SVCM MSCA PF 2026
Marie Skłodowska-Curie Actions — European Commission Cardiff University

PROVENARCH · HORIZON-MSCA-2026-PF-01

Archival Provenance and Computational Authorship: Modular Cultural Infrastructures for Digital Heritage

Provenance-aware LLM mediation in Digital Humanities

Provenance before fluency.

Draft ID SEP-211353721 · Submit 2026-09-09 16:00:00 UK

O1

Writing Provenance Layer

Bind AI-generated text to archival sources, extraction nodes, and model operations.

O2

Voice fidelity

Evaluate when fine-tuning preserves rather than flattens authorial and cultural voice.

O3

Open toolkit

Deployable open-source prototype and responsible-design methodology for practitioners.

Research programme

Computational Provenance for AI-Assisted Digital Heritage

How can an AI system preserve the genealogy of knowledge — not merely cite a source, but expose the chain of transformations through which generated text became what it is?

GraphRAG, layout-aware document parsing, and Postgres-native graph modelling are enabling technologies within this programme, not the destination.

Flagship themes

T1 · flagship

Computational Provenance

The Writing Provenance Layer (O1) formalised as a W3C PROV-O extension aligned to CIDOC-CRM — grounding the methodology in a standard two decades older than LLMs.

T5 · flagship

Document Intelligence

A falsifiable novelty test: does any published system anchor knowledge construction in precise page/bounding-box coordinates, as the applicant's own extraction pipeline already does? A rigorous negative result is itself the finding.

Also in scope

Synthetic Mediation / Algorithmic Heritage

Archives and platforms that generate afterlives without lying about how they know — memory situated as problem diagnosis, not field home.

Voice Fidelity

Restoration, interpretation, simulation, forgery — a rubric for when generated text preserves rather than flattens authorial and cultural voice.

Computational Authorship

Provenance for AI co-production — editor, conservator, or co-author?

Knowledge Engines

A research instrument, not a chatbot.

Cultural Infrastructure

Built for Europeana, DARIAH, IIIF, CIDOC CRM — the customer is the institution, not the model vendor.

The 24-month fellowship builds the instrument; the research programme continues beyond it toward an ERC-scale proposal.