As computational tools become more important for researchers, we contend that the verification and validation of researcher-built software should happen long before it is published (Nature 501, 472; 2013).

The time and effort needed to fix errors increases exponentially the later they are identified. Exposing such computational errors only when a research paper is peer reviewed threatens to undo years of hard work.

To unlock the full potential of computer-based science, software engineering must be at peak quality throughout. This depends on linking computer modelling with reality, demonstrating fitness for purpose, tracking and understanding scientific abstractions and assumptions, and considering all these factors when weighing up the value of software-derived results.

Software-based science is necessarily cross-disciplinary: scientists ensure that the research is relevant and software engineers make sure that it is correct. Their skills should be combined from the outset.