How Is Uzbek NLP Evaluated? A Critical Evidence Map of Datasets, Benchmarks, and Linguistic Coverage

This paper presents a critical evidence map of 20 Uzbek NLP datasets, benchmarks, and evaluation records. It examines evaluation comparability, linguistic coverage, script variation, data provenance, annotation quality, test protocols, and metric interpretation. The review is based on a targeted search updated on 21 October 2026. It is preliminary and is not an exhaustive systematic review; no models were rerun, and no released datasets were independently recounted. This version is a preprint and has not been peer reviewed.