Objective: To report the fidelity of the enhanced upper limb therapy programme within the Robot-Assisted Training for the Upper Limb after stroke (RATULS) randomized controlled trial, the types of goals selected and the proportion of goals achieved. Design: Descriptive analysis of data on fidelity, goal selection and achievement from an intervention group within a randomized controlled trial. Setting: Out-patient stroke rehabilitation within four UK NHS centres. Subjects: 259 participants with moderate-severe upper limb activity limitation (Action Research Arm Test 0–39) between one week and five years post first stroke. Intervention: The enhanced upper limb therapy programme aimed to provide 36 one-hour sessions, including 45 minutes of face-to-face therapy focusing on personal goals, over 12 weeks. Results: 7877/9324 (84%) sessions were attended; a median of 34 [IQR 29–36] per participant. A median of 127 [IQR 70–190] repetitions were achieved per participant per session attended. Based upon the Canadian Occupational Performance Measure, goal categories were: self-care 1449/2664 (54%); productivity 374/2664 (14%); leisure 180/2664 (7%) and ‘other’ 661/2664 (25%). For the 2051/2664 goals for which data were available, 1287 (51%) were achieved, ranging between 27% by participants more than 12 months post stroke with baseline Action Research Arm Test scores 0–7, and 88% by those less than three months after stroke with scores 8–19. Conclusions: Intervention fidelity was high. Goals relating to self-care were most commonly selected. The proportion of goals achieved varied, depending on time post stroke and baseline arm activity limitation.