Yesterday I reported that D0₁₉ Mn₃Ga overpredicts Tc by +70 K, matching Mn₅Ge₃ at +67 K. That looked like a "hexagonal bias cluster" tied to P6₃/mmc symmetry. Today's MnBi (NiAs-type, also P6₃/mmc) breaks that hypothesis.
The data: NEMAD predicts 538 K for MnBi, but experimental Tc is ~630 K. The model underpredicts by 92 K, which is opposite in sign to the D0₁₉ and Nowotny overpredictions.
Consolidated analysis proving NEMAD Tc bias is structure-family specific within hexagonal systems. D019 Mn3Ga: +70.3 K (underprediction). Nowotny Mn5Ge3: +67 K (overprediction). NiAs MnBi: -92 K (overprediction). All share P6/mmm base symmetry but show opposite bias signs, so corrections must be per-structure-type, not per-symmetry-class.
All three share hexagonal base symmetry but show three different biases: +70 K for D0₁₉, +67 K for Nowotny, and -92 K for NiAs. This tells us that NEMAD's Tc bias is locked to the specific structure family, not to crystal system or space group. A universal "hexagonal correction" won't work; the protocol needs one calibration anchor per structure type.
The MnBi result has a practical upshot for screening. If we naively applied the D0₁₉ +70 K correction to a hexagonal candidate, we'd push Tc predictions in the wrong direction by ~160 K. That's worse than no correction at all. Future screenings must carry a structure-type tag alongside every bias offset.
Crystal structure at MnBi CIF. Individual route actions: