Einstein's final classical unified field theory was described in the 5th edition of his "The Meaning of Relativity", in an appendix. To his death, he believed that this version 'might be true'. In this, he proposed a non-symmetric pseudo-tensor, similar to but not the same as an affine connection, and posited various properties to restrict possible equations (transposition invariance, lambda invariance). Field equations come from a variational principle constructed from a contracted curvature expressed in terms of this pseudo-tensor.
I have never worked through this appendix in detail. So far as I know, very little work was ever done on this theory, as no one believed its motivation and conceptual structure. Like string theory, it begged for solutions corresponding to known observations.