Crate rag_plusplus_core 

RAG++ Core Library

High-performance retrieval engine for memory-conditioned candidate selection.

Overview

RAG++ retrieves outcome-annotated trajectories and provides statistical priors for downstream conditioning. It does NOT train models - pure retrieval + statistics.

Invariants

• INV-001: Record immutability after creation
• INV-002: Welford numerical stability
• INV-003: WAL-before-buffer for durability
• INV-004: Index-record consistency

Modules

• types: Core data structures (MemoryRecord, QueryBundle, PriorBundle)
• stats: Outcome statistics with Welford’s algorithm
• error: Error types and Result aliases
• index: Vector index implementations (Flat, HNSW)
• filter: Metadata filter expressions and evaluation
• store: Record storage (in-memory, persistent)
• wal: Write-ahead log for durability
• buffer: Write buffering for batched operations
• retrieval: Query execution and reranking
• cache: Query result caching
• observability: Metrics and tracing
• eval: Evaluation metrics and benchmarking
• api: Public API contracts and traits
• trajectory: Trajectory memory algorithms (DAG traversal, phase inference, salience)
• distance: Vector distance operations with SIMD and batch support

Re-exports

pub use buffer::BufferStats;

pub use buffer::WriteBuffer;

pub use buffer::WriteBufferConfig;

pub use cache::CacheConfig;

pub use cache::CacheKey;

pub use cache::CacheStats;

pub use cache::QueryCache;

pub use error::Error;

pub use error::Result;

pub use filter::CompiledFilter;

pub use filter::FilterEvaluator;

pub use filter::FilterExpr;

pub use filter::FilterOp;

pub use filter::FilterValue;

pub use index::DistanceType;

pub use index::FlatIndex;

pub use index::FusedResult;

pub use index::FusionConfig;

pub use index::FusionStrategy;

pub use index::HNSWConfig;

pub use index::HNSWIndex;

pub use index::IndexConfig;

pub use index::IndexInfo;

pub use index::IndexRegistry;

pub use index::MultiIndexResult;

pub use index::MultiIndexResults;

pub use index::ParallelSearchConfig;

pub use index::ParallelSearcher;

pub use index::ResultsAggregator;

pub use index::ScoreFusion;

pub use index::SearchResult;

pub use index::SharedRegistry;

pub use index::VectorIndex;

pub use index::parallel_add_batch;

pub use index::rrf_fuse;

pub use index::rrf_fuse_top_k;

pub use index::shared_registry;

pub use stats::OutcomeStats;

pub use store::InMemoryStore;

pub use store::RecordStore;

pub use store::SharedStore;

pub use types::MemoryRecord;

pub use types::PriorBundle;

pub use types::QueryBundle;

pub use types::RecordId;

pub use types::VectorRef;

pub use observability::Metrics;

pub use observability::MetricsConfig;

pub use observability::QuerySpan;

pub use observability::SpanContext;

pub use retrieval::QueryEngine;

pub use retrieval::QueryEngineConfig;

pub use retrieval::QueryRequest;

pub use retrieval::QueryResponse;

pub use retrieval::Reranker;

pub use retrieval::RerankerConfig;

pub use retrieval::RerankerType;

pub use wal::WalConfig;

pub use wal::WalEntry;

pub use wal::WalEntryType;

pub use wal::WalReader;

pub use wal::WalWriter;

pub use eval::Evaluator;

pub use eval::EvaluationSummary;

pub use eval::QueryEvaluation;

pub use eval::Benchmarker;

pub use eval::BenchmarkResult;

pub use trajectory::TrajectoryGraph;

pub use trajectory::Episode;

pub use trajectory::Edge;

pub use trajectory::EdgeType;

pub use trajectory::NodeId as EpisodeId;

pub use trajectory::PathSelectionPolicy;

pub use trajectory::TraversalOrder;

pub use trajectory::BranchInfo;

pub use trajectory::PathResult;

pub use trajectory::TrajectoryPhase;

pub use trajectory::PhaseInferencer;

pub use trajectory::PhaseConfig;

pub use trajectory::TurnFeatures;

pub use trajectory::PhaseTransition;

pub use trajectory::SalienceScorer;

pub use trajectory::SalienceConfig;

pub use trajectory::SalienceFactors;

pub use trajectory::TurnSalience;

pub use trajectory::CorpusSalienceStats;

pub use trajectory::Feedback;

pub use trajectory::TrajectoryCoordinate;

pub use trajectory::NormalizedCoordinate;

pub use trajectory::compute_trajectory_coordinates;

pub use trajectory::TrajectoryCoordinate5D;

pub use trajectory::NormalizedCoordinate5D;

pub use trajectory::Ring;

pub use trajectory::RingNode;

pub use trajectory::EpisodeRing;

pub use trajectory::build_weighted_ring;

pub use trajectory::DualRing;

pub use trajectory::DualRingNode;

pub use trajectory::build_dual_ring;

pub use trajectory::build_dual_ring_with_attention;

pub use trajectory::ConservationMetrics;

pub use trajectory::ConservationViolation;

pub use trajectory::ConservationConfig;

pub use trajectory::ConservationTracker;

pub use trajectory::weighted_centroid;

pub use trajectory::weighted_covariance;

pub use trajectory::PathQuality;

pub use trajectory::PathQualityWeights;

pub use trajectory::PathQualityFactors;

pub use trajectory::ConversationPhase;Deprecated

pub use distance::l2_distance;

pub use distance::l2_distance_squared;

pub use distance::inner_product;

pub use distance::cosine_similarity;

pub use distance::cosine_distance;

pub use distance::norm;

pub use distance::norm_squared;

pub use distance::normalize;

pub use distance::normalize_in_place;

pub use distance::normalize_batch;

pub use distance::normalize_batch_flat;

pub use distance::compute_norms_batch;

pub use distance::find_unnormalized;

pub use distance::l2_distance_fast;

pub use distance::inner_product_fast;

pub use distance::cosine_similarity_fast;

pub use distance::cosine_distance_fast;

pub use distance::norm_fast;

pub use distance::normalize_in_place_fast;

pub use distance::normalize_batch_flat_fast;

pub use distance::compute_norms_batch_fast;

pub use distance::normalize_batch_parallel;

pub use distance::compute_distance;

pub use distance::compute_distance_for_heap;

pub use distance::trajectory_weighted_cosine;

pub use distance::trajectory_weighted_cosine_5d;

pub use distance::trajectory_weighted_l2;

pub use distance::trajectory_weighted_inner_product;

pub use distance::trajectory_weighted_distance;

pub use distance::TrajectoryDistanceConfig;

pub use api::RetrievalRequest;

pub use api::RetrievalResponse;

pub use api::PriorBundle as ApiPriorBundle;

pub use api::RankedCandidate;

pub use api::FilterExpression;

pub use api::FilterValue as ApiFilterValue;

pub use api::IngestRecord;

pub use api::MetadataValue as ApiMetadataValue;

pub use api::RetrievalEngine;

pub use api::Corpus;

pub use api::VectorSearcher;

pub use api::SearchHit;

pub use api::RAGBuilder;

pub use api::IndexType;

Modules

api

RAG++ Public API Contracts

buffer

Write Buffer Module

cache

Caching Module

distance

Distance Computation Module

error

RAG++ Error Types

eval

Evaluation Module for RAG++

filter

Filter Engine

index

Index Module

observability

Observability Module

retrieval

Retrieval Module

stats

Outcome Statistics with Welford’s Online Algorithm

store

Record Storage Module

trajectory

Trajectory Memory Module

types

Core Data Types

wal

Write-Ahead Log (WAL)

Macros

ragpp_event

Log structured event.