Struct ProbabilityStore 

pub struct ProbabilityStore { /* private fields */ }

Append-only probability evidence store.

Implementations

impl ProbabilityStore

pub const fn new() -> Self

pub const fn from_records(records: Vec<ProbabilityEvidence>) -> Self

pub fn record(&mut self, evidence: ProbabilityEvidence) -> String

pub fn update( &mut self, target: impl Into<String>, observation: impl Into<String>, weight: f32, provenance: impl Into<String>, recorded_at: impl Into<String>, ) -> String

pub fn reinforce_transition_path<S: AsRef<str>>( &mut self, path: &[S], reward: f32, provenance: impl Into<String>, recorded_at: impl Into<String>, ) -> Vec<String>

Reinforce a whole episode’s trajectory in one shot — the deterministic, append-only counterpart of the paper’s global feedback (episode-wide one-shot update) from arXiv:2605.00940.

Given an ordered path of visited states [s0, s1, ..., sn], this appends one ProbabilityModel::MarkovTransition record per adjacent pair (s_i -> s_{i+1}), each carrying the shared episode reward as its utility U and the same provenance/recorded_at stamp, so the entire episode is reinforced together rather than transition by transition. The recorded evidence is then visible to Self::target_weight / Self::target_evidence_count under the matching markov_from state, exactly like any other transition observation.

Returns the ids of the appended records in path order. A path with fewer than two states has no transitions, so it records nothing and returns an empty vector.

pub fn records(&self) -> &[ProbabilityEvidence]

pub fn target_weight( &self, target: &str, offline: bool, markov_from: Option<&str>, ) -> f32

pub fn target_evidence_count( &self, target: &str, offline: bool, markov_from: Option<&str>, ) -> usize

Count the number of append-only observations that support target.

This is the symbolic analogue of the evidence count C from Kolonin’s “Interpretable Experiential Learning” (arXiv:2605.00940): every recorded observation is one unit of evidence for a transition/answer, kept separate from the accumulated utility (target_weight) so that a rarely seen high-weight transition can be told apart from a frequently confirmed one. The same offline and Markov-state filters as Self::target_weight apply, so utility and count always describe the same evidence subset.

pub fn nearest_similar_evidence( &self, target: &str, offline: bool, markov_from: Option<&str>, threshold: f32, ) -> Option<SimilarEvidence>

Reuse the nearest stored target’s evidence when target has none of its own — the symbolic counterpart of the paper’s cosine-similarity SS fallback over stored situations.

Among the distinct targets that carry usable evidence under the same offline/Markov filters (excluding target itself), this returns the one whose symbolic_cosine_similarity to target is highest and at least threshold. Ties are broken by target name so the choice is deterministic. Returns None when nothing clears the threshold.

pub fn to_links_notation(&self) -> String

pub fn replay_into_event_log(&self, log: &mut EventLog, offline: bool) -> usize

pub fn append_to_link_store<S: LinkStore>( &self, store: &mut S, offline: bool, ) -> Result<usize, LinkStoreError>

Trait Implementations

impl Clone for ProbabilityStore

fn clone(&self) -> ProbabilityStore

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ProbabilityStore

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

Formats the value using the given formatter.

impl Default for ProbabilityStore

fn default() -> ProbabilityStore

Returns the “default value” for a type.

impl PartialEq for ProbabilityStore

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

Equality operator ==.

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

Inequality operator !=.

impl StructuralPartialEq for ProbabilityStore