An ethnography of SourceCred’s CredSperiment

My full ethnography of SourceCred’s CredSperiment is now on SSRN. Here’s a short summary of what I learnt from observing this astounding attempt to live in a radically different world👇

During the CredSperiment, people were paid 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. In return for building and maintaining the SourceCred product, contributors to SourceCred accepted payment for their contributions based on sophisticated calculations called “Cred scores”, created from the SourceCred software.

The SourceCred software ‘saw’ actions and people, but Cred scores were derived from dependencies and appreciation, such as whether a contribution was taken up or liked by others. Rewards were allocated according to Cred scores via a secondary token called Grain. Anyone could join, offer what they felt was needed, and even be rewarded for their contribution long after they had left (if it was picked up by others). The CredSperiment ended in 2022. The learnings are important for web3 generally.

Evan Miyazono of Protocol Labs likened the SourceCred software to a hydraulic water system. This is the Shushtar historical hydraulic water system (Iran). Image: Hassam Saeb. CC4.0

A very short summary of events

For GitHub contributions, SourceCred’s system was relatively unproblematic. However, when they created and implemented plugins for Discourse and Discord in order to reward other forms of contributions, things started to go awry.

In particular, the #props and #didathing Discord channels required people to signal their activities so the algorithm could see emoji reactions to those posts, and this changed people’s relationship to the algorithm and each other.

Attempting to reward non-code contributions was an important and necessary development. However, without adequate systems in place for managing weights and collectively steering the algorithm, rewards didn’t necessarily go where they were needed to keep the mission on track.

Community-building was amplified (not a bad thing in certain contexts!). Some people were concerned that others were taking on ‘side quests’ — community-oriented tasks that were more easily communicated to non-technical people than work on the code (hence receiving high rewards). Many of these tasks were important for needs such as onboarding, record keeping etc, while others were important for community wellbeing.

However, this took resources away from work on the product, creating tensions around ‘extractive behaviour’. This went against the original vision that SourceCred would automate rewards in the background so that “people feel free to just contribute for the joy of it, without worrying if a particular transaction gets me x$” (Dandelion).

SourceCred intended to build components that could have mitigated these tendencies, but they were not completed in time for the CredSperiment to experience the outcomes. These various solutions involved making it easier for non-technical people to use SourceCred.

Most importantly, they would have added legibility to processes such as collectively weighting what was rewarded. Contribution systems only work when people can see, understand and adjust the systems that determine value. When this happens, it’s a form of planning and governance in itself. One unresolved area is how to provide people with agency over membership boundaries.

Why it matters

For me, the CredSperiment changed how I think of web3 generally. Some kernels of that thinking are in the essay, including:

1. Systems for adding more localised information into value-formation processes are important for the problems that society faces, and SourceCred was a sophisticated attempt to do that (even if it failed). Prices as computed by markets are a poor measure of value. Web3 gives us the tools for better computation. SourceCred tried to construct a coordination mechanism that could ‘outcompete capitalism’ (Dandelion’s words). Undertaking this ethnography of SourceCred helped me to better understand regenerative economies.
2. Web3 cares about public goods, but ‘contribution goods’ is a more useful framework for how these systems can and should evolve. With contribution goods, those who put more in get more out (either resources or reputation), and the benefits to contributors rise as more high-quality contributors join. There can be a public good produced as a byproduct (such as a public blockchain), but we need ways to see and reward contributors, and which help people to coordinate in a permissionless manner. Achieving this might counteract the speculative nature of the web3 economy and token-weighted governance. In some cases, contributions will be performed by machines with human minders (validators are an example of this in my experience).
3. SourceCred dismantled the notion of the worker in what I consider to be a productive way. A contribution is not a role or a job (or necessarily performed by a human). The system removed workplace strcutures and expectations with positive outcomes for people’s relationships[1].
4. SourceCred also gave shape to some ephemeral aspects of participation. With systems like this we become less reliant on an ideal notion of the unburdened, virtuous citizen who has infinite time and attention. We don’t need to think of participation as programs that get people ‘to participate’. Participation is more about what needs doing and the value that it generates for the group.
5. Governance changes as a result. The governance that SourceCred pioneered is what I call ‘terraforming’. It’s a bit like the concept of ‘environmentality’ (a milieux-based governance where action is taken over the rules of the game rather than directly on the players), but where modulation of the system is done by the local community.

The truly transformative possibility SourceCred was to replace capitalism’s version of value-formation and its vampiric tendencies with a more sophisticated, information-abundant, locally responsive cybernetic system. I am now working with @profjasonpotts on a theory arising from this case study and others. Stay tuned.

Acknowledgements

I am grateful to the SourceCred community for letting me observe part of the journey and providing their consent for the release of the full article [2]. They also helped me pilot the Telescope bot (see this article for full background on the Telescope digital ethnography tool). Ethnography is deliberately subjective whilst also synthesising collectively-attained knowledge. Therefore, while many people contributed to the ethnography (see article for full list), they will also have their own views about what occurred.

[1] There were also negative outcomes for people including interpersonal problems. These may have been mitigated if the system had the means to be more actively steered by the contributors — or not. It’s impossible to say from this particular case study.

[2] I haven’t decided if/how I will publish this piece. Consider it an unwieldy director’s cut for now (typos and all).

I am a Professor at RMIT University in Melbourne, working across the RMIT Blockchain Innovation Hub and the Digital Ethnography Research Centre. I am also an Associate Investigator (AI 🤖) of the ARC Centre of Excellence for Automated Decision-Making and Society and a Research Director in Metagov, which is how I came to observe the CredSperiment and trial the Telescope. I acknowledge the support of the Australian Research Council, FT19010372. I use this Medium blog for work-in-progress ideas and reflections.