logic about heuristic sampling

[uestcyile]

In source code heuristic-sampling.cc for map-sparsification, I see the code of get score and cost for each landmark.
But what confuse me is that you add score and cost, as the picture bellow shows:

To my understanding, landmark with higher socre means better quality, and lower cost also means better quality, so should we use "score minus cost" as the final quality metric instead of "score plus cost" ?
And the code should be "double score_and_cost = landmark_scores[landmark_id] - landmark_costs[landmark_id];" , right ?