Crate uqa_scoring 

Relevance scoring: BM25, Lucene-style query-level Bayesian BM25, and probabilistic Boolean and log-odds combinators.

Re-exports

pub use bayesian::LegacyCompositePriorTransform;

pub use bayesian_bm25::BayesianBM25Params;

pub use bayesian_bm25::BayesianBM25Scorer;

pub use bayesian_estimator::UnsupervisedBm25ScoreEstimator;

pub use bm25::BM25Params;

pub use bm25::BM25Scorer;

pub use calibration::CalibrationMetrics;

pub use calibration::CalibrationReport;

pub use calibration::ReliabilityBin;

pub use calibration::VectorProbabilityTransform;

pub use calibration_validation::BinaryDecisionMetrics;

pub use calibration_validation::BootstrapConfig;

pub use calibration_validation::ConfidenceInterval;

pub use calibration_validation::HeldOutCalibrationGate;

pub use calibration_validation::HeldOutCalibrationReport;

pub use calibration_validation::ThresholdTransferReport;

pub use error::ScoringError;

pub use error::ScoringResult;

pub use external_prior::authority_prior;

pub use external_prior::recency_prior;

pub use external_prior::ExternalPriorScorer;

pub use external_prior::PriorFn;

pub use fusion_wand::ConfidenceScaledPoolWANDScorer;

pub use fusion_wand::TightenedConfidenceScaledPoolWANDScorer;

pub use metrics::average_precision_at_k;

pub use metrics::dcg_at_k;

pub use metrics::mean_average_precision_at_k;

pub use metrics::ndcg_at_k;

pub use multi_field::FieldConfig;

pub use multi_field::MultiFieldBayesianScorer;

pub use parameter_learner::ParameterLearner;

pub use prob::confidence_scaled_log_odds_pool;

pub use prob::cosine_to_probability;

pub use prob::logit;

pub use prob::prob_and;

pub use prob::prob_not;

pub use prob::prob_or;

pub use prob::sigmoid;

pub use prob::PROB_EPSILON;

pub use score_domain::EvidenceLogit;

pub use score_domain::PosteriorProbability;

pub use score_domain::PriorLogit;

pub use score_domain::RawBm25Score;

pub use scorer::Scorer;

pub use vector_calibration::VectorCalibrationModel;

pub use vector_calibration::VectorCalibrationProvenance;

pub use vector_calibration::VectorCalibrationStabilityReport;

pub use vector_calibration::VectorCalibrationTarget;

pub use vector_calibration::VECTOR_CALIBRATION_MODEL_SCHEMA_VERSION;

pub use vector_score::VectorScorer;

pub use wand::AdaptiveWANDScorer;

pub use wand::BlockMaxWANDScorer;

pub use wand::BoundTightnessAnalyzer;

pub use wand::CursorBlockMaxWANDScorer;

pub use wand::CursorWANDQuery;

pub use wand::CursorWANDScorer;

pub use wand::WANDQuery;

pub use wand::WANDResult;

pub use wand::WANDScorer;

pub use wand::WANDStats;

Modules

bayesian

Legacy composite-prior ranking transform.

bayesian_bm25

Query-level Bayesian calibration for BM25 scores.

bayesian_estimator

Corpus-driven estimation of an unsupervised BM25 score transform.

bm25

Standard BM25 scorer (Definition 3.2.1, Paper 3).

calibration

Calibration diagnostics (Paper 3 Section 11.3, Paper 5 Section 8.3).

calibration_validation

Held-out calibration validation with deterministic uncertainty estimates.

error

Errors raised when scoring inputs or learned parameters violate their contract.

external_prior

Bayesian BM25 with an external prior (Section 12.2 #6, Paper 3).

fusion_wand

WAND-style top-k pruning for the confidence-scaled log-odds ranking pool.

metrics

Information-retrieval relevance metrics: DCG, NDCG, average precision, and mean average precision. Inputs are graded relevance judgments aligned to a ranked list of doc ids; outputs are scalar quality scores in [0, 1] (NDCG) or [0, +inf) (DCG).

multi_field

Per-field Bayesian BM25 scoring with weighted log-odds fusion across fields (Section 12.2 #1, Paper 3).

parameter_learner

Supervised fitting for query-level Bayesian BM25 calibration.

prob

Numerically stable sigmoid / logit and probabilistic AND/OR/NOT in log space. The explicitly named confidence_scaled_log_odds_pool implements the robust n^alpha ranking heuristic; exact signed single-prior evidence fusion lives in uqa-fusion::BayesianEvidenceFusion.

score_domain

Semantic score-domain types.

scorer

Query-scoring contract shared by BM25 and Bayesian BM25.

vector_calibration

Persistable vector-calibration models and their compatibility contract.

vector_score

Vector similarity to probability conversion.

wand

WAND and Block-Max WAND top-k scorers (Section 6, Paper 3).