Reputation Scores

CrowdProof generates reputation scores by analyzing on-chain activity across multiple DeFi protocols. Each wallet receives scores in 8 independent categories plus a weighted composite.

Score Categories

Category	What It Measures	Key Protocols
OVERALL	Weighted composite across all categories	All
DEFI_LENDING	Borrows, repayments, liquidation history	Aave, Compound, MakerDAO
DEFI_TRADING	DEX volume, LP participation, slippage patterns	Uniswap, Curve
GOVERNANCE	DAO voting frequency, delegation, proposal authorship	All governance tokens
NFT	Collection curation, trading patterns, blue-chip holdings	OpenSea, Blur
IDENTITY	ENS ownership, Lens profiles, on-chain attestations	ENS, Lens, EAS
SOCIAL	Social graph size, attestation quality, community trust	Lens, Farcaster
DEVELOPER	Smart contract deployments, protocol contributions	On-chain deploys
COMPLIANCE	KYC completion, sanctions clearance	CrowdProof KYC

Score Range & Tiers

Scores range from 0 to 1000:

Range	Tier	Typical Use
0–199	Very Low	New wallets, minimal activity
200–399	Low	Some on-chain history
400–599	Medium	Regular DeFi participant
600–799	High	Active, reliable user
800–1000	Very High	Power user, protocol contributor

Confidence

Each score includes a confidence value (0.0–1.0) indicating how certain the model is about the score:

• > 0.8 — High confidence, sufficient data
• 0.5–0.8 — Moderate confidence, limited data in some areas
• < 0.5 — Low confidence, score may be unreliable

Score Decay

Confidence decays over time if no new on-chain activity is detected. The decay rate is 5% per 30 days (compound), reflecting that older data becomes less predictive of current behavior.

decayedConfidence = confidence × (0.95 ^ daysSinceLastUpdate / 30)

Querying Scores

All scores for an address

GET /api/v1/reputation/{address}

Single category

GET /api/v1/reputation/{address}/{category}

Batch query (up to 100 addresses)

POST /api/v1/reputation/batch
Content-Type: application/json

{"addresses": ["0xABCD...", "0x1234..."], "category": "OVERALL"}

How Scoring Works

1. Indexer processes on-chain events from supported protocols across 5 chains
2. Feature extraction converts raw events into behavioral features (e.g., repayment rate, voting frequency)
3. ML models (LightGBM) score each category independently
4. Confidence calibration ensures scores reflect actual prediction quality
5. Decay function reduces confidence over time without new activity
6. Composite weighted average produces the OVERALL score