But surely the possibility of Type 2 errors corrupting his findings are enormous?

Chris, you may have left one small thing out. If you can perform the test as you so lucidly described it, then you will end up with some statistics

As we know, there are "lies, damned lies, and statistics". The result is then typically left open to interpretation by the examiner, who of course, will have a subjective bias. So we are back to square one. I'll just have a bake off I think