Repository navigation
Installation
Choose the setup that matches your job:
- Running an event: install STRATHEX.
- Using the prediction library: install V2 below.
- Running Linux V3: follow the competition setup guide.
- Trying the calibration changes: see Accuracy Preview.
Use Python 3.10–3.13. Python 3.13 is recommended. Create the environment on Linux:
python3.13 -m venv .venv
source .venv/bin/activateOr in Windows PowerShell:
py -3.13 -m venv .venv
.\.venv\Scripts\Activate.ps1Then install:
python -m pip install "strathmark==2.0.1"
python -c "import strathmark; print(strathmark.__version__)"Both 2.0.0 and 2.0.1 are published on PyPI. The 2.0.1 patch changes packaging metadata and documentation; the V2 calculations are unchanged. STRATHEX pins its reviewed 2.0.0 source separately.
Next, run the Python example.
Install the API dependencies in the same environment:
python -m pip install "strathmark[api]==2.0.1"Then follow REST API to choose an explicit database path and start the server on loopback. The public calculation endpoint does not require Ollama or a residual ML model.
The repository version is 3.0.0rc7. Use Python 3.13 and install from the repository root into its own environment:
python -m pip install -e .
strathmark demoThis synthetic demo uses V2 calculations and opens no operator database. To work on V3, follow contributor setup and install the exact release lock. Installing the source build does not select V3 for an existing competition.
Home · Quick start · FAQ · Repository
Start here
Understand the predictions
- How handicaps work
- V2
- V3
- Accuracy Preview
- Wood and diameter
- Recency
- Uncertainty and simulation
- Fairness
- LLM integration
- Rulebooks
- Legacy prediction keys
Operate and develop