Test Quality

[pmaasz]

• Most tests are smoke tests: result.success == True, len(result) > 0, isinstance(x, dict). They catch crashes but not regressions in physics.
• No seeded RNG in stochastic tests. Any test that exercises SA / GA / MC is inherently flaky — it either uses a huge tolerance or asserts nothing specific.
• No convergence assertion on optimizers — i.e. no test says "merit after ≤ 0.5 × merit before". Adding even one such test per algorithm would be high value.
• Numerical tolerances are often hand‑picked rather than derived (places=4 in test_aberrations.py:60).