Struct ScoringConfig 

pub struct ScoringConfig {

    pub weights: ScoringWeights,
    pub min_raw_relevance: f32,
    pub min_rrf_relevance: f32,
    pub baseline_relevance: f32,
    pub decay_cap: f32,
    pub max_recall_candidates: usize,
    pub default_recall_limit: usize,
    pub default_token_budget: usize,
    pub chars_per_token: usize,
    pub mmr_penalty: f32,
    pub mmr_prefix_len: usize,
    pub enable_supersedes_suppression: bool,
    pub enable_multilingual_routing: bool,
    pub multilingual_model: Option<String>,
    pub adjustments: Vec<ScoreAdjustment>,
}

Complete, tunable scoring configuration for the memory recall pipeline. DoS caps: max_recall_candidates≤500, default_token_budget≤16000, default_recall_limit≤200.

Fields

weights: ScoringWeights

min_raw_relevance: f32

Minimum raw cosine similarity to include a vector hit. Hits below this are excluded before RRF fusion (#2272). Default: 0.10.

min_rrf_relevance: f32

Minimum RRF score after fusion before normalization. Default: 0.0.

baseline_relevance: f32

Relevance floor for the min-max normalization band. Default: 0.15.

decay_cap: f32

Upper cap on per-entry decay_factor before temporal recency calculation. Default: 0.05.

max_recall_candidates: usize

Maximum search candidates to retrieve. Server-side cap: 500. Default: 200.

default_recall_limit: usize

Default result limit when caller doesn’t specify. Server-side cap: 200. Default: 10.

default_token_budget: usize

Default token budget (tokens). Server-side cap: 16000. Default: 4000.

chars_per_token: usize

Approximate characters per token (for token budget). Default: 4.

mmr_penalty: f32

Score penalty applied to results whose first mmr_prefix_len characters match an earlier result. Default: 0.1.

mmr_prefix_len: usize

Character prefix length used for MMR duplicate detection. Default: 100.

enable_supersedes_suppression: bool

When true, suppress memories whose properties.supersedes value matches the ID of another memory in the result set. Default: true.

enable_multilingual_routing: bool

When true and a multilingual embedding model is registered, route non-ASCII-script queries (CJK, Cyrillic, Arabic, accented-Latin, …) to it as the primary dense model. Default: true.

multilingual_model: Option<String>

Name of the multilingual embedding model to prefer for dense routing. When None, the handler checks registered model names for substrings “multilingual” or “paraphrase”. Default: None.

adjustments: Vec<ScoreAdjustment>

Score adjustments applied after the base formula. Default: episodic bonus, semantic age penalty, entity boost.

Implementations

impl ScoringConfig

pub fn apply_dos_caps(&mut self)

Clamp candidate, result, and token budgets to their server-side maximums.

Trait Implementations

impl Clone for ScoringConfig

fn clone(&self) -> ScoringConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ScoringConfig

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

Formats the value using the given formatter.

impl Default for ScoringConfig

fn default() -> Self

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for ScoringConfig

where ScoringConfig: Default,

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for ScoringConfig

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.