f2x50 benchmark dataset + Evaluation Artifacts for ADFD-Mediated Fortran-to-Python Porting

Evaluation artifacts for the paper "Identifying Latent Declarative Representations for Code Migration" submitted to ASE 2026. This dataset accompanies an autoencoder-inspired framework that ports legacy Fortran repositories to Python via Annotated Data Flow Diagrams (ADFDs). It contains: 50 Fortran-to-Python porting results across three complexity tiers (low: 20, medium: 20, complex: 10), including inferred ADFDs, generated Python code, and per-repository evaluation metrics. 47 direct LLM translation baselines for comparison (same models, same evaluation suite, no intermediate representation). Evaluation artifacts: encoder/decoder logs, DFD round-trip comparisons, semantic judge scores, functional test results, error taxonomies, and cost tracking. f2x50 benchmark specification: repository list, GitHub URLs, complexity tier classification, selection criteria, and reproduction scripts. The 50 source Fortran repositories are publicly available on GitHub; URLs and cloning instructions are provided in f2x50/README.md. Note: This submission is anonymised for double-blind review. No author-identifying information is included in any artifact.