What lives here
- Whitepapers that formalize the main ABIS claims.
- Dated findings studies that package benchmark results into public releases.
- Research notes that keep the program active between papers.
- Validation and disclosure pages that explain how evidence is handled.
Current program
The strongest thread in the current library is ABIS: deterministic behavioral measurement for AI systems, with an emphasis on silent model change, behavioral stability, and operational governance.
How to read the library
- Start with the ABIS program page for the overall framing.
- Read the whitepapers for the main claims and evidence.
- Use the drift-monitor section for dated findings and benchmark releases.
- Follow research notes for lighter updates and release context.