Automatic text recognition for contemporary and historical printed or handwritten works has become a crucial tool in the inventory of many humanities scholars employing digital methods, but few retrodigitization software packages are adapted to this environment with its unparalleled diversity, domain-specific conventions, and low-resource settings. We examine how the design principles behind kraken, a freely-licensed ATR engine optimized for use on historical and non-Latin script documents, pertain to specific use cases encountered in the humanities and allow it to be easily adapted to process even highly non-conventional material. In addition, we shine a light on the significant functional additions, performance enhancements, and quality of life improvements added in its recent version 5 stable releases. Among these are advanced reading order support including trainable reading order, unsupervised recognition pretraining, and a new public model repository with enhanced metadata. The latest release of the software can be found at https://kraken.re