NeuroCVguard

Research-only Python software for checking whether supplied cohort identities, partitions and evaluation procedures match an intended generalization claim. It reads local CSV/TSV/JSON, checks participant/transitive dependence and requested domain separation, describes acquisition–target association, generates checked splits and runs a controlled participant-level logistic baseline with optional nested C selection. Offline HTML/JSON reports retain incomplete coverage and limitations. No account, GPU or runtime internet connection is needed.

It does not process MRI images, provide clinical advice, authenticate upstream preprocessing, prove causal confounding or certify a study as leakage-free. Repeated visits alone are not leakage; inspect actual membership and objective. Unknown upstream preprocessing remains unassessable even with a correct Pipeline.

Install and try

From an authorized local source checkout, create a dedicated Python environment. Use .venv/Scripts/python.exe on Windows or .venv/bin/python on Linux/macOS after python -m venv .venv. With that interpreter selected:

python -m pip install -e ".[dev,docs]"
python -m neurocvguard demo --out local_outputs/demo

Open local_outputs/demo/report.html. Inputs are fully synthetic, not patient data. Choose a new output path or explicitly use --overwrite. Read warnings beside coverage: association suggests reviewing acquisition imbalance, while unknown preprocessing asks for evidence rather than a passing verdict.

Private plans/evaluations are separate from projected reports. Default projection is not guaranteed anonymity; inspect artifacts before sharing. Version: 0.1.0. See the release procedure and evidence for publication status and the actual verified installation/platform results.

Documentation and development

• Installation and troubleshooting
• Executable quickstart and synthetic tutorials
• Inputs, configuration, CLI and Python API
• Objectives/warning actions, evaluation, report privacy and limitations
• Contributing/testing, changelog, security and release procedure

Build the full local site with python -m sphinx -W --keep-going -b html docs docs/_build/html and open docs/_build/html/index.html. Run python -m pytest -q --strict-markers --strict-config for the test suite. Actual stage evidence and unrun checks are recorded under state/handoffs/. Local Windows and Linux WSL2 checks and the Python 3.11 direct-dependency floor have been exercised; see installation evidence. macOS and hosted CI remain unverified until their recorded runs pass.

License, support and citation

Copyright 2026 Alireza Emad. Released under the BSD-3-Clause license. Maintainer: Alireza Emad. For private security reports, use the approved contact in SECURITY.md; share only synthetic reproductions. No response-time or long-term support commitment is claimed. Citation metadata will be added only after verified authorship and release details; no DOI or citation badge exists. AI assistance is recorded honestly. Human walkthrough, external-user testing and acceptance remain pending.