In efforts to control disease, mathematical models and numerical targets play a key role. We take the elimination of a viral infection as a case for exploring mathematical models as 'evidence-making interventions'. Using interviews with mathematical modellers and implementation scientists, and focusing on the emergence of models of 'treatment-as-prevention' in hepatitis C control, we trace how projections detach from their calculative origins as social and policy practices. Drawing on the work of Michel Callon and others, we show that modelled projections of viral elimination circulate as 'qualculations', taking flight via their affects, including as anticipation. Modelled numerical targets do not need 'actual numbers' or precise measurements to perform their authority as evidence of viral elimination or as situated matters-of-concern. Modellers grapple with the ways that their models transform in policy and social practices, apparently beyond reasonable calculus. We highlight how practices of 'holding-on' to projections in relation to imaginaries of 'evidence-based' science entangle with the 'letting-go' of models beyond calculus. We conclude that the 'virtual precision' of models affords them fluid evidence-making potential. We imagine a different mode of modelling science in health, one more attuned to treating projections as qualculative, affective and relational, as excitable matter.