pub struct RankHooks {Show 14 fields
pub seed_providers: Vec<SeedProvider>,
pub features: Vec<RankFeatureFn>,
pub rerankers: Vec<RerankerFn>,
pub edge_weights: Vec<EdgeWeightFn>,
pub similarities: Vec<SimilarityFn>,
pub coverage: Vec<CoverageProvider>,
pub candidates: Vec<CandidateProvider>,
pub rank_edges: Vec<RankEdgeProvider>,
pub rank_nodes: Vec<RankNodeProvider>,
pub criticality: Vec<CriticalityProvider>,
pub novelty: Vec<NoveltyProvider>,
pub risk: Vec<RiskProvider>,
pub semantic: Vec<SemanticProvider>,
pub profiles: BTreeMap<String, BlendWeights>,
}Fields§
§seed_providers: Vec<SeedProvider>§features: Vec<RankFeatureFn>§rerankers: Vec<RerankerFn>§edge_weights: Vec<EdgeWeightFn>§similarities: Vec<SimilarityFn>Pairwise item similarity for MMR diversification. First provider returning a nonzero value wins (deterministic chain order); the built-in default (same-group => 1.0) runs last.
coverage: Vec<CoverageProvider>Extra required-coverage providers (§124 item 26): contributed
symbol ids union with the engine required_ids base. Recorded in
reasons as required-by:<n>.
candidates: Vec<CandidateProvider>Extra candidate providers (spec 18): merged with the lexical base by canonical id, max score wins. Deterministic chain order.
rank_edges: Vec<RankEdgeProvider>Extra rank-time edges (§48): merged into diffusion, never into the canonical graph. Deterministic chain order.
rank_nodes: Vec<RankNodeProvider>Extra rank-universe nodes (§124 item 17): merged before edge indexing so provider edges can attach to provider nodes. Deterministic chain order.
criticality: Vec<CriticalityProvider>Criticality overrides (§53): first Some in chain order wins.
novelty: Vec<NoveltyProvider>Novelty overrides (§53): first Some in chain order wins.
risk: Vec<RiskProvider>Change-risk overrides (§53): first Some in chain order wins.
semantic: Vec<SemanticProvider>Semantic-score overrides (§53): first Some in chain order wins.
profiles: BTreeMap<String, BlendWeights>Named blend profiles: profile name -> per-feature weight overrides for the linear blend (feature keys: task_ppr, global_ppr, lexical, semantic, confidence, criticality, change_risk, novelty). Missing keys keep default weights. Applied inside the same linear math; recorded in reasons.