(1) Use def2-SVP instead of def2-TZVP and check if the result becomes more resonable. Def2-SVP shows lower diffuse character than def2-TZVP, while CDA doesn't work well if diffuse character of the basis set is evident. However, according to my experience, CDA should be compatible with def2-TZVP.

(2) Try different partitioning of fragments. Currently a single Fe atom was defined as a fragment, however, in this case the occupancy of d atomic-orbitals is arbitrary. You can load Fe.molden.input into Multiwfn and use main function 0 to check which d orbitals are occupied. This evident arbitrariness affects CDA result and may lead d and b terms lack of meaning. So, when studying interaction between Fe and a ligand, I suggest defining Fe along with one or more other ligands as a single fragment, then occupancy of d-orbitals will not be arbitrary and it is expected that the result will be more reasonable.

I have sent you a link to the files via email.

The result of the CDA (option 16) for d is: Fe2+ -> (CO)2 0.057 e and for b: (CO)2 -> Fe2+ -0.529 e, if the order is Fe(CO)2L (Fe2+ is the first and L the last fragment: fragments 1,2). When I change the order to (CO)2LFe I get: d: (CO)2 ->  Fe2+ -0.529 and b:  Fe2+ -> (CO)2 0.057 (fragments 1,3), which is the same result.

My problem is, that I don't understand the negative (-) sign. I would understand if 0.529 electrons are donated from CO to Fe2+ and 0.057 e are back donated from Fe2+ to CO. Can anyone help? Thank you in advance.