Central bank speeches dataset — searchable and cited

Speeches are where policymakers say the things a committee statement cannot. They are also where the earliest signals of a shift usually appear, delivered at a conference in front of an audience of two hundred people and a webcast nobody watched.

As a corpus they are scattered by construction: each institution publishes its own officials, regional banks maintain separate archives, and the format ranges from clean HTML to a scanned PDF of prepared remarks. Building a usable dataset means collecting across institutions and normalizing the metadata that makes speeches comparable — who spoke, in what role, where, and when.

Hindcast maintains that corpus with delivery timestamps preserved, roles recorded as they stood at the time, and the text chunked so retrieval returns an argument rather than a document. Every passage keeps its link back to the published original.