Replace, Don't Expand: Mitigating Context Dilution in Multi-Hop RAG via Fixed-Budget Evidence Assembly

Retrieval-Augmented Generation (RAG) systems often fail on multi-hop queries when the initial retrieval misses a bridge fact. Prior corrective approaches, such as Self-RAG, CRAG, and Adaptive-$k$, typically address this by $\mathit{\text{adding}}$ more context or pruning existing lists. However, simply expanding the context window often leads to $\mathbf{\text{context}}\mathbf{\text{dilution}}$, where distractors crowd out relevant information. We propose $\mathbf{\text{SEAL-RAG}}$, a training-free controller that adopts a $\mathbf{\text{“replace,}}\mathbf{\text{don’t}}\mathbf{\text{expand”}}$ strategy to fight context dilution under a fixed retrieval depth $k$. SEAL executes a ($\mathbf{\text{S}}$earch $\to$ $\mathbf{\text{E}}$xtract $\to$ $\mathbf{\text{A}}$ssess $\to$ $\mathbf{\text{L}}$oop) cycle: it performs on-the-fly, entity-anchored extraction to build a live $\mathit{\text{gap}}\mathit{\text{specification}}$ (missing entities/relations), triggers targeted micro-queries, and uses $\mathit{\text{entity-first}}\mathit{\text{ranking}}$ to actively swap out distractors for gap-closing evidence. We evaluate SEAL-RAG against faithful re-implementations of Basic RAG, CRAG, Self-RAG, and Adaptive-$k$ in a shared environment on $\mathbf{\text{HotpotQA}}$ and $\mathbf{\text{2WikiMultiHopQA}}$. On HotpotQA ($k=3$), SEAL improves answer correctness by $\mathbf{\text{+3–13}}\mathbf{\text{pp}}$ and evidence precision by $\mathbf{\text{+12–18}}\mathbf{\text{pp}}$ over Self-RAG. On 2WikiMultiHopQA ($k=5$), it outperforms Adaptive-$k$ by $\mathbf{\text{+8.0}}\mathbf{\text{pp}}$ in accuracy and maintains $\mathbf{\text{96\%}}$ evidence precision compared to 22% for CRAG. These gains are statistically significant ($p<0.001$). By enforcing fixed-$k$ replacement, SEAL yields a predictable cost profile while ensuring the top-$k$ slots are optimized for precision rather than mere breadth. We release our code and data at https://github.com/mosherino/SEAL-RAG.