In practice it generally comes down to demonstratable utility more than direct concerns about animal well-being. At the most basic level, if you can learn a use

In practice it generally comes down to demonstratable utility more than direct concerns about animal well-being. At the most basic level, if you can learn a useful lesson from subjecting an animal to whatever conditions, then the work is generally approvable. But if the researcher has not done their due diligence in experimental design then the whole operation is pointless, or worse than pointless if they then go and p-hack their way through a publication. For a physics analogue, think of animal experiments like a beamline proposal. In order to get time on the beamline you need to explain what you are doing, what your experiment is intended to accomplish, and why the experiment achieves those goals. With these basic checks you can hopefully ensure that the allocated time is not obviously going to be wasted. Animal experiments can easily be about as expensive in terms of money and time as a beamline-based project, including the cost for the beamline operation itself.