Skip to content

Learning Graph Quality Metrics Report

Overview

  • Total Concepts: 475
  • Foundational Concepts (no prerequisites, other concepts depend on them): 1
  • Terminal Nodes (nothing depends on them, but have prerequisites): 244
  • Orphaned Nodes (completely disconnected, no edges): 0
  • Concepts with Dependencies: 474
  • Average Dependencies per Concept: 1.47

Graph Structure Validation

  • Valid DAG Structure: ✅ Yes
  • Self-Dependencies: None detected ✅
  • Cycles Detected: 0

Foundational Concepts

These concepts have no prerequisites:

  • 1: Generative AI

Dependency Chain Analysis

  • Maximum Dependency Chain Length: 12

Longest Learning Path:

  1. Generative AI (ID: 1)
  2. Large Language Model (ID: 2)
  3. Anthropic API (ID: 76)
  4. Structured Logging (ID: 196)
  5. Log Line (ID: 198)
  6. JSON Log Format (ID: 199)
  7. Log Field (ID: 200)
  8. Required Log Field (ID: 201)
  9. Latency Field (ID: 208)
  10. Latency Metric (ID: 273)
  11. Cost-Latency Tradeoff (ID: 66)
  12. Pareto Frontier (ID: 67)

Terminal Nodes Analysis

Terminal nodes are concepts that nothing else depends on but have prerequisites. They represent natural endpoints of learning paths — culminating or specialized concepts.

  • Total Terminal Nodes: 244 (51.4% of all concepts)
  • Healthy Range: 5-40% of total concepts

Concepts at the end of learning paths:

  • 3: Foundation Model
  • 14: SentencePiece
  • 15: Vocabulary Size
  • 20: User Message
  • 21: Assistant Message
  • 27: Temperature
  • 29: Logprobs
  • 32: BPE Merge Rules
  • 36: Token-To-Char Ratio
  • 40: End-Of-Sequence Token
  • 41: Beginning-Of-Sequence Token
  • 42: Padding Token
  • 43: Token Boundary
  • 44: Pre-Tokenization
  • 45: Cross-Vendor Tokenizer Drift
  • 46: Token Counting API
  • 48: Token Count Caching
  • 49: Code Tokenization
  • 50: Multilingual Tokenization
  • 55: Output Premium

...and 224 more

Orphaned Nodes Analysis

Orphaned nodes are completely disconnected concepts with no inbound AND no outbound edges. These indicate a quality problem — every concept should connect to the graph.

  • Total Orphaned Nodes: 0

✅ No orphaned nodes detected. All concepts are connected to the graph.

Connected Components

  • Number of Connected Components: 1

✅ All concepts are connected in a single graph.

Indegree Analysis

Top 10 concepts that are prerequisites for the most other concepts:

Rank Concept ID Concept Label Indegree
1 126 Google Gemini API 18
2 12 Tokenizer 17
3 11 Token Count 14
4 101 OpenAI API 14
5 196 Structured Logging 14
6 401 Agent Budget Policy 14
7 76 Anthropic API 13
8 261 A/B Testing 13
9 311 Prompt Caching 13
10 56 Unit Economics 12

Outdegree Distribution

Dependencies Number of Concepts
0 1
1 265
2 195
3 12
4 2

Recommendations

  • ℹ️ High terminal node percentage (51.4%): Consider if some terminal concepts should be prerequisites for advanced concepts
  • ✅ DAG structure verified: Graph supports valid learning progressions
  • ℹ️ Consider adding cross-dependencies: More connections could create richer learning pathways

Report generated by learning-graph-reports/analyze_graph.py