# Package validation

Before packaging, the Suite24 source was checked with:

- Python byte-code compilation of every top-level module;
- six unit tests covering output shapes, finite all-masked attention, exact Committee fallback, correction bounds, history-only behavior, and monotone evidence gating;
- a synthetic end-to-end selection/refit smoke run for the evidence-conditioned model;
- command-line parser smoke test.

The real F1-F7 run cannot be executed in the packaging environment because the university caches exist only on the training server. `check_suite24_environment.py` therefore validates the required canonical caches and raw graph sources before the expensive run begins.
