Mathematical Foundations & Drift Detection Reliability

The Problem: Standard monitoring tools rely on "static thresholds." They only alert you when a process has already failed.

Core Engine: Relative Variance Analysis

We utilize the Coefficient of Variation (CV), a robust statistical measure of relative variability, as the core of our engine. This score is passed through an exponential decay function to produce the final 0-100% Predictability Score. This allows for the quantification of "Process Health" rather than a simple Pass/Fail binary.

Sliding Window Drift Detection

Our engine applies the core CV calculation to a moving window of data. This identifies "Process Creep" by tracking the stability score over time. It catches micro-deviations that are still "in-spec" but trending toward failure, allowing for preventative intervention.

Numba-Accelerated Inference

By compiling our core Python calculation to machine code via LLVM, we provide sub-millisecond latency on moving-window calculations. This makes the tool suitable for high-speed manufacturing lines and real-time financial data streams where every millisecond counts.