Running the example

From the root of this repository, on a Mac (the commands were run on macOS 27.0 with Python 3.14.6):

python3 -m venv .venv
.venv/bin/pip install -r host/requirements-host.txt
.venv/bin/python examples/run_example.py --check
.venv/bin/python examples/run_example.py --check --accelerator cpu
.venv/bin/python examples/run_example.py --request examples/request.json --image fixtures/images/clock_3oclock_256x240.png --tile 256

--check answers four public fixture requests (a SemIf record, a support ticket, a JSON order and a drawn picture) on the GPU at float32 precision (Metal on a Mac), compares them with the provider's own code in float32 on the CPU (run_example.expected.json), sends twelve requests the host refuses, and prints one line, PASS or FAIL. --accelerator cpu does the same on the CPU (XNNPACK, 8 threads).
--request answers your own request (a JSON file of the SystemOne-compatible request shape) and prints the response; --image adds one picture, read at a 448 px square tile (--tile 448, the default) or a 256 px one.
A graph is compiled the first time a request needs it: on the GPU that takes one to three minutes per row graph. One process keeps one row graph at a time; a request of another length closes it and compiles the next.
A request the host refuses prints refused: <error>: <message> and exits with code 3.