Observed arrival · 2026-08-28
Histpack Turns Public-Domain History into Retrieval Packs
Credibility concern recorded. The source reference remains available for verification and correction.
A small data product packages historical source files with cleaned OCR, table-aware chunks, metadata, grounded Q&A, and retrieval indexes for AI agents.
Field notes
The project converts source documents into several retrieval layers: cleaned OCR, table-boundary chunks, metadata, grounded Q&A, embeddings, and an FTS5 index. Its Radio 1926 sample reports 5,219 chunks and 92 QA items, with seed source-recall@3 of 7/8 and QA source-recall@3 of 51/92. The full archive is described as a roughly 52.4MB dated zip rather than a file hosted directly on the landing page.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue