Everyone computes the stale fraction as a share of the corpus, gets 0.03%, and moves on. Weighted by queries rather than documents it is 1.4%, roughly fifty times larger, because the documents that changed during the rebuild are the ones people are asking about.
Updated Aug 2026 · Grounded in real Forward Deployed Engineer interview loops and written to a senior-engineer editorial bar.
Everyone computes the stale fraction as a share of the corpus, gets 0.03%, and moves on. Weighted by queries rather than documents it is 1.4%, roughly fifty times larger, because the documents that changed during the rebuild are the ones people are asking about.
Watch for how the candidate prices staleness. Almost everyone computes it as a share of the corpus, gets a number like 0.03%, and concludes it does not matter. The right denominator is queries, not documents, because the documents that changed during the rebuild are exactly the ones people are asking about. Measured against a recency-weighted query distribution the same window is 1.4% of traffic, close to fifty times the corpus share, and that is the number that decides whether the second index gets funded.
No comments yet — be the first to share your approach.
