Old letters, parish registers, census pages, and wills are the hardest input in handwriting OCR: period scripts, faded ink, archaic spelling, and often a language other than modern English. General document OCR and LLM vision both fall below 50% accuracy on historical hands, so this is a specialist-only category.
Two specialists dominate, with opposite philosophies. One works accurately out of the box on the first upload; the other is designed to be trained on a specific hand and rewards a large single-writer archive after a serious time investment. Which is right depends on whether you have one consistent hand and time to train, or a mixed collection you want readable now.