Provenance-Constrained Requirements Generation for Model-Based Ship Systems Engineering: Knowledge Base and Evaluation Data

This dataset accompanies the paper "Provenance-Constrained Requirements Generation for Model-Based Ship Systems Engineering." It contains the curated domain knowledge base (1,364 evidence units spanning L1 regulatory clauses, L2 manufacturer specifications, and L3 supporting documentation, annotated with a 21-component × 45-aspect engineering taxonomy), the retrieval indexes (BM25 + dense vector), per-run language-model response caches, and the full set of generated and evaluated requirement sets (RAG v4 / no-RAG v4 / qwen-max cross-model / random-retrieval ablation / blind correctness review) needed to reproduce all numbers reported in the paper.
Companion source code: https://github.com/losyser/RAG_for_REQ