Agent context should keep background stable while routes stay open.
The route can be provisional.
The inference can be soft.
The next reader can disagree.
But not everything needs to wobble.
Useful background should stay stable enough to orient the work.
That is what makes route uncertainty usable.