JErrRAG-Eval-800

JErrRAG-Eval-800 is the public benchmark release aligned with the paper's canonical dataset for Java exception diagnosis. This release contains 800 normalized Stack Overflow-derived question-answer pairs with provenance-preserving identifiers and URLs. The package supports benchmarking Java exception-diagnosis systems, retrieval-augmented generation variants, and failure-analysis workflows. Total records: 800 Split sizes: train=639, dev=80, test=81 Difficulty distribution: easy=324, medium=388, hard=88 Exception taxonomy coverage: 264 exception types Important release note: files that originated under the local splits/ directory are uploaded to Zenodo with flattened filenames such as splits_train.jsonl and splits_train_ids.txt. This dataset is distilled from Stack Overflow / Stack Exchange user-contributed content. The release preserves source_question_id and reference_url for attribution and traceability. See LICENSE_NOTICE.md for the release-specific attribution note.