> Ultimately, it comes down to the halting problem: If there's a mechanism that can be used to alter the measured behaviour, then the system can change behaviour to take into account the mechanism.

No, that’s the opposite of the halting problem…