Abstract:  [Objective] High-elevation long-runout rock avalanches may undergo progressive damage, fragmentation, particle dilation, and a transition from block-dominated motion to debris-flow-like movement. Conventional equivalent-fluid models efficiently reproduce runout but do not explicitly resolve damage-induced disintegration, whereas fully discrete approaches are computationally expensive at field scale. This study aims to clarify how damage evolution and particle-state transition affect the post-failure dynamics of the 1991 Touzhai landslide in Yunnan, China. [Methods] Three numerical schemes were established under the same three-dimensional terrain and initial source geometry: a Dynamic Analysis of Landslides in Three Dimensions (DAN3D) equivalent-fluid model, a Landslide Post-Failure 3D (LPF3D) multiphase-state model without damage evolution and structural disintegration, and an LPF3D full-state model coupling the Holmquist-Johnson-Cook (HJC) damage model, explicit structural discontinuities, and particle-state transition. Their motion duration, velocity evolution, runout distance, run-up pattern, deposition characteristics, damage variable, and particle volume fraction were compared to evaluate the roles of structural degradation and state-dependent particle interactions. [Results] Four principal results were obtained. First, the LPF3D full-state model reproduced the main observed kinematic characteristics of the Touzhai event. The simulated motion lasted approximately 120 s and evolved from initial failure and rapid acceleration through impact-induced disintegration and channelized transport to deceleration and deposition. The maximum horizontal runout was approximately 3650 m, the vertical drop was nearly 960 m, and the peak velocity exceeded 23 m/s. The simulated mass also reproduced the observed run-up pattern on the opposite valley slope, where field investigations reported a maximum run-up of approximately 55 m. The final deposit was thicker in the central part and thinner toward both sides, broadly consistent with the mapped distribution of the source, transit, and deposition zones. Second, the damage field and particle volume-fraction field revealed a continuous mechanical-state transition rather than a thermodynamic phase change. The damage variable increased progressively from a relatively low initial level and approached 1.0 in highly damaged zones during repeated impact and shearing, indicating substantial degradation of the original load-bearing rock structure. Concurrently, the front of the moving mass experienced stronger stretching and impact, developed a lower particle volume fraction, and entered a dilute, collision- and inertia-influenced state, whereas the rear retained a higher volume fraction and a denser block- or granular-state structure. This produced a distinct longitudinal pattern characterized by a dilute front and a dense rear. As the mass entered the gentler deposition area, velocity and shear rate decreased, particle contacts were re-established, and the main body underwent re-compaction. Third, the comparison between the two LPF3D schemes isolated the influence of damage and structural disintegration under the adopted material and boundary settings. When HJC damage evolution and structural disintegration were disabled while the basic LPF3D material and contact parameters were kept unchanged, the mass stopped after a runout of approximately 1500 m. In contrast, the full-state scheme reached approximately 3650 m. The comparison indicates that damage accumulation, block separation, particle rearrangement, and dilation reduced persistent interlocking and frictional contacts and increased the relative contribution of collision and inertia, thereby modifying resistance and energy redistribution during motion. Fourth, the DAN3D equivalent-fluid scheme reproduced the macroscopic runout through a calibrated frictional rheology, demonstrating its efficiency for estimating velocity, runout, and deposition at the slope scale. However, its state variables did not directly represent rock damage, block disintegration, or particle-state transition. The three-scheme comparison therefore shows that similar macroscopic runout can be reproduced through different model representations, but only the full-state LPF3D framework explicitly links the simulated runout to progressive structural degradation and the subsequent transition among solid-like, dense granular, dilute granular, and deposition states. [Conclusions] Under the parameter and boundary conditions adopted for the Touzhai case, damage accumulation and structural disintegration are important factors promoting particle rearrangement, dilation, mechanical-state transition, and long-runout movement. The simulated transition is spatially non-uniform: the strongly disturbed front becomes dilute and increasingly collision- and inertia-influenced, whereas the rear remains comparatively dense and contact-dominated. DAN3D is suitable for efficient macroscopic runout analysis, while the LPF3D full-state model provides a more physically interpretable description of the coupled evolution from jointed rock mass to blocks and granular debris. These findings are case-specific and should not be interpreted as a universal threshold or a single mechanism applicable to all high-elevation landslides. [Significance] By integrating damage evolution, structural discontinuities, and particle-state-dependent dynamics within one computational framework, this study provides a physically constrained approach for reconstructing high-elevation rock avalanche-to-debris-flow processes. The framework can support case-specific runout assessment and hazard analysis while clarifying the model assumptions and mechanisms underlying the predicted movement range.