A Brief History of Conviction Voting

Michael Zargham
Published in
4 min readDec 26, 2020


And other musings on algorithmic policy-making

I met the Aragon founders at ETH Buenos Aires May 2018, hung out with the

, , among many others. I had spent late 2017 and early 2018, wrapping up a paper on the mathematical framework I use for designing mechanisms in dynamical systems, and working with and , applying that framework to complete the proofs underpinning the Balancer Protocol. Notably, the Balancer protocol calls for subjective parameter choices in the form of tokens supported, and weights of pools. It was already clear to many active researchers that governance of parameters was not a hypothetical problem, but one which would eventually come to dominate the political landscape of blockchain-based ecosystems.

Two months later, I attended Tech Crunch Blockchain in Zug July of 2018 and token’s role in governance was a very common conversation point. Feeling motivated, I pulled an all nighter in Zug and wrote my first conviction voting PoC, an algorithm+simulation by framing governance as dynamic average consensus. Note that this was for continuous parameter selection not discrete choices. While in Zug, I met

(founder of Giveth), someone I’ve come to respect greatly — nobody I’ve met in the crypto space has been more giving, a one-man proof that you cannot reduce all humans to selfish agents.

Diagram Framing Social Choice as a dynamic process for estimation and execution of collective preference. (figure from M. Zargham, Sensor Networks and Social Choice, 2018)

Later that July, I presented my ongoing work at the Ninth Annual Conference on Complex Systems, in Cambridge Massachusetts. There I met Frank Wang, a partner at the research firm X-Order. Over the summer, I continued to mess around with variations of conviction voting; one of the use cases that drove my explorations was the (dynamic) selection of generative art algorithms for Artonomous, a tech-art project by

. At Frank’s request, I drafted a proposal for continued research on conviction voting which contained the description of governance as “sensor fusion” which included formalizations for both the discrete choice (as in Artonomous) and continuous parameter selection as a collective preference estimation problem.

Stock and Flow diagram accounting for mechanism compositions required to achieve the full vision of the Autonomous Artist: “Artonomous” (figure from M. Zargham, Sensor Networks and Social Choice, 2018)

A few months later in Prague for Devcon IV, Griff introduced me to

and we spent hours in cafes jamming on the ideas that became the backbone of the Commons Stack.

Those discussions have gone from theoretical considerations to practical concerns. I saw this tweet today:


The reason the Commons Stack exists, and the reason the focus has been more on norm formation, understanding boundaries of communities in terms of culture, and teaching principles carried forward from Ostrom’s work is exactly this observation. Governance mechanisms don’t do anything directly, they merely convert the signals expressed by their participants into a collective decision. Acknowledging that value is inter-subjective, it is incredibly important to reflect on what values you express through your participation in governance processes — and the extent to which your community shares those value determines the direction your governance process steers your system.

To keep this article from getting exceptionally long, I am going to fast forward past some really cool things that happened in the world of governance and algorthmic policy-making. Some highlights though — in February 2019, Gitcoin Launch its first matching round and the MolochDAO was deployed. These are among the longest running social ‘signal processing’ systems in the crypto-space. In early 2019, I was also very involved with SourceCred — a project about which I am very bullish. Again, the key concepts are: algorithms and their parameter selections are subjective choices and how framing these choices in a dynamic game provides insight into governance questions.

The on-going development of cadCAD has been motivated by a desire to use dynamic games as a frame for evaluating mechanisms (and parameters) for their signal processing properties. For Conviction Voting in particular, quite a lot of modeling has taken place both before and after the deployment of the algorithm by the 1hive DAO.

Here is an outline of relevant documentation:

Discussion on Decentralized Organizations w/ @Sys_innovation & @jeffemmett

Conviction voting is one of many potential algorithms to help bridge the gap between the subjective realm of policy making. You can read more about my research on governance:

I have a range on-going research projects including computer aided governance in the Filecoin Ecosystem, DAO-to-DAO interaction patterns with

, and broader questions about governance of online spaces with Metagov.org. Specifically on the topic of time-weighted voting, the conviction voting work is currently focused on collecting and analyzing data from the 1hive DAO. Other researchers in space have addressed time weighting, eg, Commitment Voting, Chris Berg, Sinclair Davidson and Jason Potts.



Michael Zargham

Founder, Researcher, Decision Engineer, Data Scientist; PhD in systems engineering, control of networks.