A mechanical phase variable represents human gait progression that can parameterize the joint kinematic trajectories for lower-limb prostheses control. Current phase variables uses unactuated states, such as the human thigh angle and its derivative or integral, to compute the thigh phase angle to estimate gait phase during stride. With time-dependent states, the phase variable can potentially produce abnormal behavior for the prosthesis controller when the user instantaneously changes speed or inclination. We propose a probabilistic-based approach that uses a maximum likelihood estimation technique from only the thigh angle to derive a holonomic phase variable for a continuous gait phase estimation. Since it does not depend on a time-wise parameter (either derivative or integral), the phase variable can respond to instantaneous changes (e.g., unwanted disturbances) during locomotion. We evaluated the proposed phase variable algorithm across various walking speeds using able-bodied subject datasets, and introduced start and stop transitions to evaluate robustness for non-rhythmic behaviors. The analysis demonstrates a probabilistic adaptation for correcting gait phase abnormalities that can drive locomotion progression for lower-limb prostheses control.