$78.59 crowdfunded from 15 people
$1,126.59 received from matching pools
The Phantom Menace - Proposal - One click sybil scores
Sybils are a major threat to the daily operation of any decentralised organisation, as they undermine the trust and security of the network by skewing votes, gaming proposals and monopolising airdrops. Previous research has shown that sybils tend to send transactions only within their own community, and only when they intend to attack non-sybil communities. This behaviour creates clear "bridges" connecting sybils and non-sybils which can be used as a heuristic to detect attacks.
Our team has been working closely with Gitcoin to develop new techniques which combine graph analysis and machine learning for the automatic detection of patterns such as these bridges. Our grant in the prior round extracted new graph based motifs for sybil detection, providing high precision classification even without the addition of standard features like wallet age. You can read our report on this work here: https://docs.google.com/document/d/1gTvIoHd9tb-68BWFHJgJIbFnAVXMsMtw_UvUEgX2icA/
Building on this, we propose two parallel courses of action. Firstly, we intend to continue this research thread, implementing new detection methods. Secondly, these techniques should be packaged in an intuitive workflow that allows DAOs to easily implement them and understand the output.
The workflow will include the following steps. Given a new/suspicious wallet we will:
- Automatically obtain transaction data of the suspicious user.
- Run our previous sybil detection techniques to obtain their graph features and motifs.
- Identify bridges and connections between the suspicious user and other communities.
- Output a sybil detection score and report of the findings.
This workflow can also be repeated periodically for the whole community to further improve the model and remove bad actors that pass initial checks. The workflow will be available as a script which can be executed within the DAO’s infrastructure as well as within the Pometry UI currently under development.
The project will be led by a full-time research engineer (Dr. Haaroon Yousaf) for 6 months, supported by the wider Raphtory and Pometry team consisting of leading researchers in graph theory, machine learning and on-chain analytics. Dr. Yousaf is already employed by Pometry, having worked on the initial grant described above. His prior work can be seen here https://www.haaroonyousaf.com.
The total cost of this project will be £30,000, covering the salary of the researcher for the proposed period.
Nominations; DisruptionJoe and ale.k
One click sybil scores History
-
accepted into The Phantom Menace 1 year ago. 15 people contributed $79 to the project, and $1,127 of match funding was provided.