Record outcomes and turn them into next week's execution plan.
Prefer real empty states over hidden sample fallbacks.