Look for
Error detection
What differences count as deviations, and at what scale are they detected?
One way alignment happens
Feedback and correction
Can a departure from the larger pattern change what the parts do next?
Alignment becomes more robust when error is not merely present but able to propagate through the system and reorganize behavior.
Try seeing it
Error has to be able to matter.
A system may contain perfectly competent parts and still fail if those parts cannot detect or respond to departures that matter at the larger scale.
Feedback creates a path by which the consequences of local activity can alter later local activity.
Look for
Error routing
How does information about the deviation reach parts capable of changing the outcome?
Look for
Corrective authority
What actually changes once an error is detected: behavior, constraints, priorities, or the larger pattern itself?
A deeper problem
Successful correction can undermine future correction.
Once error disappears, the processes that detected it may stop receiving useful information. Monitoring may decay. Slack may vanish. Exploratory variation may be eliminated.
So maintaining alignment can require preserving not just correct behavior, but the conditions under which future errors remain visible and actionable.
A useful perturbation
Introduce an error. Where does information about it travel?
Following that path often reveals the architecture of correction.