The proposed energetic model with three parameters: (with intercalation energies $ε_{8a}$ and $ε_{16c}$ into 8a and 16c lattice site types of LTO, respectively, and the repulsion energy $J$ between adjacent 8a and 16c sites both occupied by Lithium atoms) qualitatively explains the LTO's open-circuit potential plateau as a phase transition in which Lithium atoms move from 8a sites (at 0% stoichiometry) to 16c sites (at 100% stoichiometry). Even looking only at the paper's analysis, this explanations seems certainly true because the energetic model's predictions seem to be accurate within 5%, whereas the difference between $\Deltaε = ε_{16c} - ε_{8a}$ and $J$ is much larger: 50%, and the system "should" have two distinct phases as long as $J$ is larger than $\Deltaε$ and the temperature of the particle is not crazily high (e. g., less than 400K). Besides, the phase transition dynamic was experimentally verified, too [2].
The above applies to $\mathrm{Li_{1-2}Ti_{2}O_4}$ flavour of Lithium-titanate-oxide. For $\mathrm{Li_{4/3-7/3}Ti_{5/3}O_4}$, the model developed in the paper is not sufficient for making a confident assertion that particles undergo simple phase transition between 8a and 16c sites during lithiation or delithiation, yet this is apparently what happens in reality anyway [2]. The actually measured open-circuit potential of $\mathrm{Li_{4/3-7/3}Ti_{5/3}O_4}$ also has a single big plateau, as well as $\mathrm{Li_{1-2}Ti_{2}O_4}$.
Quantitatively, the proposed energetic model predicts the position of LTO's open-circuit voltage plateau within about 3% (e. g., the prediction might be 1.45V whereas the real plateau might be 1.4V, depending on the specific LTO anode: particle size, the preparation process, binders, etc.). This precision might be sufficient for some applications and not sufficient for others, yet the main purpose of the model is a qualitative description of the material dynamics, not accurate fitting of the actual OCV relationship. This is in line with the conjecture made on the page about Physics-based vs. equivalent circuit cell models: physics-based models should guide scientific understanding, whereas the models used in a production battery management system (BMS) should not necessarily be physical if learned models are more accurate.