This paper offers a new perspective on the consistency of retrieval-augmented QA systems, highlighting the importance of the Snapshot Compatibility Audit method. It's a must-read for those interested in QA system reliability and the nuances of corpus-induced changes.
A study reveals that a retrieval-augmented QA system can produce different answers after index expansion, despite fixed model settings. The research introduces a new method, Snapshot Compatibility Audit, to estimate excess answer churn. The study finds significant churn in both Natural Questions and TriviaQA datasets, indicating the importance of auditing compatibility alongside utility in retrieval-augmented systems.
A retrieval-augmented QA system can return different answers after an index expansion even when its requested model identifier, prompt, retrieval policy, evidence depth, rendering, and exposed generation controls are held fixed. Aggregate accuracy may hide these changes when gains and losses cancel, while ordinary generation variability makes one-shot comparisons overstate update effects. We call the hidden phenomenon accuracy-blind answer churn and introduce the \emph{Snapshot Compatibility Audit}, which estimates excess answer churn by subtracting same-snapshot repeat disagreement from cross-snapshot disagreement. We instantiate it by expanding one frozen FineWeb prefix from one to seven shards. In a preregistered 400-question Natural Questions study, normalized-exact and blinded-semantic excess churn are 6.44 and 10.25 percentage points while exact-match accuracy changes by only $-1.50$ points. A post-hoc analysis finds repeat-stable semantic flips on 40/400 questions. A separately preregistered 200-question TriviaQA study yields smaller, directionally consistent excess churn while exact-match accuracy moves in the opposite direction. An outcome-blind post-hoc 100-question subset replication with a second DeepSeek generator and serving configuration finds 8.75 pp of semantic excess churn even as exact match rises by 3.00 percentage points. Answer-level compatibility can therefore fail without a conspicuous or consistently directed utility shift. Retrieval-augmented releases should audit compatibility alongside utility.