Closed-Vocabulary Multi-Label Indexing Pipeline for Historical Documents

The handwritten minutes of ETH's School Board 1854–1902 were provided with annual indexes reflecting a 19th-century taxonomy. We reconstruct this original indexing scheme by building an end-to-end pipeline that converts scanned images to Markdown and applies a closed-set multi-label classifier strictly constrained to the fixed historical label set. The resulting model recovers historical annotation practices and enables scalable, automated indexing of archival records.