Installation

GeoStep is proprietary software. Install from an authorised checkout.

git clone https://github.com/griffin-analytics/geostep.git
cd geostep
python -m venv .venv
source .venv/bin/activate
python -m pip install -e .

On Windows activate .venv\Scripts\activate instead. The methodological repair was verified with Python 3.12.11. The release candidate supports Python 3.11 and 3.12. Other interpreter versions are outside this release contract. pyproject.toml is the package dependency contract; the validation record lists the versions actually used for the statistical checks.

The private release constrains NumPy to the tested 2.3 series. The combination of pandas 2.3.3 and NumPy 2.5.3 passed tests but produced widespread timedelta deprecation warnings, so it is excluded from this release contract. The 0.5 release verification records the clean environments actually checked. Bounds do not certify every permitted combination.

Core dependencies include NumPy, pandas, SciPy, statsmodels, scikit-learn, matplotlib, seaborn, joblib, Rich and psutil. They are installed by the package. Use a project environment and retain its dependency versions with trial results. Do not upgrade dependencies in the middle of a frozen analysis without rerunning its verification.

Verify the environment

python -c "import geostep; print(geostep.__version__)"

Run the complete synthetic example in getting started. For development checks, install the tools in requirements-dev.txt and run:

PYTHONPATH=src pytest
PYTHONPATH=src python scripts/validate_parallel_inference.py

The second command is a bounded simulation, separate from ordinary unit tests. Read its scope and results.

Upgrade existing trials

Preserve the original allocation, schedule, outcomes and analysis plan. The methodological corrections intentionally change some numerical behaviour and seeded allocations. Never generate a replacement allocation for an experiment that has already started. Reanalyse using the actual assignment and strata.

For headless plotting, set MPLBACKEND=Agg. See troubleshooting for statistical input and inference errors.