Many software requirements are identified only after a product is deployed, once users have had a chance to try the software and provide feedback. Unfortunately, addressing such feedback is not always straightforward, even when a team is fully invested in user-centered design. To investigate what constrains a teams evolution decisions, we performed a 6-month field study of a team employing iterative user-centered design methods to the design, deployment and evolution of a web application for a university community. Across interviews with the team, analyses of their bug reports, and further interviews with both users and non-adopters of the application, we found most of the constraints on addressing user feedback emerged from conflicts between users heterogeneous use of information and inflexible assumptions in the team's software architecture derived from earlier user research. These findings highlight the need for new approaches to expressing and validating assumptions from user research as software evolves.