Crate ragdrift_core 

Pure-Rust core for ragdrift.

This crate computes drift scores between a baseline and a current sample across five dimensions used in RAG monitoring: data, embedding, response, confidence, and query-pattern.

The crate is split into two layers:

• stats holds the underlying statistical primitives (Kolmogorov-Smirnov, PSI, MMD, Wasserstein).
• detectors composes the primitives into batch detectors that take baseline and current samples and return a DriftScore.

use ndarray::Array1;
use ragdrift_core::detectors::ConfidenceDriftDetector;

let baseline = Array1::from(vec![0.9, 0.85, 0.92, 0.88, 0.91]);
let current = Array1::from(vec![0.6, 0.55, 0.62, 0.58, 0.61]);
let detector = ConfidenceDriftDetector::default();
let score = detector.detect(&baseline.view(), &current.view()).unwrap();
assert!(score.exceeded);

Re-exports

pub use error::RagDriftError;

pub use error::Result;

pub use types::BaselineSnapshot;

pub use types::DriftDimension;

pub use types::DriftReport;

pub use types::DriftScore;

Modules

detectors

High-level drift detectors. Each composes one or more stats primitives into a DriftScore.

error

Error type used across the crate.

stats

Statistical primitives used by the detectors.

types

Public data types: drift scores, reports, and baseline snapshots.