A reranker cannot promote a document the retriever never returned, which makes candidate-set recall a hard ceiling on the entire exercise. Compute the ceiling and the disappointment becomes arithmetic: three points of headroom were available, twelve were sitting upstream, and the latency was an implementation detail.
You added a cross-encoder reranker. p95 latency roughly doubled and answer quality did not move. Explain what happened and fix it.
A reranker cannot promote a document the retriever never returned, which makes candidate-set recall a hard ceiling on the entire exercise. Compute the ceiling and the disappointment becomes arithmetic: three points of headroom were available, twelve were sitting upstream, and the latency was an implementation detail.
Updated Sep 2026 · Grounded in real Forward Deployed Engineer interview loops and written to a senior-engineer editorial bar.
A reranker can only reorder what the retriever already handed it, so the candidate set's recall is a hard ceiling on anything reranking can achieve. Say that first and the rest of the answer writes itself. The number that should be computed on the whiteboard is the headroom: with recall@50 at 0.88 and end-to-end at 0.85, a perfect reranker is worth three points while twelve points sit upstream, and the candidate who notices the ratio is four to one has told the interviewer where they would have spent the sprint.
No comments yet — be the first to share your approach.
