spt-reasoning-core

Core reasoning engine with pattern matching and similarity scoring

Overview

Provides the foundational types and algorithms for storing, retrieving, and analyzing reasoning patterns with adaptive learning capabilities. This is the base crate of the ReasoningBank system -- all other reasoning crates depend on it for core types like Pattern, VectorEmbedding, and ReasoningEngine.

Key Features

• Pattern Matching -- Store and retrieve reasoning patterns with metadata and task outcomes
• Similarity Scoring -- Vector embedding comparison via SimilarityMetrics
• Compression Learning -- Track and analyze compressed Claude API call outcomes with CompressionTrajectory
• Incremental Statistics -- Efficient running averages and trajectory aggregation
• Configurable Engine -- ReasoningEngine with pluggable EngineConfig

Dependencies

Key internal dependencies: none (this is the leaf crate).

External: ndarray, ordered-float, uuid, chrono, serde.

Usage

use spt_reasoning_core::{
    ReasoningEngine, EngineConfig, Pattern, PatternMetadata,
    SimilarityMetrics, VectorEmbedding,
    CompressionTrajectory, CompressionInsights,
};

let config = EngineConfig::default();
let engine = ReasoningEngine::new(config);

API Reference

Core Types

Type	Description
ReasoningEngine	Main engine for pattern matching and reasoning
EngineConfig	Configuration for the reasoning engine
Pattern	A stored reasoning pattern
PatternMetadata	Metadata attached to a pattern
TaskOutcome	Result of a task execution
SimilarityMetrics	Scoring functions for vector similarity
VectorEmbedding	Dense vector representation for patterns
CompressionTrajectory	Records outcomes of compressed API requests
CompressionInsights	Derived recommendations from trajectory analysis
TrajectoryStats	Aggregated statistics over trajectories
RunningAverage	Efficient incremental averaging
ReasoningError	Error enum for all reasoning operations