Evidence Traceability UI

[mathaix]

User Story

As a manager
I want to trace every insight back to its source interview quotes
So that I can verify claims and build trust in AI-generated findings

Acceptance Criteria

Evidence Chain Visualization

For any insight, recommendation, or finding:

• Click "View Evidence" to see supporting quotes
• Evidence panel shows:
  • Quote excerpt with context
  • Speaker name and role
  • Interview ID and timestamp
  • Confidence score
  • Link to full transcript at that point
• Multiple evidence sources for each insight
• Conflicting evidence shown separately

Evidence Quality Indicators

• Strong Evidence: 3+ sources, high agreement
• Moderate Evidence: 2 sources, general agreement
• Weak Evidence: 1 source or conflicting views
• Visual badges for evidence strength
• Confidence percentage based on evidence

Interactive Evidence Explorer

• Insights organized in tree structure
• Root: Recommendation
• Level 1: Supporting insights
• Level 2: Evidence nodes
• Level 3: Interview transcripts
• Expand/collapse tree navigation

Evidence Filtering

• Show/hide low-confidence evidence
• Filter by interview source
• Filter by speaker role
• Search evidence quotes

Evidence in Reports

• Footnote-style citations in exported reports
• Appendix with full evidence details
• Hyperlinks from claim to evidence (PDF)
• Evidence table with all sources

Neo4j Evidence Architecture

The evidence chain in Neo4j:

(:Recommendation {text: "Consolidate CRM platforms"})
  -[:DERIVED_FROM]->
(:Insight {text: "CRM Pro has higher satisfaction"})
  -[:BASED_ON]->
(:Evidence {quote: "NPS is at 72...", timestamp: "..."})
  -[:FROM_INTERVIEW]->
(:Interview {id: "int_008"})
  -[:SPOKEN_BY]->
(:Person {name: "Mike Johnson", role: "CS Manager"})

Evidence Validation

• Every recommendation must have ≥1 insight
• Every insight must have ≥1 evidence
• Every evidence must link to interview
• Synthesis blocked if evidence chain incomplete

Conflict Resolution Display

When evidence conflicts:

• Show all conflicting quotes
• Explain resolution strategy used
• Display expert-weighted consensus if applicable
• Flag for manager review if unresolved

Evidence Export

• Export evidence table as CSV
• Include all fields: insight, quote, speaker, interview, confidence
• Filter before export

Technical Notes

# Evidence chain query
async def get_evidence_chain(recommendation_id: str):
    query = """
    MATCH (r:Recommendation {id: $rec_id})
    MATCH (r)-[:DERIVED_FROM*]->(i:Insight)
    MATCH (i)-[:BASED_ON]->(e:Evidence)
    MATCH (e)-[:FROM_INTERVIEW]->(int:Interview)
    MATCH (int)-[:SPOKEN_BY]->(p:Person)
    RETURN r, i, e, int, p
    """
    result = await neo4j.run(query, rec_id=recommendation_id)
    return build_evidence_tree(result)

This is Clara's most important differentiator - full evidence traceability makes synthesis defensible.

Definition of Done

• Evidence chain query working
• Evidence panel UI implemented
• Evidence indicators visible
• Conflict display working
• Evidence validation enforced
• Export functionality working
• Code reviewed and merged

Requirement ID

RS-09

---

🤖 Generated with Claude Code