Getting more insights from the Ethereum Carbonvote
This is a part of the research on the improvement of the Ethereum governance, specifically by clustering of addresses, which participated in coinvotes and EIP-186 carbonvote specifically. Results are intended to provide some additional insights on different stakeholder stances (holders, users, app developers) for that and future governance proposals. Made with Malkevych Bohdan
Ethereum governance is a complex multi-stakeholders system governing the changes to the protocol. It’s a mixture of formal proposals management (EIPs) process and less formal signalling & discussion made through calls, forums, tiwtter, reddit, events and more. Actors include foundations (like Ethereum Foundation), developers, miners, users and probably some other groups.
Tennagraph was created to improve the governance by optimizing the work around signalling stakeholder stances, making collection easier and results more insightful. The first version including coin and gas voting, as well as twitter influencer stances, was launched earlier this year. We’ve tested it with a contentious EIP-1057. Now we’re researching the further development of the tech and have identified a number of challenges around the process.
The current process shown in the picture has several actors and steps. The problems are formed in the How Might We (HMW)-statements. There’s a lot of interest around increasing the participation in governance, but for our current research, we’ve selected the data analysis, interpretation and presentation opportunities.
Also, we’ve recognized that a vast data set can be found in the carbon vote used by the network for the EIP-186 (around 3% of all Ethereum participated in that). As we can play with this data, our finding and tools can be reused for any further coinvotes at any time, including tennagraph and carbonvote platforms. Specifically, our challenges (or “how might we”) here are:
- Extract more insights from the coin vote
- Show the results in a more clear manner
So we know some patterns in the type recognition
- HODLers keep big balances and rarely transact
- Users have some transactions with exchanges and basically any other dapps
- Dapp Devs addresses could be identified through the massive amount of transactions and gas spent
- Miners by the amount of mined blocks and the distance between accounts with such actions
So we wanted to show the preference graph on the axis and also outline the votes by identified profiles.
Disclaimer: This is the first test, probably includes inaccuracies and errors. Also there is a small amount of data for the analysis so the bias can be big. To be improved after receiving more feedback.
- All calculations were done on block ~ 7957000 (14.06.2019)
- All transactions were exported from Etherscan.
- Balances for each address were discovered with the help of infura.
- Gas consumption was calculated with the help of open source explorer blockscout.
For each address that took part in the voting were calculated the consumed gas and balance. So, here’re visualizations on the voting:
Separated plots to find insights for each option separately:
Cumulative GAS spent
To find some insights by clusterization, all participants were split using a median on 3 equal groups.
User type 1 — Users which consume a small amount of gas and have a small amount of eth on their wallets.
User type 2 — The middle by the median.
User type 3 — Users which consume a lot of gas or users who have a big amount of ETH on their wallets.
- 0 — NO
- 1 — YES: 0 ≤ reward < 1.5
- 2 — YES: 1.5 ≤ reward < 2
- 3 — YES: 2 ≤ reward < 3
- 4 — YES: 3 ≤ reward < 4
- 5 — YES: reward ≥ 4
Picture 5 demonstrates the frequency of voting by each group of users.
Current findings on EIP-186 include:
- Users who use ethereum platform more than just as a wallet voted for case: YES: 1.5 ≤ reward < 2
- Users which have a lot of ETH on wallets voted for case: YES: reward ≥ 4
- Discovered the trend and active period of the voting
- Around 6 big players that took part in the voting
- Voting with gas has more equal distribution.
- Users with higher balances don’t want to change the rewards or change it to the next range 1.5 ≤ reward < 3
- The users with the lower balances want to change it to the range 0 ≤ reward < 1.5
We could automate such calculations for other research in the Tennagraph app.
The code and data is published here: https://github.com/TennaGraph/DataAnalysis/blob/master/Test%20Data%20analysis.ipynb
We’ll take them for further research in the future.
We want to validate a couple of hypotheses around this concept. If you’d like to help — let us know (in a comment or direct messages)
- How insightful can be such data?
- What amount of votes should be considered representative
- Which stakeholder groups/foundations can use this data?
- How possible/viable is using/presenting this data for future EIP decisions?
- How do our algorithms/analysis look like technically? What can be improved?
Thanks in advance. It’s an early and small test, we believe there’s a wide room for the improvement. So any criticism, suggestions and guidance is very helpful.