Struct RecallConfig 

pub struct RecallConfig {
    pub bm25_weight: f32,
    pub vector_weight: f32,
    pub importance_weight: f32,
    pub graph_weight: f32,
    pub trust_weight: f32,
    pub corroboration_weight: f32,
    pub conformal_alpha: Option<f32>,
    pub cache_embeddings: bool,
    pub max_cache_entries: usize,
    pub writer_heap_size: usize,
    pub promotion_on_read: bool,
    pub association_rerank: bool,
}

Configuration for the recall engine.

Fields

bm25_weight: f32

Weight for BM25 text score (default 0.5).

vector_weight: f32

Weight for vector cosine similarity (default 0.3).

importance_weight: f32

Weight for memory importance (default 0.2).

graph_weight: f32

Post-fusion graph-traversal boost multiplier (V8 S6, default 0.0 = OFF — evidence-gated like trust weighting). NOT part of the normalized fusion sum: when > 0, the top fused seeds are expanded one hop through association edges, neighbors are injected or boosted by seed_score * edge_weight * graph_weight, and results carry the contribution in RecallResult::graph_score.

trust_weight: f32

Post-fusion trust multiplier (V8 S8, default 0.0 = OFF).

Like the graph weight, deliberately OUTSIDE the normalized fusion sum: when > 0, every fused score is scaled by 1 + weight * (source_trust − 0.7) (user-confirmed ranks up, tool-ingested 0.7 unchanged, low trust down), and the factor is disclosed per-result in RecallResult::trust_factor.

corroboration_weight: f32

Post-fusion corroboration multiplier (bridging counter, default 0.0 = OFF). Like trust, deliberately OUTSIDE the normalized fusion sum: when > 0, every fused score is scaled by the saturating corroboration_boost over the distinct-session count, and the count is disclosed per-result in RecallResult::corroboration.

conformal_alpha: Option<f32>

Conformal-set miscoverage level (V8 S8, default None = OFF). When set in (0, 1), fused results are graded against a calibrated prediction set and the search carries a ConformalSetInfo disclosure — active with a real threshold, or uncalibrated until feedback samples exist.

cache_embeddings: bool

Whether to cache embeddings (default true).

max_cache_entries: usize

Maximum cache entries (default 1000).

writer_heap_size: usize

Tantivy IndexWriter heap size in bytes (default 50MB). Note: writer is now owned by SearchEngine; this field is kept for API compatibility.

promotion_on_read: bool

Post-fusion promotion-on-read (S5 promotion-on-read, default false = OFF). When enabled (via WM_PROMOTION_ON_READ=1 or config), top recall hits trigger Memory::recall(), updating accessed_at, access_count, recall_count, and applying Hebbian neuro_score boosting + novelty decay.

association_rerank: bool

Association-weighted recall reranking (S10, default false = OFF). When enabled (via WM_ASSOCIATION_RERANK=1 or config), candidate recall scores receive an association connectivity boost based on active edges.

Implementations

impl RecallConfig

pub fn from_env() -> Self

Create config from environment variables.

Weights are clamped to [0.0, 1.0] and normalized to sum to 1.0. NaN and Infinity values are rejected (default is kept).

pub fn weights_normalized(&self) -> bool

Validate that weights sum to approximately 1.0.

Trait Implementations

impl Clone for RecallConfig

fn clone(&self) -> Self

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for RecallConfig

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for RecallConfig

fn default() -> Self

Returns the “default value” for a type.