DacteloBy the Dactelo team at Taktek.
A diagnostic check of existing readers on corrected lines
On 30 October 2026, Dactelo audited seven existing corrected units and ran fresh Gemini, Claude and Tesseract readings. A released Muharaf recognizer was also tested through whole-page baseline detection. Visual inspection found two units were printed headings, leaving five handwritten lines from two rulings.
On that handwriting subset, fresh Gemini character error was 8.33%. The requested alias was gemini-4-argon-medium; its immutable resolved version was unavailable. This is a diagnostic result: one reviewer had corrected the references with draft assistance, and six of seven saved corrections exactly matched the earlier draft. Fresh Claude and Tesseract scored 72.22% and 79.63% CER; Muharaf with page layout scored 51.39%, including extraction errors.
The layout track also exposed fragmented and missed lines, so it cannot isolate recognizer quality. The experiment preserved outputs and prepared a blind review packet for a different reader. The next useful measurement needs independent references and broader coverage. These five lines do not establish corpus-wide accuracy, legal reliability or a winning system.