Ranking Token Curated Registries

Last month I got a free ticket for the Ethereum Camp IoT Hackathon. I love hackathons but I have never done anything with IoT. Luckily there was a cryptoeconomic mechanism design challenge which didn’t require any hardware. I came up with an IoT-related use case for the thoughts on tokenized information signalling that we are developing at Way, pitched the idea and found a group of brilliant people to work with for 4 days. We ended up winning the third prize and also met with the main sponsor Weeve a few days later to further sharpen our mechanism. The following article outlines the main thoughts that arose during the whole process. It is written in plain language, does not require any technical background, but will demand some level of concentration and interest in the topic to get through ;)

Thanks Ethereum Camp for the organization + photos and weeve for sponsorship!

In this article I first explain why ranked lists are becoming increasingly important in the world’s information landscape. I then discuss cryptoecomic mechanisms that incentivize individuals to curate high-quality ranked lists without a trusted central party. I arrive at the conclusion that decentral ranking protocols will be a major threat for any non-ranked registry and discuss directions for further development. If you are familiar with the benefits of decentralization, Curation Markets and Token Curated Registries, you might want to jump straight to the second part.

Why good lists matter

Ordered lists direct our attention and thereby our actions

The biggest problem of the information age is information overload. The amount of available information grows at an ever accelerating pace while our attention remains constant with limited potential to absorb more. Thus, the most valuable and powerful companies of today are the ones that build information filters: Google and Facebook scan through the vast jungle of the internet and present us with a short list of relevant information for our every need. The best results are always displayed right on top where the finger of our pointer is already directed at them, so we do not waste time and energy clicking through suboptimal responses. Once these lists are trusted as a reliable and fast source of information, they start to determine our behavior. The company that created them can make a lot of money, simply by adding entries that are not necessarily valuable to the top of the list whenever someone pays it to do so. This is the essence of online advertisement.

The ongoing digitization of the human experience will further increase the need for information filtering. In the future we will be immersed in information streams coming not just from our computer screens and smart phones, but also from the everyday physical world. We will be surrounded by billions of machines and they will be trying to communicate with us. Some of these machines will accompany us throughout our lives and we will know them well. But others will only cross our path rarely, so we will need helpful recommendations to decide quickly which machine we can entrust with performing a given task. Whoever creates the best rankings of machine services and data providers will have the huge power of directing our attention and choices in the Internet of Things, a.k.a. the real world.

Decentralize listmaking

Modern information technology allows groups of strangers to communicate and cooperate with each other across long distances. But only thanks to blockchain cryptocurrencies, they can now also get financially rewarded for this cooperation on a large scale. Instead of a closed company that coordinates and pays a few people to solve a problem, we can now have global armies of enthusiasts using blockchain protocols to coordinate around a shared goal — and if they succeed everyone gets paid their share. Bitcoin is only the first example of such global grassroot collaboration. There are two significant benefits associated with this new organization of work:

First, we can reason that it will lead to a more fair distribution of income, especially in the context of natural monopoly forming, network-based technologies. We perceive it as unjust and undemocratic when a handful of big companies reap all the financial benefits of having secured large networks with high switching costs. It has become obvious that these companies sometimes use their monopoly power to enrich themselves at the expense of their users. Furthermore, as good software becomes cheaper to develop, we start to realize that these monopolies are built not primarily on the quality of proprietary technology. The software algorithms are only valuable because of the vast amounts of data that we, the users, pour into them. This data is essentially the continued recording of our behavior, of our choices, of our lives. Nevertheless we do not really own these recordings. They are kept in locked silos by our service providers and we cannot take them with us to another service provider. Blockchain-powered infrastructures give ownership of the most valuable network assets back to the people and promise to compensate all participants in a fair manner.

Second, it can be argued that decentralization of networks will lead to better services being produced by the network. In many cases, distributing the work across many shoulders and incentivizing each to contribute allows for more, better and more diverse work to be done. Data curation and listmaking could be such a case. As long as contributors are coordinated efficiently, if more people are actively contributing their skills and time to find, create and curate information, better information can be found and better lists can be created. Already in the web 2.0, it is the user who does the work of surfacing information and signalling its relevance by liking, linking and commenting. If one could get paid for these activities, people would probably get even better at them.

We have set the stage now by establishing the importance of lists and the benefits of decentralization. It is time to lead the two thoughts together and turn to the main question of this article: How can groups of people be coordinated and incentivized to curate high-quality lists without the ‘help’ of a central authority?

Ranking Token Curated Registries

The importance of lists has recently been recognized by some of the leading voices in the blockchain community and this has lead to the creation of the first blockchain protocols for decentralized list making. Mike Goldin and Simon de la Rouviere of consensys have been at the forefront of this development. Working on adchain, Mike implemented the Token Curated Registry, which is now finding widespread adoption by diverse projects in the space. In this part, we briefly introduce the fundamental idea of Token Curated Registries and then go on to discuss one of its main weaknesses: the absence of ranking.

TCR 1.0

In Token Curated Registries, a rare, tradable token is programmed to be useful for a very specific purpose: It gives its holder voting power in the decision over what entries to include and which entries to exclude in a list associated with that particular token. These entries are usually service providers. When token holders collectively do a good job at voting only good services into the list and keeping bad services out, the list becomes a useful signal for consumers to determine whether a given service is of high or low quality. It will then be desirable for services to be on the list, because being on the list gets you better business. This will presumably increase the price of the token since having the power to decide who gets on the list is now worth $$$, allowing early curators to earn from their curation work in the long term. They will be incentivized to do good curation work, until the marginal cost of this curation exceeds the financial benefit that can be derived from improving the list. This system can be used for any kind of exclusive list as long as it serves as a good focal point to consumers for a particular quality signal and as long as there are token holders willing and able to do the work of curating it.

However, there is one big problem with the standard TCR mechanism: It only governs the binary choice of who gets onto the list and who does not. The protocol does not specify how entries are ordered. As we have seen above, for many consumer-facing applications it is not enough to know that all entries of a list are equally good. When there are multiple options for a given need, a consumer will want the best ones to be on top. A TCR where there are multiple, very similar entries that are not ordered might even burden consumers, because they would be left alone with the decision between them. Hence, in many cases token holders will be incentivized to include only one service of each kind, presumably the best — along some dimension, implicitly agreed upon by the token holders — of all competitors for this service. Having more alternatives would make the list less useful because it would confuse the average consumer. But a consumer with more specific needs or whose definition of ‘best’ differs from the mainstream, will want to see more options. She would depend on a separate TCR existing for services of this particular kind and then do the work of considering each entry of that TCR individually. In an era where attention is the main scarcity, this would impose significant search costs. We have here an obvious frontier for improvement.

The remainder of this blog post therefore is dedicated to the consideration of mechanisms that make a TCR more useful by incentivizing a sensible ordering of its entries. These are mostly existing ideas that are currently being discussed and experimented with, without any final verdict having been made. I list and discuss them here in a consistent framework in order to motivate a more structured debate on which way to go. Let’s think through it together and discuss in the comments below.

Cranking it out, winner’s smile

Introducing order

The easiest way to order the entries of a TCR is by some objectively verifiable criteria. This is indeed what many projects in the space are planning to do in their first stage implementations. For example, the TCR data marketplace weeve is considering the number of challenges brought forward against an entry as a sorting metric. Others are proposing the staked deposit of each entry, which gives an indication of how confident a listee is about her status as a list-member (because this is what she would lose if curators decided to kick her out). But objective criteria are unlikely to be of much value in assessing differential quality at this level. The need for a TCR only arises in the first place, when quality is hard to assess on objective grounds and we therefore accept the aggregate of subjective voting as the best signal for quality.

We can do better, and apply the same logic that is used for the in/out decision of TCRs for the ranking of TCRs: The list is most valuable when the ordering of the list is useful for consumers. Therefore, it is in the long-term interest of token holders to create a good ranking making them the best agents to decide about it. There are three different ways to coordinate this collective decision: first, with binary pairwise voting, second, by including the rank position in the TCR’s application process, and third, by using the logic of standard curation markets.

In the first design, any listee who feels that she deserves a better ranking within the TCR could put down a stake and apply to rise by one rank overtaking the listee immediately above her. If challenged, this triggers a poll allowing token holders to decide in a simple binary fashion which of the two entries they consider better than the other. There are two problems with this mechanism: Moving one step up on a list might not be worth too much so only in clear cases will a listee risk the minimum stake, motivating token holders to curate this particular binary ranking. In less clear-cut cases no one might bother to change the ranking. Also, the design is quite hard on new entries: Presumably every entry would start off in the very bottom, but then for an entry that is clearly among the best, going all the way to the top would take a very long time and incur excessive transaction costs.

The second alternative design therefore has applicants to a TCR include their desired position in the application. Token holders can then decide on whether they want to include the candidate at the desired position or outright reject the application. This appears to me rather cumbersome, too, because a given applicant might have to forfeit several deposits before getting accepted. She might even give up before finding out what ranking token holders would find acceptable and without knowing if she has any chance of being on the list at all (unless she applied to the last rank). Ranking is important but it should not distort the beautifully designed in-out mechanism of a TCR.

Back to basics

I therefore move on to discuss a new mechanism inspired by the 101 curation markets design pioneered by Simon de la Rouviere, Maciej Olpinski of userfeeds and also by steemit. In this scheme, TCR token holders apart from choosing who to let in can use their balance to give weighted votes (endorsements) to existing entries. In the simplest design, when you own 100 tokens and you vote for an entry, that entry receives an endorsement of 100. A token holder can endorse multiple entries but she can only have one endorsement for any given entry. The entries are then sorted based on the total endorsement received. When we introduce variable weighting, i.e. users can choose with how many of their token they vote for a given entry, every user could create a personal ranking to reflect her preference and then the aggregate result would capture even relative differences well (this requires a more comprehensive analysis though).

During the hackathon, we used the above mechanism to create an ordered list of temperature data providers. If there are many competing temperature sensors, which one should we trust to have the correct temperature? ToT, trust of things.

One objection to this mechanism is that it would let bribing behavior go unpunished in the short term, because there is nothing at stake for a self-interested token holder except the slight decrease in token value due to reduced usefulness of the list if the bribing party does not deserve the ranking it buys itself with the bribe. To counter this, one could introduce rewards for early curators if the trend score of an entry goes into the same direction as the early curator’s vote. A form of this has been implemented in Slava Balasanov’s curation market design for relevant. Another option is to decouple the TCR ranking from the TCR, by introducing a separate token that can only be used for weighted endorsement voting within a given TCR. This would increase the disincentive to accept bribes, because the ensuing deterioration of ranking quality would hurt a token that is only valuable for ranking much more than it would hurt the overall TCR token.

Overall, I do not think that preventing bribes should be a major concern in a system where we introduce a market for attention. In my opinion it doesn’t even make sense to speak of bribery in this context. Google has created the best ranked lists in the world and they accept money from a few entries to be shown on top, but this is not exactly a bribe. The acceptance of money from listees or candidates in exchange for upvotes should be accepted as a legitimate (and fundamentally undetectable) way for token holders to monetize their curation work once they have developed an attractive ranking. At the same time, a given TCR will only work if token holders do not accept too many bribes, because that would render the list and thereby the token worthless. TCR ranking communities that engage in excessive bribe taking would be outcompeted by more honest ones. Google would be stupid to take too many bribes and plaster the entire screen with advertisement. Of course, it remains to be seen, if any cooperative of token holders can reach the same level of wisdom collectively.

I have discussed different mechanisms for the decentral curation of ordered lists. Let us conclude now by looking at where and how these lists might be applied in the future.


Like all crypto-economic mechanisms, TCRs thrive in competition. If in the medium term, a certain TCR does not deliver on its promise to curate a useful list for a consumer demand, another community can be expected to start a new TCR or fork the existing one, for the very same purpose with improved governance and parameters. In this article I have argued that the ordering of entries is of major importance for consumers and therefore a new TCR that offers a useful ranking without distorting or excessively complicating the basic mechanism, will render the existing TCR obsolete. It is therefore in the strong interest of every TCR community to start experimenting with the extensions discussed here.

Not only search results and lists of machines can be seen as ordered lists. Social media, news feeds, dating apps, marketplaces, maps, augmented reality layers — lists are everywhere. And whenever a user does not already know what exactly she is looking for or what path of action to take, i.e. in the realm of discovery, these lists are immensely powerful. If they are well curated and we trust them to help us quickly discover the most satisfying answer to our needs, they direct human behavior and attention whenever there is room for these to be directed at all. In a world where our options are abundant, nothing else is more valuable.

Today, big companies earn all the money that can be made with this power. With the help of well-designed TCRs, people like you and me can pool our resources, invest them in work and earn the financial rewards if we manage to create lists that people trust. Let’s figure this out together. Comment or email to achill@way.network.

Thanks for your attention.

Acknowledgements: Many of the above thoughts were inspired by or picked up from chats, emails, blog posts, videos and conversations with people on the following unordered list: Meher Roy, Maciej Olpinski, Simon de la Rouviere, Mike Goldin, Luke Duncan, Felipe Gaucho, Trent McConaghy, Sebastian Gajek, Sidd Bhasin, Chris Haug, Claudio Weck, David Terry, AB Rao, Max Roessner, Eckart Burgwedel.