Struct ProbabilityRankingConfig 

pub struct ProbabilityRankingConfig {
    pub temperature: f32,
    pub offline: bool,
    pub markov_from: Option<String>,
    pub counted_utility: bool,
    pub min_transition_utility: Option<f32>,
    pub min_transition_count: Option<usize>,
    pub similarity_threshold: Option<f32>,
}

Ranking controls shared by Bayesian and Markov-style helpers.

The optional fields below port the decision-policy hyperparameters from Kolonin’s “Interpretable Experiential Learning” (arXiv:2605.00940). Their defaults (counted_utility = false, both thresholds None) reproduce the paper’s recommended CU = False, TU = 0, TC = 1 baseline, which is exactly the additive behavior this module shipped before they were added, so existing callers are unaffected unless they opt in.

Fields

temperature: f32

offline: bool

markov_from: Option<String>

counted_utility: bool

Counted-utility policy (the paper’s CU). When true, a candidate’s learned utility is scaled by its evidence count (U becomes U * C), so a frequently confirmed transition outranks a rarely seen one of equal per-observation weight. When false the ranking uses the accumulated utility directly (argmax(U)).

min_transition_utility: Option<f32>

Minimum accumulated transition utility (the paper’s TU). A candidate whose evidence weight is below this threshold has its learned evidence withheld and falls back to its structural prior. None disables the gate.

min_transition_count: Option<usize>

Minimum evidence count (the paper’s TC). A candidate observed fewer times than this threshold has its learned evidence withheld and falls back to its structural prior. None disables the gate.

similarity_threshold: Option<f32>

Similarity threshold for the inexact-state fallback (the paper’s SS). When a candidate has no exact evidence of its own, the ranker reuses the nearest stored target whose symbolic cosine similarity to the candidate is at least this threshold, scaling the borrowed utility by the similarity. None disables the fallback, so only exact evidence counts.

Implementations

impl ProbabilityRankingConfig

pub const fn with_decision_policy( self, policy: ProbabilityDecisionPolicy, ) -> Self

Overlay the paper’s decision-policy hyperparameters (CU/TU/TC/SS) onto this config, leaving the deterministic transport knobs (temperature, offline, markov_from) untouched. This is the seam every call site uses to honour a centrally configured ProbabilityDecisionPolicy without re-spelling each field.

pub const fn decision_policy(&self) -> ProbabilityDecisionPolicy

Extract the decision-policy hyperparameters from this config.

Trait Implementations

impl Clone for ProbabilityRankingConfig

fn clone(&self) -> ProbabilityRankingConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ProbabilityRankingConfig

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

Formats the value using the given formatter.

impl Default for ProbabilityRankingConfig

fn default() -> ProbabilityRankingConfig

Returns the “default value” for a type.

impl PartialEq for ProbabilityRankingConfig

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

Equality operator ==.

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

Inequality operator !=.

impl StructuralPartialEq for ProbabilityRankingConfig