Look for
Detection
What lets the system distinguish situations that require different responses?
One way alignment happens
How does a difference in the world become a difference in what the system does?
Alignment can depend on mechanisms that make some distinctions detectable, stable, salient, and connected to possible responses.
Try seeing it
A feature can physically exist without functionally existing for the system. What matters is whether the system can pick it up in a form that changes its available action.
Signals, perceptual invariants, attention, interfaces, labels, boundaries, and environmental structure can all make a difference locally usable.
Look for
What lets the system distinguish situations that require different responses?
Look for
Of everything detected, what becomes important enough to alter action?
Look for
Which responses appear available, possible, permitted, or natural from the system's current state?
A characteristic failure
Adding information does not guarantee alignment. A warning can be ignored, a metric can be observed but disconnected from action, or an anomaly can be detected without being treated as evidence of error.
The relevant question is not merely whether information is present, but how it reorganizes the field of possible behavior.
A useful perturbation
If not, the difference may exist for us without existing functionally for the system.