ReproAudit v0.1.0

ReproAudit v0.1.0

ReproAudit v0.1.0 is the first public release of a deterministic, rule-based tool for auditing consistency across structured machine-learning experiment artifacts.

Highlights

• Audits experiment.yaml, claims.yaml, raw_results.csv, and summary_results.csv
• Includes ten deterministic integrity, statistical, configuration, and conclusion checks
• Recomputes means and sample standard deviations using ddof=1
• Supports absolute and relative numeric tolerances
• Produces console, Markdown, and strict JSON reports
• Provides deterministic exit codes suitable for CI
• Includes clean and deliberately corrupted MEC examples
• Includes reproducible example generation and distribution-artifact verification
• Licensed under MIT with CFF 1.2.0 citation metadata

Installation

Download the wheel attached to this release:

python -m pip install reproaudit-0.1.0-py3-none-any.whl

Python 3.11 or later is required.

Quick Start

reproaudit audit examples/mec_clean \
  --output-dir /tmp/reproaudit-clean \
  --format all

The deliberately corrupted example exits with code 2 because it contains expected audit errors:

reproaudit audit examples/mec_corrupted \
  --output-dir /tmp/reproaudit-corrupted \
  --format all

Release Assets

• Python wheel
• Source distribution
• SHA256SUMS

GitHub also provides source archives for this tag.

Current Scope

This release audits structured YAML and CSV experiment packages. It does not parse PDF or Word papers, use OCR or LLMs, provide a Web UI, automatically repair research artifacts, or publish to PyPI.

Citation and License

Citation metadata is provided in CITATION.cff. ReproAudit is licensed under the MIT License.