Skip to main content

Source: ebisu/docs/adr/0059-trust-score-implementation.md | ✏️ Edit on GitHub

ADR-059: Trust Score Implementation

Status​

Implemented

Context​

The vessel trust scoring system was incomplete with placeholder functions. Trust scores are critical for the intelligence platform to assess data reliability and detect potential deception.

Decision​

Implemented comprehensive trust scoring that:

  1. Weights sources by authority level (AUTHORITATIVE=1.0, BLACKLIST=-0.5)
  2. Considers data completeness and consistency
  3. Tracks sighting frequency as confirmation value
  4. Provides summary statistics for source analysis

Implementation​

Created /scripts/create_trust_score_functions.sql with:

Core Functions​

  • calculate_trust_scores_for_source() - Batch trust score calculation
  • get_source_trust_summary() - Statistical analysis of trust distribution
  • calculate_vessel_trust_score() - Individual vessel scoring (already existed)

Scoring Algorithm​

trust_score = (
    0.4 * source_score +      -- Source authority weight
    0.3 * completeness +      -- Data completeness
    0.2 * consistency +       -- Cross-source agreement  
    0.1 * temporal           -- Recent vs old data
)

Results​

CCSBT Import:

  • 685 vessels (47%) achieved trust score ≥ 0.7
  • Average trust score: ~0.65
  • Validates multi-source confirmation value

Consequences​

Positive​

  • Quantifiable vessel reliability metrics
  • Automated risk assessment capability
  • Foundation for deception detection
  • Supports intelligence-based decision making

Negative​

  • Requires regular recalculation as new intelligence arrives
  • Complex scoring may need tuning based on real-world validation

Future Enhancements​

  1. Admiralty scoring integration (A1-F6 reliability/credibility)
  2. Deception probability calculations
  3. Behavioral pattern analysis
  4. Machine learning optimization of weights