"Ein Freinttlichen Trunck": The Zechzettel Dataset for Handwritten Text Recognition and Information Retrieval

The Zechzettel dataset provides a focused resource for evaluating Handwritten Text Recognition (HTR) under real-world historical variation. It contains over 1,800 digitized petitions submitted in 16th-century Augsburg, featuring diverse handwriting styles, dialectal language, and non-standard orthography. Expert transcriptions and document-level annotations preserve uncertainties and contextual features, while image-level labels describe visual attributes such as legibility and ink quality. Designed to go beyond standard metrics like CER and WER, Zechzettel enables systematic analysis of how manuscript variability affects HTR performance and downstream tasks such as information retrieval.