Skip to content

Precedent Network

The precedent network provides case-based reasoning through semantic similarity matching. Cases with similar ethical situations connect through embedding-based similarity scores.

Overview

Precedent discovery identifies cases with similar:

  • Fact patterns
  • Ethical reasoning
  • Code provisions
  • Professional contexts

Accessing Precedent Discovery

From Navigation

Navigate to Case Relations > Similar Cases in the navigation bar.

Direct URL: /cases/precedents/

From Cases List

Each case card includes a Find Similar link (blue badge) that opens precedent discovery with that case pre-selected.

From Case Structure

On any case Structure page, click Find More Precedents in the Similar Cases sidebar.

Precedent Discovery Precedent discovery page showing similar cases ranked by component-aware similarity

Using Precedent Discovery

Select Source Case

  1. Open Precedents page
  2. Select a case from the dropdown
  3. View similar cases ranked by similarity

View Matches

Results are presented as ranked cards, one per similar case. Each card shows:

Element Description
% Similar Overall component-aware similarity
Retrieval Scores Component similarity (the weighted nine-component embedding), provision overlap, and tag overlap
Component Breakdown Expandable per-component similarity across the nine D-tuple components
Section Comparison Collapsible facts/discussion similarity, shown for reference and not used in scoring

Explore Matches

Click any match to view case details and compare analysis.

Interpreting Results

High Similarity (0.8+)

Cases with very similar fact patterns, ethical issues, and professional contexts. Suitable for direct precedent comparison.

Moderate Similarity (0.5-0.8)

Cases sharing related ethical concepts, similar role structures, or comparable dilemmas. Provides broader context.

Low Similarity (< 0.5)

Cases with limited connection but potential insights.

Similarity Network Visualization

The Similarity Network provides a visual overview of all cases and their relationships.

Navigate to Case Relations > Similarity Network or direct URL: /cases/precedents/network

Similarity Network Force-directed graph showing case relationships with color-coded outcomes

Node Colors (by Outcome)

Color Outcome
Green Ethical
Red Unethical
Orange Mixed
Gray Unknown/Unclear

Edge Colors (by Similarity Score)

Color Score Range Meaning
Green > 0.5 High similarity
Yellow 0.3 - 0.5 Moderate similarity
Red < 0.3 Low similarity

Interacting with the Network

Click a Node - Displays case details panel showing title, outcome type, NSPE Code provisions, entity count, and connection count.

Click an Edge - Displays similarity breakdown with component scores for facts, discussion, provisions, outcome, tags, and principles.

Navigation - Drag nodes to reposition, scroll to zoom, click background to pan.

Filtering Options

Minimum Score Threshold

Use the Min dropdown to filter edges:

Threshold Result
0.2 Most relationships
0.3 Default balanced view
0.4 Higher similarity only
0.5 Strong relationships only

Similarity Component Filters

Filter Description
Provisions Cases sharing NSPE Code sections
Tags Shared subject tags

Entity-overlap filters restrict edges to cases sharing extracted entities.

Entity Overlap Filters

Filter by shared entities using the concept buttons (R, P, O, S, Rs, A, E, Ca, Cs).

Layout Options

Layout Description Best For
Force Physics-based simulation Discovering clusters
Circular All cases in a circle Seeing all connections
Grid Regular grid pattern Dense networks
Radial Grouped by outcome Comparing outcomes

Focus Mode

Focused Network Network with a specific case focused and highlighted

Use URL parameter to focus: /cases/precedents/network?case_id=7

Similarity Components

The network uses six similarity factors:

Component Method Description
Facts Similarity Cosine Semantic similarity of case facts
Discussion Similarity Cosine Semantic similarity of ethical analysis
Provision Overlap Jaccard NSPE Code section overlap
Outcome Alignment Categorical Ethical/unethical match
Tag Overlap Jaccard Subject tag overlap
Principle Overlap Jaccard Ethical principle conflicts

Technical Details

Embedding Model

Default: all-MiniLM-L6-v2 (sentence-transformers)

  • 384 dimensions
  • Optimized for semantic similarity

Similarity Calculation

Cosine similarity between embeddings:

  • 1.0 = Identical
  • 0.8+ = Very similar
  • 0.5-0.8 = Moderately similar
  • < 0.5 = Dissimilar

The overall score is a component-aware weighted metric over the nine-component embeddings, combined with provision and tag overlap. Facts/discussion section similarity is displayed for reference and does not contribute to the score.