SAFETY
For archival AI, safety starts with evidence: what source, what page, what language, and what uncertainty.
SAFETY
studio1804's safety work is practical and source-grounded. For Rasin, that means citations stay visible, archival provenance is preserved, and generated answers are treated as research aids rather than replacements for primary sources.
Principles
01
Source Visibility
Rasin answers should preserve the source document, page, collection, and language behind each claim.
02
Transparency
We publish methods, name limitations, and make research workflows reproducible where source rights allow.
03
Data Control
Archives should retain usable exports: OCR text, metadata, embeddings, indexes, and citation records.
04
Measured Development
Evaluation, review, and historical context matter more than shipping unsupported summaries quickly.