you do effort and do not get rewards but others get while doing something that is “worse” in your opinion, you will feel less inclined to participate and may even drop out.
One more quick suggestion…how about we take that leader board and place it somewhere prominently on gov.yearn.fi page…it would be cool to keep track of all the best contributors and bring some spot light, plus would be a neat gamification “nudge” encouraging good contributors without going too overboard?