For every algorithmic change (and every public function/class), require automated tests that lock in both correctness and failure modes. Minimum standard: - Prefer assertions/doctests over “print and eyeball.” (Printing-only checks are not sufficient.) - Add doctests or pytest cases for: - happy path(s) - edge cases (empty input, boundaries, negative values, single element) - invalid inputs / rejected preconditions (at least one case that should fail or return a sentinel) - Don’t weaken confidence with tests that all assert the same output for different inputs—add cases that meaningfully differ. - Keep existing tests; don’t remove doctests “just because.” - Ensure doctests actually execute in the module (e.g., call `doctest.testmod()` in `__main__`). - Avoid nested helper functions when you rely on doctest coverage inside their docstrings; keep testable code at module level. Example pattern (doctest with edge + failure): ```python def interpolation_search(sorted_collection: list[int], item: int) -> int | None: """ >>> interpolation_search([1, 2, 3, 4, 5], 2) 1 >>> interpolation_search([1, 2, 3, 4, 5], 6) is None True >>> interpolation_search([], 1) is None True """ ... if __name__ == "__main__": import doctest doctest.testmod() ``` This standard directly prevents regressions (new validation rules, edge-case fixes, and invariants) and makes behavior verification repeatable and reviewable.