fix(experiments): reject an OMARS design that never varies a factor

[kgdunn]

Summary

• omars_properties checked that each factor reaches the middle level at least once, so that its pure quadratic is not the constant column a two-level factor gives. The mirror case was unchecked: a factor left at the middle level in every run has the constant quadratic 0. Both are inestimable; only the first was caught, so is_omars was returning True for designs whose main-and-quadratic model matrix is rank deficient.
• That permissiveness fed a bias in the search. max_second_order_correlation skips constant columns, which is correct on its own terms, but a factor pinned at the centre removes its own quadratic and all of its interaction columns from the comparison, so the score improves. generate_omars(selection_criterion="min_second_order_correlation") minimises exactly that quantity, so it had a direct incentive to produce the degenerate designs the verifier was accepting, and did: across a three-to-six factor sweep it returned one in roughly a third of cells, including a spurious perfect 0.000 at three factors in nine runs.
• The fix requires an outer level as well as a middle one. With the verifier corrected the ILP rejects those candidates during the search and spends its budget on real ones: degenerate results fell from twelve to three, and two sizes that previously produced no usable design at all (three factors in nine runs, five factors in thirteen) now produce one.

uses_middle_level = np.any(np.abs(matrix) <= tol, axis=0)
uses_outer_level  = np.any(np.abs(matrix) >  tol, axis=0)      # new
quadratics_estimable = bool(np.all(uses_middle_level & uses_outer_level))

Behaviour change

is_omars now returns False for matrices it previously accepted. Every such matrix has an inestimable quadratic, so no correct caller can depend on the old answer. It is corrected in place rather than put through the deprecation schedule.

Test plan

How this was verified:

• New regression test, test_factor_never_leaving_the_middle_is_not_omars, the mirror of the existing test_full_factorial_is_not_omars. Five runs in three factors is the smallest case that shows it: with the third factor pinned at the centre, is_omars returned True while the main-and-quadratic model matrix had rank 4 of 7. The test asserts the rank deficiency alongside the verdict, and that is_balanced and main_effects_orthogonal are both still True, so it is clear which check was the missing one.
• pytest -k omars: 291 passed, 2 skipped.
• The 13-run definitive screening design still verifies, so the new condition does not reject valid designs.
• ruff check . and ruff format --check . clean; mypy src/process_improve clean over 146 source files.

Checklist

• Version bumped in pyproject.toml (PATCH for fixes/docs/config, MINOR for new features)
• Tests added or updated where relevant
• ruff check . passes
• CHANGELOG.md updated

---

Generated by Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!