The Anatomy of a Generative Prototyping Session

Anonymous author(s)

In generative prototyping, users formulate a request, the system realizes it, and users inspect the prototype before the next request. In a preregistered study, 100 non-experts used Mark, our instrumented generative prototyping environment, to complete three tasks in familiar and unfamiliar domains. We dissected 1,643 of the generations into prompt formulation, extent of change, verification, and next move. Results show that prompt formulation predicts the extent of change and is associated with the likelihood of repair. Verification time shows no clear association with repair, while prompt vocabulary varies with domain familiarity. [X] further non-experts and [X] designers discussed selected events of their sessions in replay interviews. The interviews provide an account of how reflective practice extends to the verification of the result, and how expertise informs both formulation and evaluation. The anatomy contributes an instrumented environment, an analytic lens, and empirical results to the study of generative prototyping.