[Atlas] Analysis-to-KG impact scoring done analysis:8 coding:7 reasoning:7

[Atlas] Analysis-to-KG impact scoring

Goal

Compute world model impact per analysis: new edges contributed, new entities introduced, centrality change, gaps resolved. Store as kg_impact_score on analyses table. Acceptance: all analyses scored; recalculated on KG changes; visible and sortable on /analyses/.

Acceptance Criteria

Approach

Read relevant source files to understand current state

Plan implementation based on existing architecture

Implement changes

Test affected pages with curl

Commit with descriptive message and push

Work Log

2026-04-25 - Implementation complete

• Added kg_impact_score column to analyses table (REAL, default 0.0) via migration 126
• Created calculate_kg_impact() function in post_process.py:

- 1 pt per edge contributed (from knowledge_edges where analysis_id matches)
- 2 pts per unique entity introduced (source_id or target_id from those edges)
- 0.5 pts per high-degree node connection (>5 connections)
- 10 pts bonus if analysis addressed a gap (status resolved/partially_addressed)

• Created update_all_kg_impact_scores() function to recalculate all scores
• Updated regenerate_dashboard() to:

- Include kg_impact_score in SELECT query (COALESCE to 0 for nulls)
- Sort analyses by kg_impact_score DESC, then by created_at DESC
- Display score as purple "KG: X.X" tag on each analysis card

• All 419 analyses scored (173 with nonzero scores, range: 0.0 to 651.0)
• Top scoring analysis: "SDA-2026-04-01-gap-20260401-225149" (651.0)
• Committed as 201bf5fa1 to branch orchestra/task/2382f69a-analysis-to-kg-impact-scoring

2026-04-01 - Starting implementation

• Assigned to slot 4
• Reading current codebase to understand analyses table, KG structure, and /analyses page
• Will implement kg_impact_score calculation and display

2026-04-01 - Implementation complete (prior attempt)

• Added kg_impact_score column to analyses table (REAL, default 0.0)
• Created calculate_kg_impact() function in post_process.py:

- 1 point per edge contributed
- 2 points per unique entity introduced
- 10 points bonus if analysis addressed a gap (status=partially_filled/resolved)
- 0.5 points per high-degree node connection (>5 connections)

• Created update_all_kg_impact_scores() function to recalculate all scores
• Updated regenerate_dashboard() to:

- Include kg_impact_score in SELECT query
- Sort analyses by kg_impact_score DESC
- Display score as purple tag on each analysis card

• All 21 analyses scored (range: 0.0 - 91.0)
• API endpoint /api/analyses automatically includes score (uses SELECT a.*)
• Tested: /analyses/ page loads correctly (200) with KG impact scores visible
• Top scoring analysis: "Selective vulnerability of entorhinal cortex layer II neurons in AD" (91.0)