Comment by ethbr1

The difference is that phased safety/efficacy trial costs are sunk.

If it fails, you've produced near $0 value, aside from some more information (maybe) about how something doesn't work. Because those costs are strictly for the execution of the trial, not for the development work before/during/after it.

That's in contrast to your hardware prototyping example, where a failed prototype still usually represents >$0 development value.

The appropriate analogy would be if every time a hardware prototype failed testing, it was required by regulation to scrap it, go back to the design stage, and restart from there.