Connection-Oriented Cluster Matching (COCM)

COCM is our sybil detection system. It identifies clusters of strongly connected accounts based on shared identity signals, behavioral patterns, and graph relationships. It ensures fair funding by detecting and limiting the influence of suspicious voter groups.

---

Purpose

COCM ensures sybil resistance by:

• Detecting collusive or duplicate voter accounts

• Quantifying connection strength between accounts

• Disqualifying or penalizing clusters with sybil-like behavior

If a group of accounts appears artificially connected (e.g. sharing too many identity traits or behaviors), they can be:

• Flagged for reduced voting power

• Excluded from voting or matching altogether

---

How It Works

COCM builds a connection graph where each node is a voter and each edge represents a weighted connection.

For any two voters i and j:

$$C(i,j)= ∑k=1 n ​ w k ​ ⋅f k ​ (i,j)$$

Where:

• fk(i,j): normalized score for feature k (e.g. shared credentials, co-voting frequency, wallet linkages)

• w_k: weight assigned to feature k (tuned by the platform)

---

Clustering & Filtering

1. Edges are drawn if connection weight C(i,j)≥TC(i, j)

2. Voters are grouped into clusters

3. Each cluster is evaluated for:

   • Internal density (how tightly connected the members are)

   • Behavioral diversity (do they act like unique humans?)

If a cluster fails diversity checks and has high internal coherence, it is flagged as a sybil cluster.

---

What Happens to Suspicious Clusters?

• High-risk clusters may be fully excluded

• Or their voting power is reduced via suspicion score adjustment

• This ensures one-person-one-vote dynamics without revealing private identities

---

Benefits

• Prevents vote manipulation

• Protects matching pool integrity

• Enhances trust in the funding process