Run tests

Install the development dependencies, then run the smallest meaningful tier first and expand by blast radius.

python -m pip install -e ".[dev]"

# Daily development baseline
python tools/ci_tier.py pr-default

# Local API, runtime, planner and compiler changes
python tools/ci_tier.py pr-runtime

# Distributed planner, audit and benchmark contracts on CPU
python tools/ci_tier.py pr-distributed

# Accelerator-backed and device-bound Triton tests
python tools/ci_tier.py gpu-scheduled

Plain pytest entry points work as well:

python -m pytest tests/unit -q
python -m pytest tests/qec -q
python -m pytest -m qiskit
python -m pytest -m pennylane

What each tier proves

Tier	Proves	Does not prove
Push gate	Imports, minimal circuits, autograd, pure planner and audit helpers	Runtime integration, distributed behaviour, performance, release readiness
Local runtime	Seeded integration coverage for local runtime and API behaviour	Multi-process transport, GPU execution, scalability
Distributed CPU	CPU distributed semantics, fail-closed gates, benchmark contracts, release-gate validation	Real multi-GPU or multi-node capacity expansion
GPU scheduled	Accelerator-backed and device-bound kernel tests	Multi-node transport or release scalability by itself

An empty marker selection is not verification. CPU distributed tests prove semantics only and are never used as scalability release evidence.