Problem. numbers is sorted in non-decreasing order. Return the 1-based indices [i, j] (i < j) of the two
values that add up to target. Exactly one solution exists; use O(1) extra space.
numbers = [2, 7, 11, 15], target = 9 → [1, 2]
numbers = [2, 3, 4], target = 6 → [1, 3]
numbers = [-1, 0], target = -1 → [1, 2]
ML context. "Two items whose costs fit a budget" is a common shape: two model calls whose latencies add up to an SLO, two features whose combined memory fits a device, two experiments whose sample sizes fill an allocation. With the costs sorted, two pointers answer it, and the counting variant (how many pairs fit) shows up in capacity planning.