benchmark_rag_comparison.json

{
  "benchmark": "STYX vs RAG vs GraphRAG Comparison",
  "patent": "Pending #63/975,190",
  "dataset": "facebook/react",
  "document_count": 300,
  "timestamp": "2026-02-05T02:20:24Z",
  "query": "How do I handle concurrent state updates in React?",
  "token_counts": {
    "full_context": 150720,
    "basic_rag": 9510,
    "rag_reranking": 9446,
    "graphrag": 16731,
    "styx_on_full": 9141,
    "styx_on_rag": 1312,
    "styx_on_reranking": 678,
    "styx_on_graphrag": 2002
  },
  "reductions": {
    "rag_vs_full": "93.7%",
    "styx_vs_full": "93.9%",
    "styx_vs_rag": "86.2%",
    "styx_vs_graphrag": "88.0%",
    "styx_on_rag_vs_full": "99.1%",
    "styx_on_graphrag_vs_full": "98.7%"
  },
  "key_finding": "STYX provides 86-88% additional reduction on top of RAG/GraphRAG",
  "methodology": {
    "rag_method": "Semantic similarity retrieval (top-k chunks)",
    "reranking_method": "Cross-encoder reranking after retrieval",
    "graphrag_method": "Knowledge graph entity extraction + community summaries",
    "styx_method": "Structural intelligence extraction (decisions, constraints, tensions, anti-patterns)"
  }
}