Struct BridgeConfig 

pub struct BridgeConfig {
    pub threshold_base: f64,
    pub threshold_evr_penalty: f64,
    pub overlap_artifact_territorial: f64,
    pub balanced_affinity_quantile: f64,
    pub min_evr_for_classification: f64,
}

Parameters controlling bridge detection and classification.

Fields

threshold_base: f64

Constant term in the EVR-adaptive bridge threshold threshold = threshold_base + (1 − evr)² · threshold_evr_penalty.

threshold_evr_penalty: f64

EVR-penalty coefficient in the bridge threshold formula.

overlap_artifact_territorial: f64

Percentile of the observed territorial factor distribution below which a bridge is classified as OverlapArtifact rather than Genuine or Weak. 0.3 = the bottom 30 % of bridge pairs by territorial separation are labeled artifacts. Expressed as a percentile so that dense corpora (where all exclusivities collapse toward zero) do not classify every bridge as an artifact.

balanced_affinity_quantile: f64

Quantile of the home-affinity distribution that sets the genuine-bridge floor. For each member item, “home affinity” is the cosine similarity between the item’s embedding and its own category’s centroid. A bridge is classified Genuine when min(affinity_to_source, affinity_to_target) exceeds the quantile-q of those home affinities; otherwise Weak.

Why a quantile and not an absolute cosine: home affinity scale varies with the projection layout. After stratified PCA spreads imbalanced corpora, home affinities can drop into the 0.3–0.6 band where a fixed 0.5 cosine floor labels almost every cross- domain item Weak. A quantile-based floor adapts to the corpus’s own affinity scale: tight corpora get a strict floor, spread ones get a permissive one, without per-corpus tuning.

Smaller q = stricter (only bridges matching the strongest home affinities qualify). Larger q = more permissive. Default 0.25: a bridge is Genuine if it has at least as much affinity to both sides as the bottom-25% of items have to their own home category.

min_evr_for_classification: f64

EVR below which bridge classification is unreliable. When the outer projection’s EVR is below this threshold, all bridges are labeled Weak (honest uncertainty) rather than attempting territorial-factor-based classification — which collapses to 100% OverlapArtifact when caps overlap everywhere on a low-EVR projection, flattening the tuner landscape. Default 0.20.

Implementations

impl BridgeConfig

pub fn evr_adaptive_threshold(&self, evr: f64) -> f64

EVR-adaptive bridge threshold.

Higher EVR → looser threshold (projection is more trustworthy). At EVR=0.19: 0.5 + 0.81² × 0.4 = 0.76 (strict). At EVR=0.90: 0.5 + 0.01 × 0.4 = 0.50 (essentially unchanged).

Trait Implementations

impl Clone for BridgeConfig

fn clone(&self) -> BridgeConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for BridgeConfig

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

Formats the value using the given formatter.

impl Default for BridgeConfig

fn default() -> Self

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for BridgeConfig

where BridgeConfig: 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 BridgeConfig

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

where __S: Serializer,

Serialize this value into the given Serde serializer.