the schema is fine for ENR. and claude's deep research run showed another possibility if all things gets loose with ENR - functional data analysis (FDA) for true temporal model architecture and data scheme.
the problem, however, is the redundancy of certain features because they are either near constant, or derivative of actual independent features. so the rawest, true independence in morphological data is painfully narrower.
not a big issue if it's ENR or anything similar like Ridge, but it will have fewer variables to play with. it might be good practice to offload both circularity and solidity altogether during preprocessing of modeling.
the FE pipeline itself is clean though.