Skip to content

Config reference: fusion

  • Enterprise tuning surface

    Defaults + constraints are rendered directly from Pydantic.

  • Env keys when available

    Many fields have an env-style alias (from TriBridConfig.to_flat_dict()).

  • Tooltip-level guidance

    If a matching glossary entry exists, you’ll see deeper tuning notes.

Config reference Config API & workflow Glossary

Total parameters: 6

Group index
  • (root)

(root)

JSON key Env key(s) Type Default Constraints Summary
fusion.graph_weight FUSION_GRAPH_WEIGHT float 0.3 ≥ 0.0, ≤ 1.0 Weight for graph search results (Neo4j)
fusion.method FUSION_METHOD Literal["rrf", "weighted"] "rrf" allowed="rrf", "weighted" Fusion method: 'rrf' (Reciprocal Rank Fusion) or 'weighted' (score-based)
fusion.normalize_scores FUSION_NORMALIZE_SCORES bool true — Normalize scores to [0,1] before fusion
fusion.rrf_k FUSION_RRF_K int 60 ≥ 1, ≤ 200 RRF smoothing constant (higher = more weight to top ranks)
fusion.sparse_weight FUSION_SPARSE_WEIGHT float 0.3 ≥ 0.0, ≤ 1.0 Weight for sparse BM25/FTS search results
fusion.vector_weight FUSION_VECTOR_WEIGHT float 0.4 ≥ 0.0, ≤ 1.0 Weight for vector search results (pgvector)

Details (glossary)

fusion.graph_weight (FUSION_GRAPH_WEIGHT) — Graph Weight

Category: general

Weight assigned to graph (Neo4j) search results in weighted fusion mode. Higher values (0.4-0.6) favor structural relationships, lower values (0.2-0.3) reduce graph influence. Weights must sum to ~1.0 with vector and sparse weights. Recommended: 0.3 for balanced tri-brid retrieval.

Sweet spot: 0.3 for balanced systems. Use 0.4-0.5 when graph relationships are critical (e.g., finding code that calls or imports specific functions). Use 0.2 when vector and sparse search are more important.

• Range: 0.0-1.0 (must sum with vector + sparse ≈ 1.0) • Vector/sparse-focused: 0.2 (lower graph weight) • Balanced: 0.3 (recommended) • Graph-focused: 0.4-0.5 (higher graph weight) • Effect: Higher = more weight to graph search results • Symptom too high: Graph matches dominate, other modalities buried • Symptom too low: Graph relationships undervalued

Badges: - Weighted Mode

Links: - Neo4j GraphRAG - Graph Traversal - Weighted Fusion

fusion.method (FUSION_METHOD) — Fusion Method

Category: general

Method for combining results from vector, sparse, and graph search: "rrf" (Reciprocal Rank Fusion) or "weighted" (score-based weighted sum). RRF combines ranking positions without score normalization, making it robust to different score scales. Weighted fusion requires normalized scores and allows fine-grained control over modality weights.

Sweet spot: "rrf" for most use cases. RRF is simpler, more robust, and doesn't require score normalization. Use "weighted" when you need precise control over modality weights or when score distributions are well-calibrated.

• RRF: Position-based fusion, robust to score scales, simpler • Weighted: Score-based fusion, requires normalization, more control • Effect: Determines how tri-brid results are combined • Symptom wrong method: Suboptimal result ranking

Badges: - Core Setting

Links: - Reciprocal Rank Fusion - RRF Paper - Data Fusion

fusion.normalize_scores (FUSION_NORMALIZE_SCORES) — Normalize Scores

Category: general

Normalize scores from vector, sparse, and graph search to [0,1] range before fusion. This ensures scores from different modalities are comparable when using weighted fusion. When disabled, raw scores are used directly (may cause one modality to dominate). Recommended: enabled for weighted fusion, not needed for RRF.

Sweet spot: enabled for weighted fusion mode. Normalization prevents one modality from dominating due to different score scales. For RRF mode, normalization is unnecessary since RRF uses ranking positions, not scores.

• Enabled: Scores normalized to [0,1], comparable across modalities • Disabled: Raw scores used, may cause modality imbalance • Effect: Controls score normalization before weighted fusion • Symptom if disabled: One modality may dominate due to score scale differences

Badges: - Weighted Mode

Links: - Score Normalization - Normalization Methods - Data Fusion

fusion.rrf_k (FUSION_RRF_K) — RRF k Parameter

Category: general

RRF constant for tri-brid fusion when method is rrf. Formula: sum(1/(k+rank)). Lower k emphasizes top ranks more; higher k distributes weight more evenly. Default: 60. Range: 1-200.

Badges: - RRF Mode

Links: - RRF Original Paper - RRF Research - Reciprocal Rank Fusion

fusion.sparse_weight (FUSION_SPARSE_WEIGHT) — Sparse Weight

Category: general

Weight assigned to sparse (BM25) search results in weighted fusion mode. Higher values (0.4-0.6) favor keyword matches, lower values (0.2-0.3) reduce keyword influence. Weights must sum to ~1.0 with vector and graph weights. Recommended: 0.3 for balanced tri-brid retrieval.

Sweet spot: 0.3 for balanced systems. Use 0.4-0.5 when exact keyword matching is critical (e.g., finding specific function names or error codes). Use 0.2 when semantic and graph search are more important.

• Range: 0.0-1.0 (must sum with vector + graph ≈ 1.0) • Semantic-focused: 0.2 (lower keyword weight) • Balanced: 0.3 (recommended) • Keyword-focused: 0.4-0.5 (higher keyword weight) • Effect: Higher = more weight to sparse search results • Symptom too high: Keyword matches dominate, semantic matches buried • Symptom too low: Keyword matches undervalued

Badges: - Weighted Mode

Links: - BM25 Algorithm - Hybrid Search - Weighted Fusion

fusion.vector_weight (FUSION_VECTOR_WEIGHT) — Vector Weight

Category: general

Weight assigned to vector (pgvector) search results in weighted fusion mode. Higher values (0.5-0.7) favor semantic matches, lower values (0.2-0.4) reduce semantic influence. Weights must sum to ~1.0 with sparse and graph weights. Recommended: 0.4 for balanced tri-brid retrieval.

Sweet spot: 0.4 for balanced systems. Use 0.5-0.6 when semantic matching is critical (e.g., finding conceptually similar code). Use 0.2-0.3 when keyword matching is more important than semantics.

• Range: 0.0-1.0 (must sum with sparse + graph ≈ 1.0) • Keyword-focused: 0.2-0.3 (lower semantic weight) • Balanced: 0.4 (recommended) • Semantic-focused: 0.5-0.6 (higher semantic weight) • Effect: Higher = more weight to vector search results • Symptom too high: Semantic matches dominate, keyword matches buried • Symptom too low: Semantic matches undervalued

Badges: - Weighted Mode

Links: - Hybrid Search - Weighted Fusion - Fusion Strategies