The delivery of precision medicine in rehabilitation will require not only precise measurement of participant response, but also precise measurement of the "ingredients" of intervention and their dose. As an example, we report the measurement of motor error in two treatment groups from a randomized controlled trial in toddlers (mean age 26.3 months) with cerebral palsy (CP). Our objective was to measure the type and amount of motor error during physical therapy sessions in young children with CP. Participants were stratified by motor function and age and randomly allocated to "conventional" physical therapy that generally prevented falls or to an intervention that encouraged error experience by not preventing falls (experimental group). Baseline motor and cognitive function were measured using the Gross Motor Function Measure-66 (GMFM-66) and Bayley 3 cognitive subscale (B3-C) prior to randomization. Randomly selected video recorded therapy sessions were manually coded to identify losses of balance defined as falls (child contacted floor), rescues (therapist prevented fall) or saves (child recovered their balance independently). Average number of losses of balance per session were higher in the experimental group than the conventional group due to significantly greater falls. Saves were infrequent in both groups but were also significantly higher in the experimental group. Average number of rescues did not differ between groups. In the experimental group, greater frequency of falls was significantly related to GMFM-66. In both groups, greater frequency of saves was related to GMFM-66. Neither total losses of balance per session nor rescues were related to GMFM-66 in either group. There were no significant relationships between losses of balance and baseline cognition in either group, except greater frequency of saves was related to higher cognitive ability in the experimental group. Our observations suggest that motor error experience is lower in toddlers with CP compared to peers with typical development but can be manipulated to higher doses of error during therapy sessions. Future work should investigate the relationship between type and amount of error experience and rehabilitation outcomes, as well as other "ingredients" of rehabilitation therapy. Tools to automate the precise measurement of intervention content are necessary for broad scale implementation.