Designs with 100% D-efficiency in CVA are both perfectly orthogonal and level balanced (equal number of exposures per level within each attribute). Note that orthogonal arrays (from Hahn and Shapiro) aren't always level balanced within each attribute (for example, with an asymmetric design where the number of levels differs across attributes).
Note that with our latest CVA software, it by default assumes you want to generate multiple versions of the questionnaire, where each version is as close to 100% D-efficient as it can get. This is good such that different respondents often get different concept arrays, which helps control for psychological order & context effects.
To add practice (throw-away) tasks prior to your real tasks, I suggest exporting your already generated questionnaire design with 8 tasks to .csv file (from the Design tab). Then, modify the .csv file so that you add 1 or 2 practice tasks to the beginning of the questionnaire, meaning that your .csv file now supports 9 or 10 tasks.
Now, change your CVA project within SSI Web to have either 9 or 10 tasks total, matching your modified .CSV file. Now, use the Import capability on the Design tab to browse to the modified .CSV file and read it into your CVA project.
Now, when you compute utilities you can omit the first 1 or 2 tasks as warm-ups, but the remaining 8 tasks are the orthogonal, highly efficient design.