Accuracy is meaningless for ranking, the whole point is the order. The real question behind the question is whether your relevance is graded or binary and whether one good result is enough, because that picks the metric. Then the trap: your offline NDCG can climb while online engagement falls.
Updated Aug 2026 · Grounded in real Forward Deployed Engineer interview loops and written to a senior-engineer editorial bar.
Accuracy is meaningless for ranking, the whole point is the order. The real question behind the question is whether your relevance is graded or binary and whether one good result is enough, because that picks the metric. Then the trap: your offline NDCG can climb while online engagement falls.
The grading follow-up is 'why not just use precision@k?' The strong answer is that precision@k ignores order within the top k and ignores graded relevance, so a list with the best result at position 10 scores the same as one with it at position 1. NDCG fixes both with a position discount and graded gains. Candidates who can't say why position discounting matters are answering a classification question, not a ranking one.
No comments yet — be the first to share your approach.
