Recall@10 is 0.94 and climbing, and answer accuracy is going down. The two facts are consistent, and reconciling them is the whole question: every passage you add to raise recall also adds a distractor the generator has to survive.
Updated Sep 2026 · Grounded in real Forward Deployed Engineer interview loops and written to a senior-engineer editorial bar.
Recall@10 is 0.94 and climbing, and answer accuracy is going down. The two facts are consistent, and reconciling them is the whole question: every passage you add to raise recall also adds a distractor the generator has to survive.
The question is engineered so that every retrieval instinct is a dead end, and the candidates who keep proposing retrieval fixes after being told the gold passage is in the context are failing the actual test. What earns the round is naming an experiment that isolates the generator, and the cleanest one is a sweep over k. Watching recall climb while end-to-end accuracy peaks and falls is the moment the room decides whether you have run one of these systems or read about them.
No comments yet — be the first to share your approach.
