METAFOR RAG research data: algorithm performance observations and convergence evidence

This dataset is the versioned research-data release supporting metafor-rag v0.4.0. It contains METAFOR algorithm-design candidates and benchmark performance observations from a portable 137-file HYBRIDS input snapshot, together with compact convergence evidence for retrieval-augmented algorithm-design recommendations. The convergence component includes the trusted dat_index.pkl.gz analyzer cache and a derived Parquet dataset containing 2,218,980 convergence-curve records covering 31 algorithms, four benchmark competitions, and multiple problem dimensions. The package also provides source and convergence manifests, SHA-256 checksums, provenance metadata, schema documentation, and a data dictionary. The release supports reproducible ingestion, verification, retrieval, and analysis with metafor-rag. Raw optimization traces and generated plots are not included. Because dat_index.pkl.gz is a Python pickle file, it should only be loaded after verifying its checksum and explicitly trusting its provenance. The research data are licensed under CC BY 4.0.