Most commercial software has goals like “never crash”, “be easy to use”. These goals are usually not top of the list for scientific software. “Give the right result or crash” is a common one. Standard software engineering techniques have been developed to meet typical commercial quality goals. I suspect that this is one of the reasons why attempts to apply these techniques to scientific software development can generate more heat than light.

A lot of new scientific software is used to generate hypotheses, which then will be tested by experiment. Some of it, when it becomes more mature, is used to guide practical decisions, even safety-critical ones like clinical decisions, or judgements about whether a new alloy is suitable for use in an airframe. The quality goals for the software can therefore change during its lifetime. It is better to be explicit about which goals you have tried to meet.