Scoring Model

CrowdProof's reputation scoring system transforms raw blockchain activity into normalized 0–1000 scores across multiple categories.

Score Categories

Category	Code	What It Measures
DeFi Lending	DEFI_LENDING	Lending/borrowing history, liquidation ratio, collateral management
DEX Trading	DEX_TRADING	Trading volume, consistency, MEV exposure
Governance	GOVERNANCE	Proposal voting, delegation activity, DAO participation
NFT Activity	NFT	Collection diversity, holding periods, marketplace behavior
Social Reputation	SOCIAL	On-chain social graph, attestations, endorsements
Credit History	CREDIT_HISTORY	Cross-protocol debt repayment, outstanding obligations

Score Tiers

Score Range	Tier	Meaning
800–1000	Excellent	Top-tier on-chain reputation
650–799	Good	Established, reliable participant
500–649	Fair	Moderate activity with some history
350–499	Poor	Limited or inconsistent history
0–349	Very Poor	New address or negative signals

Scoring Pipeline

Raw On-Chain Data
       │
       ▼
  Data Ingestion (Indexers)
       │
       ▼
  Feature Extraction
  ├── Transaction count & volume
  ├── Protocol diversity
  ├── Time-weighted activity
  ├── Liquidation events
  ├── Governance participation
  └── Debt repayment history
       │
       ▼
  ML Model Inference
  (per-category models)
       │
       ▼
  Score Normalization (0–1000)
       │
       ▼
  Confidence Calculation
       │
       ▼
  Merkle Root Commitment (on-chain)

Confidence Score

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

Confidence	Meaning
0.9–1.0	High confidence — extensive on-chain history
0.7–0.89	Moderate confidence — reasonable data points
0.5–0.69	Low confidence — limited history
< 0.5	Very low confidence — insufficient data for reliable scoring

Confidence is influenced by:

• Data volume — more transactions = higher confidence
• Data recency — recent activity weighted more heavily
• Cross-chain coverage — activity on multiple chains increases confidence
• Data consistency — conflicting signals lower confidence

Score Decay

Scores are not static. Inactive addresses experience time-based decay:

decayed_score = base_score × decay_factor
decay_factor = e^(-λt)

Where:

• λ (lambda) = 0.001 per day (configurable via governance)
• t = days since last on-chain activity

Days Inactive	Score Retention
30 days	~97%
90 days	~91%
180 days	~84%
365 days	~69%

Decay resets when new on-chain activity is detected.

Model Versioning

Each score response includes a modelVersion field (e.g., "v2.1.0"). Model updates are:

1. Tested on historical data for accuracy
2. Announced 7 days before activation
3. Deployed with the old model running in parallel for 48 hours
4. Governed — major model changes require DAO approval via GovernanceToken