Conventional wisdom on utilizing cation redox in intercalation materials for energy storage has been challenged by a new class of cathode materials where oxygen anions actively participate in the reversible electrochemical reaction. This type of Li-rich layered oxides (LRLO) have captured great interests in the field of energy storage materials as they are considered as one of the most promising cathode materials for future high energy density lithium-ion and lithium metal batteries. However, the activation of solid-state oxygen redox reaction usually causes some detrimental effects on the electrochemical performance (e.g., O2 loss, voltage fade) preventing the practical application of LRLO (Y. S. Meng et al., Nature Energy, 2018, 3, 641-647; X. Yu et al., Nature Energy, 2018, 3, 690-698). Till now, the charge-compensation mechanism associated with oxygen reduction/oxidation is still under fierce debate. For instance, J.M. Tarascon et al. proposed the presence of peroxo-like species (O2 n-, O-O: 2.45 Å) in lattice oxygen redox based on the research results of 4d/5d-TM LRLO (Science, 2015, 350, 1516-1521; Nature Materials, 2017, 16, 580-586). In contrast, P. Bruce et al. clearly demonstrated that there is no peroxide species formed in 3d-TM LRLO (Nature Chemistry, 2016, 8, 684-691). The controversial viewpoints regarding oxygen reduction/oxidation at solid state may be originated from the intrinsic complexity of this class of materials and the limitations of the characterization techniques. Our work reveals the origin of stable and reversible lattice oxygen redox in 3d TM LRLO, specifically, the localized structure distortion (i.e., short O-O pairs) accompanying the oxygen redox is favorable for the stability of the average layered lattice structure, maintaining the high-voltage character and topotactic structure-reversibility of LRLO. With a model compound (Li1.2Mn0.54Co0.13Ni0.13O2), we for the first time observe a local occurrence of short O-O pair (~2.39 Å) concomitant with the lattice oxygen redox via neutron pair distribution function. It was found that the distance of this short O-O pair is very similar to that of O-O dimer in peroxo-like species, but our quantum mechanical calculation results suggest that the bonding nature of this short O-O pair (in which potential π-type overlap dominated) is quite different from that of peroxo-like O-O dimer (in which σ-type bond formed). Based on these understandings in our work, an optimization guidance for LRLO is unambiguously presented: designing a robust layered structure with high-covalency TMs while constructing a flexible local structure (e.g., honeycomb structure) with high-ionicity TMs to accommodate the local distortion induced by oxygen redox, thus achieving the reversible high energy density in LRLO. Figure 1