# How Arkham Labels Wallets

Arkham labels wallets by combining **on-chain behavior** with **public off-chain signals**, running that through our proprietary attribution system (**ULTRA**), and only publishing labels when they meet strict confidence thresholds.

## The Process (High Level)

### 1. Data Ingestion (Signals)

  On-chain
  
- Transfers and interactions
- Counterparty networks
- Contract behavior
- Timing patterns
  

  Off-chain (public)
  
- Public disclosures
- Official wallet posts
- Protocol docs
- Exchange/protocol announcements
- Public attribution sources
  

### 2. Attribution & Clustering (ULTRA)

ULTRA applies a mix of:

- **Address clustering** - linking related addresses
- **Address modeling** - behavioral patterns
- **ML-driven heuristics** - probabilistic attribution
- **Intelligence augmentation** - enrichment from external public signals
- **Internal consistency checks** - across multiple data sources

### 3. Confidence Gating

  Verified labels
  ≥98%

  Prediction labels
  ≥80%

Prediction labels are explicitly marked as lower-confidence.

### 4. Human Verification

Labels that are high visibility or high impact can be reviewed **case-by-case** by analysts before publication.

## Why This Matters

Labeling isn't a single "source of truth" problem. It's a **signal fusion** problem: ULTRA combines multiple independent signals and only publishes attribution when the evidence crosses defined confidence thresholds.

This approach provides:

- **Accuracy** - High confidence thresholds reduce false positives
- **Coverage** - ML-driven predictions expand coverage beyond manual labeling
- **Transparency** - Confidence scoring lets you decide how to weight attribution
- **Adaptability** - Continuous updates as new signals emerge