Treat OCR identifiers as fallible dataset grouping keys

When splitting an object-detection dataset by physical asset, treat historical OCR strings as fallible grouping keys. A character substitution can put different views of the same asset into training and validation even when normalized identifiers are disjoint. Exact file hashes and whole-image perceptual hashes can also miss this case when viewpoint and lighting differ. Review plausible identifier neighbors alongside visual evidence, merge confirmed entity groups, and preserve that grouping through dataset version creation. Similar strings alone are insufficient to merge distinct assets; this check reduces known leakage without proving exhaustive entity separation. Applies when: Repeated physical assets are grouped using extracted text. Keywords: OCR dataset leakage split grouping identifiers images validation