Checks/asserts in code base versus test suite

[ttimbers]

In data analysis and machine learning code, many checks/asserts are in code base themselves (e.g., data validation, data leakage, etc) and there might not be tests for these exceptions in the test suite. Is this sufficient for robust ML code? Or one of the necessities, but sufficiency would also require tests for these exceptions in the test suite?

[ttimbers]

ChatGPT agrees that checks/asserts are not sufficient: https://chatgpt.com/share/671aae62-6140-800d-8996-f7cb23459e53