From Provincial Minutes to Regional Archives: Scaling a Specialised HTR Model for 19th-Century Santa Catarina
A highly accurate model for 1880s Brazilian legislative minutes — but CER values between 15% and 40% when tested on other regional archives. This doctoral research confronts the gap between assumed and actual palaeographic homogeneity, asking a question that matters to every HTR practitioner: when a specialised model reaches its limits, is it better to fine-tune incrementally or start again with a richer ground truth?