Does a different random arrangement of the same composition give the same DFT energy?

E179 result — prediction confirmed: the −71 is the composition's, not the decoration's. Seed-1 decoration (19/54 sites shared), same Vegard a, JOB DONE in 23 iterations: E/atom −6463.52808 eV, E_form(DFT) = −72.5 meV/atom vs E177's −71.1 — ΔE = 1.4, an order of magnitude inside the 12.4 meV decoration floor. Two things follow. (i) The DFT formation energy of a random 54-atom decoration at this composition is reproducible to ~1–2 meV/atom, so single-decoration rung-4 verdicts are sound at this cell size. (ii) The eCE scored the two decorations 6 meV apart (−17.9 / −11.9) where DFT puts them 1.4 apart — the model resolves decoration differences that are not there, at the same time as it misses the composition's energy by 54. Both are what an embedding fitted to RHEA's rattled, strain-cut labels would do at a composition 0.27 from its nearest training point. Four QE cells now exist; six more before E175's refit.