SYSE.35:4.2 - Choose the comparison and exposure unit

Choose a population or unit that can receive the candidate without invalidating the comparison: for example, a supported user cohort or a bounded set of work items. Identify the control and explain why its observations answer the same outcome question.

Account for task mix, resource conditions, data, time variation and other changes. Where requests from one task can cross candidate and control, preserve enough attribution to understand the result. A shared database or dependency can transmit the candidate’s effects to the control; isolation is a condition to examine, not a label to assume.

Compare bounded exposure with alternatives such as a controlled representative exercise, independent feature activation through a qualified feature-flag or experiment configuration, or a planned cutover. Choose an arrangement that can both limit consequence and supply useful evidence. Do not enlarge exposure merely to make the charts look statistically busy.