Haeyong Chun, Youngsoo Park, Hyo Young Kim

The primary purpose of this study was to test whether within-session engagement in a mission-based fitness app exhibits (a) a mid-sequence motivational lull and (b) goal-proximal acceleration as users approach a reward-contingent completion point. Using large-scale behavioral log data from an exercise application “Daysquat”, we examined how stopping likelihood and effort intensity change across a structured squat program. We modeled stopping behavior with a discrete-time proportional hazard model and indexed motivational intensity using inter-set completion intervals. Results supported a non-monotonic, U-shaped engagement signature. The hazard of stopping at the third set was significantly lower than at the second (Hazard Ratio; HR ≈ 0.58), indicating reduced cessation around the midpoint. In contrast, stopping risk increased sharply at the reward boundary, producing a pronounced “cliff effect” at the fourth set (HR ≈ 28.2). Timing analyses converged with these patterns: mean inter-set intervals peaked at the midpoint transition (set 2→3: 532 s) but dropped substantially during the final push (set 3→4: 143 s), consistent with goal-gradient acceleration. Additionally, previous-day activity was associated with lower stopping risk (HR ≈ 0.28), and higher platform-level engagement was also protective (HR ≈ 0.98). In contrast, longer user tenure was associated with a small positive effect on stopping (HR ≈ 1.09). These findings demonstrate that micro-incentives and salient goal structure can shape motivation within a single session, offering design implications for digital health interventions targeting sustained exercise adherence.