We publish what the fieldwork teaches us — grounded in real collections, reviewed by practitioners.
Drafted internally, under review, not yet public.
OCR & HTR
A case study on per-line confidence, cataloger review, and the cost of trust in archival OCR workflows. Draws on 120K folios across three historical scripts.
Arabic · HTR · Error tolerance
Description
How to generate scope notes, biographical sketches, and container lists that pass institutional review — and what fails silently when you don't design for it.
DACS · LLM · Finding aids
Discovery
A unified discovery layer over five institutional catalogs without schema migration. Translation happens at query time, not ingest.
Search · Cross-modal · Multi-institutional
Principles
A design principle for AI systems that process records: assume the first pass is wrong in known ways. Revisable without losing provenance.
AI · Archival ethics · Error tolerance
We don't publish for citation count. Every paper is grounded in real fieldwork, reviewed by practitioners who catalog records for a living, and held back until the claims survive that review.
We take a small number of research collaborations per year. Tell us what you're working on.