code-quality-corpus / README.md
CatQualia's picture
Upload folder using huggingface_hub
3a792b7 verified
|
Raw History Blame Contribute Delete
2.57 kB
metadata
license: other
license_name: csil-3.0
license_link: https://catqualia.com/licensing
language:
  - en
tags:
  - code-quality
  - code-review
  - static-analysis
  - repair
  - sft
pretty_name: CatQualia code-quality corpus — semantic smell classes with before/after fixes
size_categories:
  - 10K<n<100K
configs:
  - config_name: default
    data_files:
      - split: train
        path: data.jsonl

CatQualia code-quality corpus — semantic smell classes with before/after fixes

39,383 rows · 25,541,792 bytes · JSON Lines, one object per line.

What this is

Real code smells paired with the fix: a smell_class that names the semantic problem (not just the syntax), the original lines, the corrected lines, the file and line it came from, and a rationale explaining why the original was wrong. Useful for code-review or repair training where the label has to say what is wrong rather than merely flagging a line.

Schema

Fields of the first record, read from the file in this repository:

Field Type
id str
smell_class str
severity str
file str
line_before int
before str
after str
rationale str
grade str
provenance str
consumer str

Measured

Property Value
Rows 39,383
Bytes 25,541,792
Distinct top-level fields 11

The row count was measured with wc -l against the file staged here, not against a source copy.

How it was produced

By the author's own generation pipeline: candidate rows are produced from seed material, graded, and the survivors assembled. Several fields (notably grade and any alien_or_typical judgement) are the pipeline's own assessment, not an external label. The method is described at https://catqualia.com/forest.

Known limitations

  • Synthetic and self-graded. grade, alien_or_typical and failure_class come from the generating pipeline. They are not human-annotated ground truth.
  • Generated at scale. Volume is not evidence of quality; a large fraction of rows may be unremarkable.
  • English only.
  • Provenance of individual rows varies and is recorded per row where the pipeline captured it.
  • Not peer-reviewed, not independently evaluated.

Licence

CatQualia Structural Isomorphism License (CSIL) v3.0 — attribution: Christopher Betances, catqualia.com. Non-commercial use is free with attribution; commercial use requires a commercial licence. Full text: LICENSE in this repository and catqualia.com/licensing