The proof library

The plan is ten public artifacts, each built on documents anyone may republish. Today one is built and none has passed a sealed acceptance test. This page lists what exists and what is still planned. It shows no scores because there are no certified results yet.

#ArtifactDocuments & licenceStatus
01Enterprise policy questions in five file formats
The same policies as PDF, Word, PowerPoint, HTML and Markdown
Synthetic · MIT licenceBuilt, not certified
02Contracts & clause questions
Clause lookup across real contracts
CUAD · CC-BY-4.0Planned
03Technical & equipment manuals
Troubleshooting, tables and scanned pages
RAGBench eManual, DelucionQA · CC-BY-4.0Planned
04Financial & compliance reasoning
Questions that need tables and text together
FinQA, TAT-QA · CC-BY-4.0Planned
05Healthcare administration guidance
Non-clinical policy answers
CMS Medicare manuals · public domainPlanned
06Government procurement regulations
Section-level citations at scale
Federal Acquisition Regulation · public domainPlanned
07Support knowledge base
Customer-support articles
Bitext · CDLA-Sharing-1.0 (legal review pending)Planned
08Multilingual documentation
Questions and documents in different languages
MultiEURLEX · CC-BY-4.0Planned
09Versioned policy corpus
Updates, deletions and point-in-time answers
eCFR snapshots · public domainPlanned
10Retrieval alone versus retrieval plus a tuned model
Measured value of tuning, on the same documents
RAGBench slice or artifact 01 · CC-BY-4.0Planned

Development measurements for artifact 01 are on the performance page. They are not acceptance results.