Archival Data Systems

The State of the Archive

The art-historical record is not a single archive — it is thousands of them, each with its own conventions, its own cataloguing systems, and its own idiosyncratic relationship with the objects it documents. Auction houses maintain ledgers dating to the eighteenth century. Museums hold accession records, conservation reports, and correspondence files that have never been digitised. Private collections pass inventories from one generation to the next in formats that range from meticulous card catalogues to handwritten notes in the margins of exhibition catalogues.

This fragmentation is not merely inconvenient; it is the single greatest obstacle to systematic art-historical research. A provenance gap that could be closed with a single ledger entry remains open because the ledger sits undigitised in a storage facility, its contents unknown to the researcher who needs it.

From Document to Data

Our archival data pipeline transforms raw documents into structured, interlinked records through a series of carefully validated stages:

• Digitisation and OCR — converting physical documents to machine-readable text, with specialised models for historical handwriting, period typefaces, and degraded materials.
• Entity extraction — identifying and tagging the people, places, artworks, dates, and transactions mentioned in each document.
• Schema normalisation — mapping the extracted entities into a unified data model that can accommodate the enormous variety of archival formats without losing the specificity of each source.
• Provenance linking — connecting records across archives to build continuous ownership and exhibition histories for individual works.

Data Quality and Provenance

Every record in our system carries metadata documenting its source, the method by which it was extracted, and the confidence level of each extracted field. We distinguish rigorously between what a source states, what we have inferred from a source, and what remains uncertain. This transparency is not optional — it is the foundation of scholarly trust.

When two sources contradict each other, both are preserved with their provenance intact. Resolution is a scholarly act, not an engineering decision, and our systems are designed to surface contradictions rather than silently resolve them.

Scale and Ambition

The scale of this undertaking is difficult to overstate. Centuries of accumulation have produced millions of documents relevant to the attribution, provenance, and valuation of fine art. Our goal is not to digitise all of them — that would take decades — but to build the infrastructure that makes each new document immediately useful the moment it enters the system, connecting it to everything already known and flagging the gaps it might help to close.