It occurs to me that a system, trained on peer-reviewed applied-machine-learning literature and Kaggle winners, that generates candidates for structured feature-engineering specifications, based on plaintext descriptions of columns' real-world meaning, should be considered a requisite part of the "meta" here.

Ah, and then you could iterate within the resulting feature-engineering-suggestion space as a hyper-parameter between experiments, which could be optimized with e.g. https://github.com/HIPS/Spearmint . The papers write themselves!