A Provenance-First Framework for Evidence-Linked FAIR Data and Visual Analytics of Chaoshan Drawnwork Heritage

Chaoshan drawnwork is a nationally designated intangible cultural heritage of China. It encompasses thread work, lace making, and embroidery techniques with a broader historical background extending to the 1860s; this study focuses on documented development from 1886 to 2000. Historical records are scattered across archives and exist mostly as scanned documents with mixed text and visual plates. A structured openly reusable dataset for computational research is currently missing. This paper asks how restricted, heterogeneous, and OCR degraded heritage archives can be transformed into reusable structured evidence while preserving provenance, uncertainty, and human interpretive authority. We present a two stage framework. First, a FAIR compliant processing pipeline creates citable snippets, canonical entities, confidence scored candidate relations, and Chinese CLIP based visual retrieval links. Second, a web based visual analytics system helps experts inspect evidence chains rather than accept computational outputs as ground truth. The resulting dataset comprises 180 processed documents, 70,863 text snippets, 19,197 visual snippets, and 1,753 evidence backed candidate relations. The evaluation reports provenance, leakage, confidence, and expert verification checks, while leaving full precision, recall, top-k accuracy, and cross-domain validation to future work.