PerturbGuard 1.0.3

1.0.3

• Renames the user-facing validation workflow to expected-findings checks while keeping validation-benchmark as a deprecated CLI alias.
• Adds real-data case-study manifests and expected output snapshots for SciPlex, Jorge, and LINCS-style examples.
• Adds README scope table clarifying what PerturbGuard can and cannot prove.
• Aligns release metadata for GitHub and PyPI.

PerturbGuard 1.0.1

Patch release responding to early review feedback.

Changes:

• Clarifies maturity language: first public release, beta-quality guardrail software until validated on user data.
• Changes PyPI classifier from Production/Stable to Beta.
• Adds descriptive CLI help for commands and options.
• Adds balanced-random split generation with optional --balance-column stratification.
• Documents heuristic limitations for confounding checks, split generation, model evaluation, and target mapping.
• Updates package URLs to the actual GitHub repository.

Install:

pip install perturbguard==1.0.1

PerturbGuard 1.0.0

First stable PerturbGuard release.\n\nInstall from PyPI:\n\n\\�ash\npip install perturbguard\n\\n\nPyPI: https://pypi.org/project/perturbguard/1.0.0/\n\nIncludes dataset validation, AnnData repair, config inference, audits, split generation, claim checking, adversarial leakage checks, model prediction evaluation, benchmark manifest validation, large-file profiling, dataset-card generation, interactive reports, CI, docs, and packaged workflow assets.