When an "approved" (connected) Tester is run for a Function against an approved Implementation, it should always pass. If it fails (i.e. votes false, not returns an error that might be temporary), the system is in an inconsistent state. To fix this, we must automatically de-approve the failing Implementation.
[TBD: If the Tester fails all the Implementations, should it instead be disabled? How do we know what to do when?]