---
id: 43cb893a-c577-407e-a41c-51d7f81e80f6
type: solution
title: "Retroactive edge creation: edge-proposer.py script and 20 curated edges"
tags: [cognitive-memory, edges, automation, python, graph]
importance: 0.7
confidence: 0.8
created: "2026-02-20T05:39:47.119778+00:00"
updated: "2026-02-20T05:40:00.105088+00:00"
relations:
  - target: c3f6fbc2-04d0-41ef-87c2-8685d8769ac3
    type: CAUSES
    direction: outgoing
    strength: 0.8
    edge_id: 6cecf4c4-e3c4-44f7-bba8-ad69a10f9c86
---

# Retroactive Edge Creation for Cognitive Memory

## Problem

Memory graph had only 34 edges across 473 memories — very sparse, limiting graph traversal usefulness.

## Solution

Built `~/.claude/skills/cognitive-memory/scripts/edge-proposer.py` that analyzes all active/fading memories and proposes candidate edges.

### Scoring Algorithm

Three signals combined:
- **Type heuristics** (40%): e.g., fix + problem → SOLVES relationship
- **Tag Jaccard similarity** (30%): shared tags indicate related topics
- **Content keyword overlap** (30%): common terms in body text

### Results

- First run on 473 memories produced 5,186 candidates
- Manually reviewed top 80 candidates
- Selected 20 high-quality edges across 8 project clusters:
  - Gitea CI/CD: 3 edges
  - Paper Dynasty: 3 edges
  - Major Domo: 4 edges
  - Database patterns: 2 edges
  - Cognitive memory: 1 edge
  - Proxmox: 1 edge
  - Vagabond RPG: 3 edges
  - OpenSCAD: 3 edges
- Grew edge count from 34 to 54

### Noise to Skip

Session-log FOLLOWS pairs (auto-generated episode entries) were identified as low-value noise and excluded from selection.