Reference project · Museum Ulm

Directorate files
1933 – 1945

The museum directorate's correspondence, digitized and searchable. Provenance research as a living archive — opened up page by page.

≈ 30,000 pagesSütterlin · Kurrent · FrakturFunded by the German Lost Art FoundationMuseum Ulm · City of Ulm · 1tm solutions
Fig. 01

Starting point

A collection full of answers — but impossible to query.

The directorate files of Museum Ulm document the museum leadership's correspondence from 1933–1945: acquisitions, loans, letters to dealers, authorities, and collectors. For provenance research this collection is a central source — for instance to clarify under what circumstances works entered the collection.

The files had been digitized, but only as image files: not searchable, not structured, not linked to authority data. Every inquiry meant manually leafing through tens of thousands of pages.

Fig. 02

The challenge

Historical scripts

A large share of the pages is handwritten in Sütterlin and Kurrent; printed and typewritten documents use Fraktur. Standard OCR fails at both.

Scholarly standards

Provenance research needs verifiable statements: every extracted piece of information must remain traceable to its source — the individual scanned page.

Archival structure

Fonds, folder, correspondence, page: the archival order must be preserved and exportable to standards such as EAD and CIDOC CRM.

Fig. 03

The approach

  1. 01

    Preserve the structure

    On ingest, the archival order — fonds, folder, correspondence, page — is reconstructed from file names and the filing plan. Deterministic metadata is captured without AI.

  2. 02

    Knowledge becomes an analysis profile

    The museum describes its knowledge — cataloging tables, rules, examples. arcafind generates a versioned analysis profile from it: prompt, schema, and test cases, validated on real pages before activation.

  3. 03

    AI analysis with evidence

    The AI transcribes Sütterlin and Fraktur and extracts people, places, dates, and events — per document and per page, with an evidence anchor, confidence, and provenance per field. Manual corrections always take precedence.

  4. 04

    Authority data, search, publication

    People and places are linked to the GND. Full-text and semantic search are fused; curated results can be shared as publications and exported to EAD or CIDOC CRM.

Fig. 04

From the collection

Fig. 05

The result

  • 01

    The collection is searchable — literally and by meaning, across transcript and structured metadata.

  • 02

    Every extracted statement remains traceable to the scanned page: evidence anchor, confidence, and provenance per field.

  • 03

    People and places are linked to GND authority data and ready for archival standards (EAD, CIDOC CRM).

  • 04

    The research team curates the results; publications make curated selections available internally or publicly.

“Every page is evidence. arcafind makes the archive legible.”

Provenance research · Museum Ulm

A similar collection at your institution?

Whether records, collection documentation, or media data — we will show you what arcafind can do with your own material.

Request access