Domain-Specific HTR and Field Model Integration for 11th–12th Century Church Slavonic Uncial

Generic Church Slavonic models in Transkribus fail on 11th–12th century Cyrillic uncial — CER above 20% and systematic layout errors make them practically unusable for this material. EastSlavCyrillic11-12 is the answer: a domain-specific model trained on thirteen manuscripts, achieving 3.41% CER and demonstrating that combining a tailored HTR model with a specialised polygon-based Field Model is the decisive factor for early Cyrillic uncial. A concrete proof of what specialisation can achieve where generic approaches fall short.