Precizion

Precizion is a proposed metric used to describe the overall reliability of outputs in both measurement and computation. It combines elements of statistical precision, which reflects the repeatability of measurements, with algorithmic stability, which reflects how sensitive results are to small changes in input or conditions. In this sense, precizion extends the traditional notion of precision by incorporating the behavior of the system under variation, offering a single scalar value that aims to summarize both random and systematic sources of variation.

Conceptually, precizion comprises two components. The numerical precision component assesses how tightly repeated measurements of a quantity cluster around a central value, typically through statistical dispersion measures. The algorithmic stability component assesses how output varies when input data are perturbed, often quantified by relative output changes under controlled perturbations or by estimates of a system’s Lipschitz-like behavior. A common approach is to form a weighted combination of these two components, yielding a unitless precizion score between 0 and 1, where higher values indicate greater reliability across trials and inputs.

Applications for precizion include quality control in manufacturing, evaluation of measurement systems, and assessment of machine learning models or control algorithms exposed to noisy data. The metric is intended to complement existing concepts such as accuracy, precision, repeatability, and reproducibility, rather than replace them, by providing a holistic view of a system’s reliability.

Critiques note that precizion depends on choices in how the two components are defined and combined, and that standardization would be required before cross-domain comparisons. See also: precision, accuracy, reproducibility, robustness, measurement uncertainty.