Skip to main content
Help improve BevorAI’s accuracy and effectiveness by contributing high-quality labeled security data. Your expert knowledge directly enhances the AI models that power security analysis for the entire ecosystem.
Data labeling will be eligible for compensation in the future. If you’re an auditor interested in becoming an official labeling partner, reach out to contact@bevor.io to be considered.

​
Why Data Labeling Matters

BevorAI’s effectiveness depends on high-quality training data. When security experts like you label findings, you’re:
  • Improving Model Accuracy: Teaching AI to recognize subtle security patterns
  • Reducing False Positives: Helping models distinguish between real threats and benign code
  • Expanding Coverage: Contributing to detection of new vulnerability types
  • Building Community Knowledge: Sharing expertise to benefit all users

​
How to Label Findings

​
Through the Dashboard

  1. Access the Labeling Interface
    • Open app.bevor.io
    • Navigate to the Dashboard
    • Enable “Expert Mode” in settings
  2. Review and Label Findings
    • Run security analysis on contracts
    • Review each finding in the results panel
    • Use the labeling interface to mark findings as:
      • ✅ True Positive: Confirmed vulnerability
      • ❌ False Positive: Incorrect detection
      • ⚠️ Missed Vulnerability: Missed a human-identified vulnerability
  3. Add Detailed Annotations
    • Provide specific vulnerability classifications
    • Add severity assessments (Critical, High, Medium, Low)
    • Include remediation suggestions
    • Note any special context or edge cases

​
Labeling Quality Guidelines

​
Be Specific and Accurate

  • Use precise vulnerability classifications
  • Provide clear, actionable descriptions
  • Include relevant code snippets or line numbers
  • Reference established security standards (e.g., SWC registry)

​
Consider Context

  • Account for intended behavior vs. actual vulnerabilities
  • Consider the broader protocol context
  • Note any assumptions or prerequisites
  • Identify edge cases and boundary conditions

​
Maintain Consistency

  • Use standardized terminology
  • Follow established severity classifications
  • Apply consistent labeling criteria
  • Document any special cases or exceptions

​
Recognition and Rewards

Contributors to BevorAI’s security dataset receive:
  • Recognition: Public acknowledgment in our contributor hall of fame
  • Labeling Rewards: Earn for high-quality contributions
  • Early Access: Priority access to new features and capabilities
  • Community Status: Special roles in our Discord community

​
Best Practices

Focus on quality over quantity. A few well-labeled, thoroughly analyzed findings are more valuable than many hastily labeled ones.

​
Before Labeling

  • Understand the contract’s intended functionality
  • Review the broader protocol context
  • Check for existing similar vulnerabilities
  • Verify your analysis with multiple approaches

​
During Labeling

  • Be thorough in your analysis
  • Document your reasoning process
  • Include relevant references and sources
  • Consider multiple attack vectors

​
After Labeling

  • Review your labels for consistency
  • Update labels if you discover new information
  • Share insights with the community
  • Track the impact of your contributions
⌘I