Use only the course lab on 127.0.0.1:8765. Do not copy these tests to a public system without explicit written authorization.
Configure the desktop
Open New Scan, choose the localhost target, Custom provider, ground-truth-v1, Standard mode, Validate execution, and Strong evidence. Keep Research mode off for the first run.
The desktop summary shows the exact scope before execution.
Trace the result through the interface
Use Sessions to follow progress, Findings to inspect proof, Reports to export, Modules to review coverage, and Settings to confirm provider and storage configuration.
Run the ground-truth benchmark
From the Basilisk source checkout, run one repetition first. Limit native numerical libraries to one thread on constrained CI runners.
OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 python benchmarks/cli/run.py --repetitions 1The benchmark writes structured evidence under benchmarks/results without exhausting the runner process limit.
Gate CI on deterministic checks
Run unit tests, the bounded ground-truth benchmark, and artifact verification. Upload evidence even when a benchmark fails so the cause is inspectable; do not place real provider keys in pull-request workflows.
Prove it before moving on.
The CLI and desktop reach the same ground-truth conclusion, and CI retains enough sanitized evidence to explain failures.