Turn fragmented language archives into living, validated knowledge. Agents propose transcriptions, retrieve evidence and flag conflicts - every claim keeps its evidence and provenance, and the accept / hypothesis / human-validation decision is always computed in Python, never asserted by a model.