Scientific research often depends on structured comparative datasets, but much of the evidence needed to build them remains locked inside narrative sources: books, encyclopedias, unstructured reports, scanned PDFs, and other heterogeneous scholarly texts. 

Traditionally, this kind of data collection required human researchers to painstakingly work through scattered and sometimes incomplete reports in order to extract and code this information. 

Recent advances in large-context reasoning models create new possibilities for working with these materials directly and at scale. 

These models can read long documents, identify relevant passages, compare them to explicit coding definitions, and produce structured candidate data linked to textual evidence in minutes. This talk explores how such models can support the translation of unstructured narrative-based knowledge into structured, auditable databases using the Seshat: Global History Databank as an example. 

Photos and/or videos may be taken at this event and used by the Complexity Science Hub for press coverage, publications, and social media.