The biggest advantage is imho the automated calibration to the data. Static rules are often created by data engineers without knowing the business impact. They might set it e.g. at ±5% level.

However, for some columns ±5% is too strict, for others it might be killing the whole business and you should be raising even ±0.5%.

I agree that the predicted ranges & unsupervised detection is the way forward.

Great article, Jeremy!