01
Evidence-centered
Keep claims traceable to the archival sources from which they emerge.
SOURCE ↔ CLAIM
OPEN-SOURCE RESEARCH PROTOTYPE · MIT LICENSE
From fragmented
archives to
connected histories.
Provenance Wiki helps researchers investigate historical relationships across photographs, correspondence, catalogues, and museum records—with AI-assisted discovery and evidence at the center.
ARCHIVE / 001
c. 1930–1940
SOURCE RECORD
TRANSCRIPTION
DISCOVERCONNECTQUESTIONVERIFY
01 / THE APPROACHResearch principles,
Research principles,
not just features.
Provenance is more than a linear chain of ownership. Historical evidence reveals networks of people, objects, places, and institutions—and the uncertainties within those networks.
02
Relational histories
Investigate connections beyond any single artwork or institutional collection.
OBJECT ↔ NETWORK
03
Human-guided AI
Use AI to discover research leads; let scholars interpret, correct, and verify.
SUGGEST ↔ REVIEW
04
Revisable knowledge
Preserve the possibility of uncertainty, correction, and reinterpretation.
REVISE ↔ REVISIT
02 / RESEARCH WORKFLOWFrom sources
From sources
to relationships.
Five connected stages turn fragmented documents into a structured research environment. Verification is continuous: researchers return to the evidence whenever interpretations change.
01 / INPUT
Archival sources
Historical photos, correspondence, catalogues, and records.
02 / PROCESS
AI assistance
Information extraction, entity recognition, and model-assisted inquiry.
03 / ORGANIZE
Local archive
Source-linked records, entity identities, and search indexes.
04 / EXPLORE
Discover & ask
Hybrid search, network and timeline, source-grounded Q&A.
05 / INTERPRET
Human review
Check passages, merge entities, annotate, and revise findings.
REVIEW → CORRECT → REINDEX Interpretation feeds back into the research archive
THE AI PROCESSING LAYER
Select a path to exploreTwo paths, different stages of development.
Today: Local storage of research data, with selected photographs or text submitted to external services for multimodal extraction, embeddings, and language-model assistance. Researchers must assess provider policies before using restricted archival material.
Experimental direction: Testing locally deployed AI models for selected processing tasks. The goal is more deployment flexibility and greater institutional control over where archival data is processed—not a claim of full feature parity or production readiness.
03 / CORE FUNCTIONALITYTools for asking
Tools for asking
better questions.
Different ways of entering the same evidentiary environment: browse the sources, explore connected entities, and ask questions grounded in document passages.
DOCUMENT / EXTRACT
HISTORICAL RECORD
EXTRACTED INFORMATION
PERSON
OBJECT
PLACE
linked to source passage ↗
NETWORK / TIMELINE
EVENTS OVER TIME
RESEARCH ASSISTANT / RAG
QUESTIONWhich sources support this connection?
EVIDENCE-BASED RESPONSE
[SOURCE 01] [SOURCE 02] VIEW PASSAGES ↗
ILLUSTRATIVE INTERFACE DIAGRAMS
Turn archival images into readable, editable records with extracted entities and source context. The illustration is conceptual, not a screenshot of the current application.
Explore the implementation on GitHub ↗ALSO IN THE PROTOTYPE
Keyword and semantic search · Entity aliases and merging · Researcher annotations · Edit history · PDF dossier exports
04 / RESEARCH CONTEXTBuilt for archival
Built for archival
complexity.
Provenance Wiki began as an independently developed software experiment by Mengge Cao. Its research questions and demonstrations draw in part on work connected with the University of Chicago's Dispersed Chinese Art Digitization Project (DCADP).
The goal is broader than one collection or geographic field: to support responsible, source-aware research into the movements and relationships of art objects, people, and institutions.
From fragmented evidence to connected histories — and back to the sources.
OPEN SOURCE / MIT LICENSEExplore the code.
Visit the GitHub repository ↗Explore the code.
Join the conversation.
Read the documentation, run the prototype locally, suggest improvements, or discuss research applications. This website is informational; the application itself runs in a local research environment.