What can be true
The parts succeed
Individual components satisfy their local goals, rules, or criteria of competence.
A dissociation
Aligned parts ≠ aligned whole
Success at one scale does not guarantee success at another.
Components can each behave competently, rationally, or even exactly as intended while their interactions produce an unwanted collective outcome.
≠
What does not follow
The whole succeeds
Their interaction need not produce a desirable or coherent higher-scale outcome.
Try seeing it
The same behavior can change meaning when you change scale.
Growth and resource acquisition may be successful behavior for an individual cell. In the context of an organism, unconstrained growth can become cancer.
Similar patterns occur elsewhere. Individually reasonable choices can generate segregation. Individually aligned artificial agents can interact in ways that produce an unwanted organization-level result.
What this reveals
Alignment does not automatically compose.
To understand the whole, we need more than facts about the parts. We need to know how they are coupled, what information crosses scales, which constraints bind them, and which larger variables their activities collectively preserve.
Try this elsewhere
Change the scale at which you judge success.
Does the behavior that looked aligned before still look aligned now?