Observing the CredSperiment

SourceCred’s CredSperiment was a long running attempt at a permissionless organisation. At the commencement of the CredSperiment, SourceCred’s ‘Temporary Benevolent Dictator’ described it as akin to laying the train tracks while the train is already in motion. In March 2022, SourceCred hit the brakes. This post looks back at what the CredSperiment achieved.

Introduction

The CredSperiment paid people according to their contributions within a permissionless organisation, meaning an organisation where there is no boss and people can enter and exit at their will. During the CredSperiment, contributions were recorded in a way that was both relational and observable using a sophisticated software product called CredRank. It was designed so the community could determine the value of contributions – not just keeping humans in the loop but giving them real governance power over the protocol and their shared endeavours.

SourceCred, the creators of the software, undertook the Credsperiment, paying people with funding provided by Protocol Labs. In a post announcing the CredSperiment, founder Dandelion Mane wrote that the aim was to “learn more about how SourceCred works in practice”, including testing the incentives related to Cred with real money involved, and understanding what features would be needed to represent the different kinds of contributions. The CredSperiment was announced in August 2019, about a year after development and community building had commenced. In March 2022, SourceCred decided that the organisation would be wound down and that the product would revert to being a regular open source project.

The Credsperiment demonstrated that software for calculating contributions can alter people’s sense of purpose (with implications for the organisation), the tasks they take on, and their relationships with each other. Permissionless organisations cannot scale without such tools, yet we can only know their outcomes by examining them in action (laying the train tracks while the train is already moving). This post is an attempt to bring together the various views on what happened during the Credsperiment rather than present a fully formed analysis. Those who were involved in SourceCred should consider it a draft article for feedback.

First, it is important to note that the SourceCred product continues to be used in web3 projects that have implemented it to varying degrees and in their own way. At least one of those groups has intentions to keep developing SourceCred the product. It’s entirely possible that SourceCred will endure. Maybe it can still change how we value and reward what gets done in the world.

The CredSperiment refers only to SourceCred’s use of their own product, not the experience of these other groups. For a full analysis we would need to unbundle the CredSperiment into the following components:

  • SourceCred Product (data collect and CredRank implementation)
  • The choice of plugins and weighting parameters
  • Grain monetary policy (grain being the token used for paying people)
  • The business model (the means by which SourceCred might have become financially sustainable)
  • Community norms (not everyone held the same values, which influences who remains involved).

I touch on some of these components below.

1 The Contributions of the CredSperiment

Even though the CredSperiment is over, it achieved a great deal.

1.1 A pre-DAO innovation in DAO tooling.

The software used during the Credsperiment took information about contributions and encoded them with degrees of importance. Information about contributors and contributions was fed from plugins (initially GitHub, followed by Discourse and Discord) and built into a graph data structure. A contribution would receive Cred according to whether it was validated by future contributions, with time dimensions to ensure that current impact was factored in. The CredRank algorithm was then run over the graph to generate scores that were used for payouts. The nodes of the graph (contributions) and edges (connection between the nodes) could be weighted by the community, effectively allowing them to set parameters according to their own values and priorities. The code was sophisticated and interesting enough to attract smart people.

1.2 People got paid.

The intended outcome was that people would be rewarded for what they did rather than their role in the organisation. As was pointed out by a contributor in one meeting, the big achievement of the CredSperiment was that people got paid, many from marginalised groups, and they were paid to work on a software product without needing to face Silicon Valley’s hiring barriers.

1.3 Permissionless organisations can create different working relationships, loyalties and opportunities.

Although the community did not use all the affordances of the software to alter what was valued when problems arose (see below), those who participated got to live in what someone described as ‘a radically different world’. SourceCred helped people to see work and need in new ways, and with the kind of understanding that only comes through experience. Even as SourceCred was on the verge of winding down, one member proposed that they sell “source pills”, helping others to shake the habits (and damage) accrued through work in the capitalist economy.

1.4 SourceCred was an experiment in the custodianship of a protocol by those it affects.

People were vocal about what went down in the CredSperiment. Problems were raised and debated. For instance, the #didathing and #props channels – workarounds that lasted too long - were contentious because they encouraged people to ‘audit each other’. Even if the problem was never resolved, people were actively concerned about designing systems that compounded good emotions and mutual benefit rather than APY.

1.5 Identity and Cred weighted voting.

Seth Benton wrote a blog post arguing that SourceCred could resolve at least some of the identity problems associated with web3. It did, and not just for the Credsperiment (see MakerDAO’s use of the Discourse plugin).

2. Difficulties and contentious issues

The following are things that people expressed frustration or regret about:

2.1 CredRank was complex and making it legible didn’t happen.

During the CredSperiment, developer Thena decided to simplify the SourceCred product, resulting in what was known as CredEquate. Their justification was that “it’s hard to make any real meaningful analytics about a contribution without traversing the whole graph and seeing a contribution in the wider context of how cred is flowing through the graph”. Whether this change would have altered the outcomes of the Credsperiment is hard to know as CredEquate was completed only weeks before the organisation decided to wind down.

Some I spoke to believe that CredRank was workable, but that SourceCred did not develop the tools that would enable communities to observe contributions and easily alter weightings. It seems possible that the CredSperiment ended before SourceCred could complete the element that might have aligned contributors around a shared goal. Making contributions more legible, along with an easy interface for assigning value to grouped tasks (the Creditor), was frequently discussed but not realised.

Meanwhile, ‘the algorithm’ turned attention into anxiety as people had no means to easily see and alter the system as needed. The Discord plugin was at the centre of that tension, not just because it was the primary tool for signalling contributions that were non-technical in nature, but because it required people to spruik their own efforts and assess each other’s contributions.

2.2 What got rewarded was not necessarily what progressed the product

In their 2019 comments, Dandelion and William hint at the possibility of people gaming the system.

While I do not believe that people did game the system, certain behaviours may have led to some having more power than others. For instance, knowing what was recognised and being forthright about one’s contributions could be interpreted as manipulation – being loud while others held back. Such dynamics are typical of community organisations, but perhaps they were amplified through the algorithm. Some people told me that social interactions were rewarded over work that helped to progress the product. It’s also possible that there was no conscious or unconscious power play, just the perception of it, which may have been mitigated through tools such as the Creditor.

2.3 Invisible work

An unresolved tension was the degree to which SourceCred rewarded otherwise invisible labour, such as convening meetings or looking out for each other. Discord could be used to signal and acknowledge this work (with the plugin responding to emoji reactions) and people did use it for this purpose. However, those who were not comfortable promoting their contributions risked not getting paid. Introversion led to feeling undervalued.

At the same time, people wanted to express gratitude to each other without that being tied to financial allocations. People changed their behaviour rather than changing the system.

2.5 Membership boundaries

In a permissionless organisation, anyone can arrive and offer their skills, even if those skills are not what’s needed at a particular point in time. While this is discussed in web3 commentary as an ‘onboarding issue’, when that work is rewarded it fundamentally changes the organisation. It may redirect energy and attention to the detriment of things that need doing. Finding alignment around a goal, and having the tools and governance to stick to the plan, is even more important when anyone can make a claim on the organisation’s funds.

3. Conclusion: Looking back at looking forward

In two podcasts episodes released in late 2019, SourceCred contributor Seth Benton asks his guests “if SourceCred fails, what do you think will be the cause?”.

Dandelion Mane, the founder, and William Chargin, a developer who worked on the first iteration with Dandelion, provided the following predictions:

3.1 If SourceCred fails it will be because they were unable, or unready, to anticipate how the product would work when new people entered the organisation.

At this point, Dandelion is the Temporary Benevolent Dictator of SourceCred and they worry the project is reliant on them, having not reached the point of “self-sustaining progress, where it’s able to get its own resources and keep unfolding its own potential”.

William thinks failure will not be due to lack of people, but because a lot of people will come into SourceCred when “we don’t have the mechanisms to help those people engage productively”. New people will not work in pursuit of the goals that the current group was trying to achieve. Avoiding this scenario would require “having positive feedback loops to help bring people into the community and negative feedback loops to help direct that energy”. William adds that these new people could be “completely well-meaning but don’t understand the limitations and try to push it too far. And become disillusioned with it and then become detractors”.

3.2 SourceCred will have failed because the code is not finished or fails to work as intended.

It might fail because “there were some fundamental mistakes in the conceptual set of SourceCred some contradictions we can’t work our way out of through clever governance or clever community coordination or design”, says Dandelion. William posits that some people could also game the system while it was still “in trust level 2” (meaning it relied on people knowing each other) and before SourceCred had the systems in place to counteract such attacks. These could be attacks on “the regulatory system itself” or something unforeseen.

“We will just have to play this game to find out where it goes”, says Dandelion.

In hindsight, were Dandelion and William’s answers Cred-ible?

Let me know your thoughts. You can do this by joining the Wind Down forum on 12 May 2022 at 6.30pm PT. You can also leave comments below or contact me direct to talk further: ellie.rennie@rmit.edu.au or Ellie R. on the SourceCred Discord (DMs open).

About the research: I am an ethnographer based at RMIT university. I was permitted to observe SourceCred for what was to be the final 9 months of the CredSperiment. My colleagues and I in the metagov research collective created a Discord bot to assist us with that ethnographic endeavour, which the SourceCred community generously helped us to pilot. In addition to Discord posts, I read through the Discourse (governance) discussion forum and sat in on, or listened back to, several key meetings. I had one-on-one discussions with some former and current members who I cannot identify or cite here (I am awaiting consent), but their inputs undoubtedly informed what I was able to see in the data I collected. Michael Zargham, an early contributor to SourceCred, has been collaborating with me on this research, providing important insight into the design and intentions of the project. For the next stage of the research we intend to triangulate the ethnographic observations presented here with analysis of data generated by SourceCred.

6 Likes

thanks for providing these observations & gathering feedback across the community.

3 Likes