🛡️Anti-Gaming & Sybil Resistance
We build on validated P2P lending research showing that friends who vouch with capital and lenders who evaluate borrowers significantly reduce defaults [12].
Phase 0 tests these findings on Farcaster and tracks emerging attack patterns, allowing us to deploy additional defenses if needed.
Security Layers
On-chain reputation
🤷 Logical
Makes Sybils expensive
Our Approach
Principles:
Economic alignment > algorithms — Lenders risk money, so lender evaluation matters most
Data-driven iteration — Phase 0 reveals what works, we adapt accordingly
Layered defense — Multiple protections create resilience
Expected evolution:
Phase 0: Social trust + basic Sybil resistance
Phase 1: Add cashflow verification (harder to fake bank statements than social graphs)
Phase 2: Auto-repayment reduces strategic default incentives
Phase 3+: Machine learning on repayment patterns, cross-platform reputation
Defense Layers
Layer 1: Economic Protections (Strongest Defense)
Friends vouch with capital:
Every contribution requires actual money, not endorsements
Friends risk their capital if borrower defaults
Eliminates "cheap talk" problem [16]
Lender evaluation:
Lenders use social and financial signals to assess borrowers [12]
Early lenders signal quality, attracting others to participate [84]
Trust scores + on-chain reputation provide the information lenders need
Layer 2: Bot & Spam Filtering (Validated)
Farcaster-native quality signals:
Neynar scores: 0-1 scale measuring account quality [87]
Farcaster spam labels: ML predictions based on activity patterns and community moderation [90]
OpenRank trust scores: Graph-based reputation, updated every 2 hours [89]
These systems filter low-quality accounts (bots, spammers, low-quality AI) before they impact trust scores.
Layer 3: Temporal & On-Chain Signals
Time-based protections:
Account age and connection stability
Loan size limits for new borrowers
Growth patterns flagged if suspicious
Short loan terms (30-90 days) provide fast feedback on what predicts repayment
Permanent on-chain reputation:
Default history visible to all future lenders
Can't create fresh identity after default
Makes Sybil attacks expensive (need capital + reputation for each identity)
Layer 4: Network Analysis (Validated, Available If Needed)
Available if sophisticated attacks emerge:
Graph neural networks (HGNNs) deployed at major financial institutions
Behavioral pattern analysis: posting cadence, interaction networks, account evolution
Community detection algorithms identify coordinated fraud rings
Farcaster-specific signals: FID history, reaction patterns, content repetition
Economic protections (friends vouch with capital + lender evaluation) come first. These methods are deployed by major platforms and available if needed.
Attack Vectors We Track
Basic Sybil attacks:
Fake accounts to boost trust scores
Artificial social connections
Coordinated bot networks
Sophisticated attacks:
Coordinated small networks (real people creating tight clusters of fake accounts)
Time-based reputation farming (build trust with small loans, default on large one)
Collusion rings (real accounts coordinate to defraud lenders)
Unusual network topology (circular vouching patterns)
Default clustering (multiple defaults from same social cluster)
Phase 0 data reveals which attacks emerge. We adapt defenses based on real patterns.
Next: Social Trust Scoring · Risk Scoring · Risk & Default Handling
Last updated