Enum RankingStrategy 

#[non_exhaustive]
pub enum RankingStrategy {
    Hybrid {
        alpha: f32,
        decay: DecayFn,
    },
    Blended {
        weights: BlendWeights,
        decay: DecayFn,
    },
}

How Client::query orders the candidate set.

Reach for Self::Hybrid when relevance and recency are the only signals that matter (it ignores confidence and category); reach for Self::Blended to also reward high-confidence facts and preferred categories — the typical choice once the categorize/confidence pipeline is populated. Blended is a strict superset, so Hybrid is just the w_conf = 0, no-category special case. For raw nearest-neighbor hits with no re-ranking at all, use Client::search instead of query.

Constructing a variant explicitly is the caller’s opt-in to that specific behavior — those parameter values become part of the stability contract for the caller’s code. The default behavior (when no strategy is passed) is allowed to drift pre-1.0.

Variants (Non-exhaustive)

Non-exhaustive enums could have additional variants added in future. Therefore, when matching against variants of non-exhaustive enums, an extra wildcard arm must be added to account for any future variants.

Hybrid

Blend of cosine similarity and recency.

score = alpha * cosine + (1 - alpha) * decay(age). alpha in [0.0, 1.0]; 0.0 is pure recency, 1.0 is pure cosine. Age is computed against event_at when present, falling back to created_at.

Fields

alpha: f32

Weight on cosine; 1 - alpha weights recency.

decay: DecayFn

Decay shape applied to the memory’s age.

Blended

Linear blend of cosine, confidence, recency, and a category bonus.

score = w_cos·cosine + w_conf·(confidence/100) + w_rec·decay(age), plus category_bonus when the memory’s category is preferred. The superset of Self::Hybrid (which is the w_conf = 0, no-category case). Confidence is normalized from 0-100 to [0, 1] before weighting so no single signal dominates by scale. Pure math + indexed lookups — no inference in the read path.

Fields

weights: BlendWeights

Signal weights and the optional category preference.

decay: DecayFn

Decay shape applied to the memory’s age.

Implementations

impl RankingStrategy

pub fn default_hybrid() -> Self

The default Hybrid configuration query() uses when no strategy is passed. Parameter values may drift pre-1.0.

pub fn blended(weights: BlendWeights) -> Self

A Self::Blended strategy with the given weight preset and the default exponential recency decay.

Convenience over constructing Blended { weights, decay } by hand when the default half-life is fine: RankingStrategy::blended(BlendWeights::trust_first()).

Trait Implementations

impl Clone for RankingStrategy

fn clone(&self) -> RankingStrategy

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for RankingStrategy

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

Formats the value using the given formatter.

impl PartialEq for RankingStrategy

fn eq(&self, other: &RankingStrategy) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl StructuralPartialEq for RankingStrategy