Smartwatch heart-rate (HR) displays may be inaccurate during rapid post-exercise transitions. This study evaluated timepoint-specific smartwatch–electrocardiography (ECG) error and its reproducibility across three resistance-exercise sessions. Sixty healthy young men completed three identical sessions five days apart using the Apple Watch Series 7, Galaxy Watch 4, Fitbit Charge 5, and Garmin Vivosmart 5. Displayed HR was recorded before each set and at 5, 30, and 60 s afterward. The primary outcome was participant-level session-mean error across eight 5 s observations. Mean absolute error increased in the range of 1.94–2.32 bpm before exercise to 8.71–25.43 bpm at 5 s and decreased to a range of 3.58–10.06 bpm at 30 s. Session-mean 5 s error intraclass correlation coefficients were 0.892, 0.771, 0.879, and 0.834 for Apple, Galaxy, Fitbit, and Garmin, respectively, indicating moderate-to-good through good-to-excellent reliability. Session effects were not statistically significant. Raw single-observation limits of agreement were substantially wider than session-mean limits, and all devices exhibited proportional bias. Participant-level variance estimates remained similar in exploratory models including ECG-response and wrist-position covariates. Therefore, early post-set error contained a reproducible session-level component, but individual 5 s readings were not interchangeable with ECG measurements. These findings support cautious interpretation of post-set HR displays without establishing a stable participant-specific error trait.