How armada-dn/proton-qa works
PROTON-QA is a knowledge-grounded benchmark dataset for question answering over nuclear technical reports. A benchmark user consults annotated questions and answers, follows their provenance to supporting report evidence, and can evaluate a QA system with the supplied evaluation script. Its knowledge graph also represents answer units and types, and links question entities to controlled concepts. The repository documents a RAG prompt and Croissant metadata as additional resources. No source excerpts were provided, so runtime behavior and unevidenced connections are not inferred.
Main entry points: questions.ttl QA annotations · corpus Action reports · schemaorg-current-https.ttl Question-answer model · oa.ttl Evidence annotations
Components 14
Benchmark knowledge
- QA annotations
questions.ttl - Action reports
corpus - Dataset metadata
croissant.jsonld
Semantic vocabularies
- Question-answer model
vocab/schemaorg-current-https.ttl - Evidence annotations
vocab/oa.ttl - Provenance model
vocab/prov-o.ttl - Units and quantities
vocab/qudt.ttl - Unit vocabulary
vocab/qudt-unit.ttl - EuroVoc concepts
vocab/eurovoc_in_skos_core_concepts.ttl - Domain concepts
vocab/concept_extensions.ttl - QA concepts
vocab/qanary.ttl
Benchmark use
- RAG prompt
benchmark/rag_prompt.txt - Benchmark evaluator
benchmark/evaluate.py
Other components
- Benchmark user
Connections 9
- QA annotations → Action reports: links evidence
README.md - QA annotations → Question-answer model: uses question model
README.md - QA annotations → Evidence annotations: uses annotations
README.md - QA annotations → Provenance model: records provenance
README.md - QA annotations → Units and quantities: represents units
README.md - QA annotations → EuroVoc concepts: links concepts
README.md - QA annotations → Domain concepts: links concepts
README.md - QA annotations → QA concepts: uses QA concepts
README.md - Benchmark user → Benchmark evaluator: evaluates QA