{"id":6634,"date":"2026-07-27T13:57:44","date_gmt":"2026-07-27T13:57:44","guid":{"rendered":"https:\/\/qyrus.com\/qapi\/?p=6634"},"modified":"2026-07-29T09:04:18","modified_gmt":"2026-07-29T09:04:18","slug":"stop-losing-context-practical-fixes-for-problems-in-rag","status":"publish","type":"post","link":"https:\/\/qyrus.com\/qapi\/stop-losing-context-practical-fixes-for-problems-in-rag\/","title":{"rendered":"Stop Losing Context: Practical Fixes for\u00a0Problems\u00a0in RAG\u00a0"},"content":{"rendered":"\t\t
\n\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

You\u00a0finally\u00a0built the RAG pipeline correctly. The retriever found the right documents. The relevant information is sitting right there in the context window. The model still got the answer wrong.\u00a0<\/p>

This is the “lost in the middle” problem, and three years after\u00a0Liu et al. first documented it at Stanford and UC Berkeley<\/a>, it\u00a0remains\u00a0one of the most consistent production failures in RAG systems. The original paper showed that LLMs achieve highest accuracy when relevant information appears at the beginning or end of the context window \u2014 and that performance degrades by more than 30 percentage points when critical information is in the middle.\u00a0<\/p>

What’s\u00a0changed since 2023? Models now support million-token context windows. The research community has proposed several mitigations. And the problem still\u00a0hasn’t\u00a0gone away.\u00a0<\/p>

This piece covers\u00a0what’s\u00a0actually been\u00a0solved, what\u00a0hasn’t, and what the fixes look like in 2026.\u00a0<\/p>

What the 2025\u20132026 Research Actually Shows\u00a0<\/h2>

The\u00a0MIT 2025 follow-up research<\/a> on causal masking\u00a0\u2014 sometimes called “Found in the Middle” \u2014 finally explained the architectural mechanism behind the U-shape with precision. The core finding: position bias in transformer models\u00a0isn’t\u00a0just about distance\u00a0decay in positional encodings.\u00a0It’s\u00a0structurally embedded in causal attention itself, where each token can only attend to tokens that came before it. Tokens in the middle have a fundamentally different attention profile than tokens at the edges.\u00a0<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t

\n\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\"The\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t
\n\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

A\u00a02025 paper on layer-specific scaling of positional encodings\u00a0showed that applying layer-specific scaling to enhance middle-context attention produced an average accuracy improvement of +20% on key-value retrieval tasks and +2.7% on multi-document question answering \u2014 without retraining. This is meaningful progress.\u00a0It’s\u00a0also not yet in any production model\u00a0you’re\u00a0using today.\u00a0<\/p>

A broader\u00a02025 survey on transformer context extension\u00a0reviewed the full landscape of mitigation approaches \u2014 from efficient training schemes to inference-time context extension methods \u2014 and concluded that even models specifically trained for long-context use still\u00a0exhibit\u00a0the lost-in-the-middle problem. Extended context windows reduce the severity. They\u00a0don’t\u00a0eliminate\u00a0the underlying bias.\u00a0<\/p>

The bottom line as of mid-2026:\u00a0no production model has fully eliminated position bias. The architectural mechanism is structural. The question\u00a0isn’t\u00a0whether your model is affected \u2014\u00a0it’s\u00a0how much, and what you can do about it now.\u00a0<\/p>

The Myths Worth Debunking in 2026\u00a0<\/h4>

“Long context windows solve this”<\/b>\u00a0<\/p>

This is the most persistent misconception, and\u00a0it’s\u00a0understandable \u2014 if the model can hold 200,000 tokens, surely it can pay attention to all of them?\u00a0<\/p>

The research says otherwise. A\u00a02025 analysis by Yen et al.\u00a0tested frontier long-context models including Claude 3.5 and GPT-4o on needle-in-a-haystack tasks at 200-document contexts. The U-shape was still clearly measurable. Edge positions showed 90%+ recall. Middle positions showed a statistically significant dip \u2014 the exact\u00a0magnitude\u00a0varying by model and chunk size, but consistently present.\u00a0<\/p>

The\u00a0Atlan analysis published June 2026\u00a0summarizes the current state well: newer models have improved long-context capacity, but they still do not use every position equally. Research on context rot and effective context windows shows that performance can degrade before the advertised token limit. Larger windows still need\u00a0selection, ordering, compression, and governance.\u00a0<\/p>

Longer windows change where the problem lives and reduce its visible severity. They\u00a0don’t\u00a0make context position irrelevant.\u00a0<\/p>

“New attention calibration techniques have solved this in production”<\/b>\u00a0<\/p>

Techniques like Multi-scale Positional Encoding (Ms-PoE) and attention calibration \u2014 covered in the\u00a0“Found in the Middle” paper\u00a0\u2014 do reduce the bias in controlled research settings. The +20% accuracy improvement from layer-specific scaling mentioned above is real and reproducible.\u00a0<\/p>

But as of 2026, none of these techniques are deployed in the production LLMs most teams are using.\u00a0They’re\u00a0research results, not product features. The gap between “this works in a research paper” and “this is live in the OpenAI or Anthropic API you’re calling” is significant. Applying academic mitigations to your RAG pipeline means waiting for providers to adopt them or fine-tuning your own models \u2014 neither of which is a near-term\u00a0option\u00a0for most teams.\u00a0<\/p>

“Agentic frameworks have made this irrelevant”<\/b>\u00a0<\/p>

This argument has some merit, and\u00a0it’s\u00a0worth taking seriously. A\u00a02025 analysis in Towards AI\u00a0makes the case that instead of a linear retrieve-rerank-generate flow, agentic AI\u00a0operates\u00a0in a reason-act-observe loop. An agent can recognize when its attentional focus is cluttered, decompose the question, retrieve different sub-contexts iteratively, and assemble an answer across multiple passes rather than trying to attend to everything in one long prompt.\u00a0<\/p>

That’s\u00a0accurate\u00a0\u2014 and\u00a0it’s\u00a0genuinely one of the more promising architectural directions for long-context problems. But it applies to agentic systems specifically. Most RAG applications in production in 2026 are not agentic.\u00a0They’re\u00a0single-pass retrieve-then-generate pipelines. For those systems, position bias\u00a0remains\u00a0a live problem that no architectural shift has eliminated.\u00a0<\/p>

“Randomizing document order fixes the bias”<\/b>\u00a0<\/p>

No. The bias is positional, not\u00a0content-based. Randomizing order randomly assigns relevant documents to middle positions, which on average makes things worse. You want deliberate ordering based on relevance \u2014 the opposite of random.\u00a0<\/p>

The Fixes: What Actually Works in 2026\u00a0<\/h2>

Fix 1: Strategic document reordering \u2014 still the highest-leverage, lowest-effort fix\u00a0<\/h4>

Nothing has replaced this. The research consensus from\u00a0Liu et al.’s original paper\u00a0through the 2025 follow-on work\u00a0remains\u00a0consistent: placing the most relevant document first and the second most relevant document last, with supporting context filling the middle, recovers\u00a0a significant portion\u00a0of accuracy lost to position bias.\u00a0<\/p>

LangChain\u00a0implements this as\u00a0LongContextReorder. The underlying logic: the model attends most strongly to the beginning and end of the context window. Give it what it needs most at both edges.\u00a0<\/p>

Implementation cost: a few lines of sorting logic in your context assembly step. No architecture changes, no re-indexing, no\u00a0additional\u00a0latency. This is still the first fix to implement because the cost is near zero and the impact is measurable\u00a0immediately.\u00a0<\/p>

What’s\u00a0new in 2026: the\u00a0Position Engineering paper (He et al., 2024)\u00a0showed that systematic positional information manipulation \u2014 not\u00a0just simple\u00a0first\/last ordering but deliberate placement of different information types at\u00a0different positions\u00a0\u2014 can further boost performance. For teams that want to go beyond simple relevance-based ordering, this is worth reading.\u00a0<\/p>

Fix 2: Reduce retrieved context volume \u2014 the case has gotten stronger\u00a0<\/h4>

The argument for retrieving fewer documents has only gotten stronger as context windows have gotten longer. The instinct to “use the full context window” leads teams to pass 50 or 100 chunks when 3 to 5 highly relevant ones would produce better results.\u00a0<\/p>

The\u00a0Towards AI analysis\u00a0frames this well: the best enterprise fix is governed context delivery \u2014 fewer, higher-signal context objects, ranked by relevance, placed intentionally. Prompt tactics help, but durable improvement comes from controlling what goes into the context before the model starts reasoning.\u00a0<\/p>

Shi et al.’s work on distracting context\u00a0showed that irrelevant-but-plausible passages in the context window significantly degrade model performance even when the correct answer is also present. Every\u00a0additional\u00a0chunk you pass is a potential source of distraction \u2014 and a potential source of middle-position penalty if it pushes relevant content away from the edges.\u00a0<\/p>

Practical starting point: if\u00a0you’re\u00a0currently passing 15\u201320 chunks, cut to 5 and measure accuracy. Most teams see improvement, not degradation.\u00a0<\/p>

Fix 3: Two-stage retrieval with cross-encoder reranking \u2014 now table stakes for production\u00a0<\/h4>

What was an advanced technique in 2023 is now a standard production pattern. Two-stage retrieval works as follows: use your dense retriever to pull a broad candidate set of 20\u201330 documents, then run a cross-encoder\u00a0reranker\u00a0to re-score and\u00a0rerank\u00a0based on precise relevance to the query.\u00a0<\/p>

Cross-encoders attend to both the query and the document simultaneously, which produces significantly more\u00a0accurate\u00a0rankings than first-stage semantic similarity.\u00a0Cohere’s Rerank API\u00a0and\u00a0the\u00a0ms-marco-MiniLM models on Hugging Face\u00a0are the most widely used options.\u00a0<\/p>

What’s\u00a0new in 2026:\u00a0RankRAG (Yu et al., 2024)\u00a0proposed unifying context ranking with RAG generation, training the LLM to jointly rank and generate rather than treating them as separate steps. This showed\u00a0state-of-the-art\u00a0results on several benchmarks.\u00a0It’s\u00a0not widely deployed in production yet, but\u00a0it’s\u00a0the direction the field is moving \u2014 from separate retrieval and generation to unified ranking-generation.\u00a0<\/p>

Then apply Fix 1 to the reranked results \u2014 place the top result first, second-most-relevant last, fill the middle with the rest. The combination of reranking and strategic ordering is the most reliable pattern for production systems in 2026.\u00a0<\/p>

Fix 4: Contextual retrieval and smarter chunking \u2014 the re-indexing investment worth making\u00a0<\/h4>

Anthropic’s contextual retrieval approach\u00a0(released late 2024) adds a short contextual summary to each chunk before embedding. The summary tells the embedding model where in the document the chunk appears and what broader topic it relates to, which significantly improves retrieval precision.\u00a0<\/p>

The\u00a0AutoRAG paper (Choi et al., 2024)\u00a0benchmarked chunking strategy variations and found recall differences of up to 9% between fixed-size and semantic chunking on the same corpus.\u00a0<\/p>

What’s\u00a0new in 2026:\u00a0LIFT (Long Input Fine-Tuning, Mao et al., 2025)\u00a0proposes improving long-context understanding through fine-tuning on long-input examples rather than improving chunking. This is a model-level fix rather than a pipeline-level fix \u2014 relevant for teams fine-tuning their own models, less relevant for teams using commercial API endpoints.\u00a0<\/p>

For most teams using commercial models: invest in chunking improvements before model-level fixes. The pipeline-level gains from contextual retrieval are more\u00a0immediately\u00a0actionable.\u00a0<\/p>

Fix 5: Multi-query retrieval for complex questions \u2014 increasingly automated\u00a0<\/h4>

Multi-query decomposition is more accessible in 2026 than it was two years ago.\u00a0LangChain’s MultiQueryRetriever\u00a0generates sub-queries from the original query automatically. What used to require custom implementation is now a one-line setup.\u00a0<\/p>

The underlying value\u00a0hasn’t\u00a0changed: for queries that require information from multiple sections of a document or multiple documents, single-query retrieval systematically misses relevant content. Decomposing into sub-queries retrieves more complete context.\u00a0<\/p>

The trade-off\u00a0remains: at least 2\u00d7 the retrieval latency. Use this for query categories that consistently fail single-query retrieval, not universally.\u00a0<\/p>

Fix 6: Agentic context management \u2014 the emerging fix for complex multi-document tasks\u00a0<\/h4>

This is the most\u00a0significant change\u00a0from the 2023 version of this problem. For complex, multi-hop tasks \u2014 where the answer requires reasoning across multiple documents or sources \u2014 agentic approaches that iteratively retrieve and reason have shown\u00a0substantially better\u00a0results than single-pass RAG.\u00a0<\/p>

The\u00a0PAM QA approach (Never Lost in the Middle, 2024)\u00a0trained models to use position-agnostic decomposition: rather than attending to one long context, models learn to decompose queries, retrieve sub-contexts, and assemble answers across multiple passes. The results on long-context QA benchmarks were significantly better than single-pass approaches.\u00a0<\/p>

For production systems: this\u00a0doesn’t\u00a0mean you need to rebuild your RAG pipeline as an agent overnight. It means that for the subset of queries in your system where single-pass retrieval consistently fails \u2014 usually complex multi-hop questions \u2014 a targeted agentic retrieval loop is now a practical option rather than a research experiment.\u00a0<\/p>

How to Detect and Measure the Problem in Your System\u00a0<\/h2>

The detection\u00a0methodology\u00a0is straightforward and worth running before implementing any fixes. Place a known relevant document at position 1 in your context window and measure accuracy. Repeat with the same document at positions 5, 10, 15, and 20. Plot the accuracy curve.\u00a0<\/p>

If you see the U-shape \u2014 high accuracy at positions 1 and 20, degraded accuracy at positions 5\u201315 \u2014 you have quantified the problem. The depth of the dip tells you how\u00a0much accuracy\u00a0you’re\u00a0losing from position bias and how much the fixes above are likely to recover.\u00a0<\/p>

The\u00a0Yen et al. (2025) long-context analysis\u00a0provides a replication framework you can adapt to your specific retrieval configuration and query distribution.\u00a0<\/p>

What to add to your evaluation pipeline:<\/b>\u00a0Track context recall by chunk position over time. Did implementing strategic reordering flatten the U-shape? Did the cross-encoder\u00a0reranker\u00a0reduce the number of relevant chunks landing in middle positions? These are the metrics that tell you the fixes are working \u2014 not just aggregate faithfulness or\u00a0groundedness\u00a0scores, which can look fine even when position bias is degrading specific query categories.\u00a0<\/p>

This is where\u00a0qAPI’s\u00a0RAG evaluation module provides direct value. The chunking strategy A\/B comparison lets you measure the impact of a retrieval configuration change without re-indexing production \u2014 run both configurations against your eval set and compare recall at each position.\u00a0\u00a0<\/p>

Index freshness monitoring catches the related failure mode where stale retrieval produces confident answers about outdated information. And because\u00a0qAPI\u00a0sits alongside your API testing, when a retrieval endpoint changes its response structure and causes a quality drop, that surfaces in the same report as your RAG metrics rather than requiring a separate investigation.\u00a0<\/p>

The Implementation Sequence for 2026\u00a0<\/h2>

The right order\u00a0hasn’t\u00a0changed, but the context around each step has:\u00a0<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t

\n\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\"week\"\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t
\n\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Week 1:<\/b>\u00a0Run the position-detection test. Quantify the U-shape on your actual queries. If the accuracy drop at middle positions is under 5%, other optimizations are higher priority. If\u00a0it’s\u00a0over 10%, Fix 1 and Fix 2 are urgent.\u00a0<\/p>

Week 2:<\/b>\u00a0Implement strategic document reordering (Fix 1). A few lines of sorting logic. Measure the U-shape again. For most teams, this alone recovers 50\u201370% of the accuracy lost to position bias.\u00a0<\/p>

Week 3:<\/b>\u00a0Reduce retrieved context to 3\u20135 highly relevant chunks (Fix 2). Measure. Most teams see further improvement.\u00a0<\/p>

Week 4+:<\/b>\u00a0Add cross-encoder reranking (Fix 3) if the residual gap justifies the latency cost. Pairwise comparison with and without reranking on your eval set should make this decision straightforward.\u00a0<\/p>

Ongoing:<\/b>\u00a0Invest in contextual retrieval and chunking improvements (Fix 4) when you have operational bandwidth to re-index. Evaluate agentic approaches (Fix 6) for specific query categories where single-pass retrieval consistently fails.\u00a0<\/p>

At each step, re-run the position-detection test. You want to see the U-shape flattening.\u00a0<\/p>

The Takeaway\u00a0<\/h2>

The lost-in-the-middle problem looks different in 2026 than it did in 2023. The research community understands the architectural cause with much more precision. Mitigation techniques like attention calibration and layer-specific positional scaling show real promise. Agentic frameworks offer a structural alternative for complex multi-hop tasks.\u00a0<\/p>

What\u00a0hasn’t\u00a0changed: no production model\u00a0you’re\u00a0deploying against has fully\u00a0eliminated\u00a0position bias. The U-shape is still there. The accuracy drop at middle positions is still measurable and still consequential.\u00a0<\/p>

The teams that know about this problem and measure it have a clear, actionable path to improving answer accuracy without touching the model. The teams that assume million-token context windows solved it are flying blind on an accuracy deficit they\u00a0don’t\u00a0know exists.\u00a0<\/p>

Run the detection test. Quantify the dip. Fix it layer by layer.\u00a0It will work\u00a0with\u00a0qAPI.\u00a0<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t","protected":false},"excerpt":{"rendered":"

You\u00a0finally\u00a0built the RAG pipeline correctly. The retriever found the right documents. The relevant information is sitting right there in the context window. The model still got the answer wrong.\u00a0 This is the “lost in the middle” problem, and three years after\u00a0Liu et al. first documented it at Stanford and UC Berkeley, it\u00a0remains\u00a0one of the most…<\/p>\n","protected":false},"author":9,"featured_media":6658,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"content-type":"","inline_featured_image":false,"footnotes":""},"categories":[17,10],"tags":[],"class_list":["post-6634","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-blog","category-resources"],"acf":[],"yoast_head":"\nStop Losing Context: Practical Fixes for\u00a0Problems\u00a0in RAG\u00a0 - qAPI<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/qyrus.com\/qapi\/stop-losing-context-practical-fixes-for-problems-in-rag\/\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Stop Losing Context: Practical Fixes for\u00a0Problems\u00a0in RAG\u00a0 - qAPI\" \/>\n<meta property=\"og:description\" content=\"You\u00a0finally\u00a0built the RAG pipeline correctly. The retriever found the right documents. The relevant information is sitting right there in the context window. The model still got the answer wrong.\u00a0 This is the “lost in the middle” problem, and three years after\u00a0Liu et al. first documented it at Stanford and UC Berkeley, it\u00a0remains\u00a0one of the most...\" \/>\n<meta property=\"og:url\" content=\"https:\/\/qyrus.com\/qapi\/stop-losing-context-practical-fixes-for-problems-in-rag\/\" \/>\n<meta property=\"og:site_name\" content=\"qAPI\" \/>\n<meta property=\"article:publisher\" content=\"https:\/\/www.facebook.com\/profile.php?id=61571758838201\" \/>\n<meta property=\"article:published_time\" content=\"2026-07-27T13:57:44+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2026-07-29T09:04:18+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/qyrus.com\/qapi\/wp-content\/uploads\/2026\/07\/The-U-shaped-attention-curve-1.png\" \/>\n\t<meta property=\"og:image:width\" content=\"4455\" \/>\n\t<meta property=\"og:image:height\" content=\"2271\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/png\" \/>\n<meta name=\"author\" content=\"R Varun\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:creator\" content=\"@testwithqapi\" \/>\n<meta name=\"twitter:site\" content=\"@testwithqapi\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"R Varun\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"12 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\/\/qyrus.com\/qapi\/stop-losing-context-practical-fixes-for-problems-in-rag\/#article\",\"isPartOf\":{\"@id\":\"https:\/\/qyrus.com\/qapi\/stop-losing-context-practical-fixes-for-problems-in-rag\/\"},\"author\":{\"name\":\"R Varun\",\"@id\":\"https:\/\/qyrus.com\/qapi\/#\/schema\/person\/33d511c123d8cd9b9e9dc5ee9e0e5c90\"},\"headline\":\"Stop Losing Context: Practical Fixes for\u00a0Problems\u00a0in RAG\u00a0\",\"datePublished\":\"2026-07-27T13:57:44+00:00\",\"dateModified\":\"2026-07-29T09:04:18+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\/\/qyrus.com\/qapi\/stop-losing-context-practical-fixes-for-problems-in-rag\/\"},\"wordCount\":2353,\"publisher\":{\"@id\":\"https:\/\/qyrus.com\/qapi\/#organization\"},\"image\":{\"@id\":\"https:\/\/qyrus.com\/qapi\/stop-losing-context-practical-fixes-for-problems-in-rag\/#primaryimage\"},\"thumbnailUrl\":\"https:\/\/qyrus.com\/qapi\/wp-content\/uploads\/2026\/07\/thumbnail-f-1-1.png\",\"articleSection\":[\"Blog\",\"Resources\"],\"inLanguage\":\"en-US\"},{\"@type\":\"WebPage\",\"@id\":\"https:\/\/qyrus.com\/qapi\/stop-losing-context-practical-fixes-for-problems-in-rag\/\",\"url\":\"https:\/\/qyrus.com\/qapi\/stop-losing-context-practical-fixes-for-problems-in-rag\/\",\"name\":\"Stop Losing Context: Practical Fixes for\u00a0Problems\u00a0in RAG\u00a0 - qAPI\",\"isPartOf\":{\"@id\":\"https:\/\/qyrus.com\/qapi\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\/\/qyrus.com\/qapi\/stop-losing-context-practical-fixes-for-problems-in-rag\/#primaryimage\"},\"image\":{\"@id\":\"https:\/\/qyrus.com\/qapi\/stop-losing-context-practical-fixes-for-problems-in-rag\/#primaryimage\"},\"thumbnailUrl\":\"https:\/\/qyrus.com\/qapi\/wp-content\/uploads\/2026\/07\/thumbnail-f-1-1.png\",\"datePublished\":\"2026-07-27T13:57:44+00:00\",\"dateModified\":\"2026-07-29T09:04:18+00:00\",\"breadcrumb\":{\"@id\":\"https:\/\/qyrus.com\/qapi\/stop-losing-context-practical-fixes-for-problems-in-rag\/#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/qyrus.com\/qapi\/stop-losing-context-practical-fixes-for-problems-in-rag\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\/\/qyrus.com\/qapi\/stop-losing-context-practical-fixes-for-problems-in-rag\/#primaryimage\",\"url\":\"https:\/\/qyrus.com\/qapi\/wp-content\/uploads\/2026\/07\/thumbnail-f-1-1.png\",\"contentUrl\":\"https:\/\/qyrus.com\/qapi\/wp-content\/uploads\/2026\/07\/thumbnail-f-1-1.png\",\"width\":1280,\"height\":720,\"caption\":\"Stop losing Context: Practical fixes for Problems in RAG\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/qyrus.com\/qapi\/stop-losing-context-practical-fixes-for-problems-in-rag\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\/\/qyrus.com\/qapi\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Stop Losing Context: Practical Fixes for\u00a0Problems\u00a0in RAG\u00a0\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/qyrus.com\/qapi\/#website\",\"url\":\"https:\/\/qyrus.com\/qapi\/\",\"name\":\"qAPI\",\"description\":\"\",\"publisher\":{\"@id\":\"https:\/\/qyrus.com\/qapi\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/qyrus.com\/qapi\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"},{\"@type\":\"Organization\",\"@id\":\"https:\/\/qyrus.com\/qapi\/#organization\",\"name\":\"qAPI\",\"url\":\"https:\/\/qyrus.com\/qapi\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\/\/qyrus.com\/qapi\/#\/schema\/logo\/image\/\",\"url\":\"https:\/\/qyrus.com\/qapi\/wp-content\/uploads\/2025\/02\/qAPI-Youtube-DP-98-x-98.png\",\"contentUrl\":\"https:\/\/qyrus.com\/qapi\/wp-content\/uploads\/2025\/02\/qAPI-Youtube-DP-98-x-98.png\",\"width\":409,\"height\":409,\"caption\":\"qAPI\"},\"image\":{\"@id\":\"https:\/\/qyrus.com\/qapi\/#\/schema\/logo\/image\/\"},\"sameAs\":[\"https:\/\/www.facebook.com\/profile.php?id=61571758838201\",\"https:\/\/x.com\/testwithqapi\",\"https:\/\/www.linkedin.com\/company\/testwithqapi\/?viewAsMember=true\",\"https:\/\/www.instagram.com\/testwithqapi\/\",\"https:\/\/www.youtube.com\/@testwithqapi\"]},{\"@type\":\"Person\",\"@id\":\"https:\/\/qyrus.com\/qapi\/#\/schema\/person\/33d511c123d8cd9b9e9dc5ee9e0e5c90\",\"name\":\"R Varun\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\/\/qyrus.com\/qapi\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/secure.gravatar.com\/avatar\/62344175a96575918f882055650fdf8d3c6c18886a2248ce250f7cd05e3ca866?s=96&d=mm&r=g\",\"contentUrl\":\"https:\/\/secure.gravatar.com\/avatar\/62344175a96575918f882055650fdf8d3c6c18886a2248ce250f7cd05e3ca866?s=96&d=mm&r=g\",\"caption\":\"R Varun\"},\"url\":\"https:\/\/qyrus.com\/qapi\/author\/rvarunqyrus-com\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Stop Losing Context: Practical Fixes for\u00a0Problems\u00a0in RAG\u00a0 - qAPI","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/qyrus.com\/qapi\/stop-losing-context-practical-fixes-for-problems-in-rag\/","og_locale":"en_US","og_type":"article","og_title":"Stop Losing Context: Practical Fixes for\u00a0Problems\u00a0in RAG\u00a0 - qAPI","og_description":"You\u00a0finally\u00a0built the RAG pipeline correctly. The retriever found the right documents. The relevant information is sitting right there in the context window. The model still got the answer wrong.\u00a0 This is the “lost in the middle” problem, and three years after\u00a0Liu et al. first documented it at Stanford and UC Berkeley, it\u00a0remains\u00a0one of the most...","og_url":"https:\/\/qyrus.com\/qapi\/stop-losing-context-practical-fixes-for-problems-in-rag\/","og_site_name":"qAPI","article_publisher":"https:\/\/www.facebook.com\/profile.php?id=61571758838201","article_published_time":"2026-07-27T13:57:44+00:00","article_modified_time":"2026-07-29T09:04:18+00:00","og_image":[{"width":4455,"height":2271,"url":"https:\/\/qyrus.com\/qapi\/wp-content\/uploads\/2026\/07\/The-U-shaped-attention-curve-1.png","type":"image\/png"}],"author":"R Varun","twitter_card":"summary_large_image","twitter_creator":"@testwithqapi","twitter_site":"@testwithqapi","twitter_misc":{"Written by":"R Varun","Est. reading time":"12 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/qyrus.com\/qapi\/stop-losing-context-practical-fixes-for-problems-in-rag\/#article","isPartOf":{"@id":"https:\/\/qyrus.com\/qapi\/stop-losing-context-practical-fixes-for-problems-in-rag\/"},"author":{"name":"R Varun","@id":"https:\/\/qyrus.com\/qapi\/#\/schema\/person\/33d511c123d8cd9b9e9dc5ee9e0e5c90"},"headline":"Stop Losing Context: Practical Fixes for\u00a0Problems\u00a0in RAG\u00a0","datePublished":"2026-07-27T13:57:44+00:00","dateModified":"2026-07-29T09:04:18+00:00","mainEntityOfPage":{"@id":"https:\/\/qyrus.com\/qapi\/stop-losing-context-practical-fixes-for-problems-in-rag\/"},"wordCount":2353,"publisher":{"@id":"https:\/\/qyrus.com\/qapi\/#organization"},"image":{"@id":"https:\/\/qyrus.com\/qapi\/stop-losing-context-practical-fixes-for-problems-in-rag\/#primaryimage"},"thumbnailUrl":"https:\/\/qyrus.com\/qapi\/wp-content\/uploads\/2026\/07\/thumbnail-f-1-1.png","articleSection":["Blog","Resources"],"inLanguage":"en-US"},{"@type":"WebPage","@id":"https:\/\/qyrus.com\/qapi\/stop-losing-context-practical-fixes-for-problems-in-rag\/","url":"https:\/\/qyrus.com\/qapi\/stop-losing-context-practical-fixes-for-problems-in-rag\/","name":"Stop Losing Context: Practical Fixes for\u00a0Problems\u00a0in RAG\u00a0 - qAPI","isPartOf":{"@id":"https:\/\/qyrus.com\/qapi\/#website"},"primaryImageOfPage":{"@id":"https:\/\/qyrus.com\/qapi\/stop-losing-context-practical-fixes-for-problems-in-rag\/#primaryimage"},"image":{"@id":"https:\/\/qyrus.com\/qapi\/stop-losing-context-practical-fixes-for-problems-in-rag\/#primaryimage"},"thumbnailUrl":"https:\/\/qyrus.com\/qapi\/wp-content\/uploads\/2026\/07\/thumbnail-f-1-1.png","datePublished":"2026-07-27T13:57:44+00:00","dateModified":"2026-07-29T09:04:18+00:00","breadcrumb":{"@id":"https:\/\/qyrus.com\/qapi\/stop-losing-context-practical-fixes-for-problems-in-rag\/#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/qyrus.com\/qapi\/stop-losing-context-practical-fixes-for-problems-in-rag\/"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/qyrus.com\/qapi\/stop-losing-context-practical-fixes-for-problems-in-rag\/#primaryimage","url":"https:\/\/qyrus.com\/qapi\/wp-content\/uploads\/2026\/07\/thumbnail-f-1-1.png","contentUrl":"https:\/\/qyrus.com\/qapi\/wp-content\/uploads\/2026\/07\/thumbnail-f-1-1.png","width":1280,"height":720,"caption":"Stop losing Context: Practical fixes for Problems in RAG"},{"@type":"BreadcrumbList","@id":"https:\/\/qyrus.com\/qapi\/stop-losing-context-practical-fixes-for-problems-in-rag\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/qyrus.com\/qapi\/"},{"@type":"ListItem","position":2,"name":"Stop Losing Context: Practical Fixes for\u00a0Problems\u00a0in RAG\u00a0"}]},{"@type":"WebSite","@id":"https:\/\/qyrus.com\/qapi\/#website","url":"https:\/\/qyrus.com\/qapi\/","name":"qAPI","description":"","publisher":{"@id":"https:\/\/qyrus.com\/qapi\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/qyrus.com\/qapi\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Organization","@id":"https:\/\/qyrus.com\/qapi\/#organization","name":"qAPI","url":"https:\/\/qyrus.com\/qapi\/","logo":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/qyrus.com\/qapi\/#\/schema\/logo\/image\/","url":"https:\/\/qyrus.com\/qapi\/wp-content\/uploads\/2025\/02\/qAPI-Youtube-DP-98-x-98.png","contentUrl":"https:\/\/qyrus.com\/qapi\/wp-content\/uploads\/2025\/02\/qAPI-Youtube-DP-98-x-98.png","width":409,"height":409,"caption":"qAPI"},"image":{"@id":"https:\/\/qyrus.com\/qapi\/#\/schema\/logo\/image\/"},"sameAs":["https:\/\/www.facebook.com\/profile.php?id=61571758838201","https:\/\/x.com\/testwithqapi","https:\/\/www.linkedin.com\/company\/testwithqapi\/?viewAsMember=true","https:\/\/www.instagram.com\/testwithqapi\/","https:\/\/www.youtube.com\/@testwithqapi"]},{"@type":"Person","@id":"https:\/\/qyrus.com\/qapi\/#\/schema\/person\/33d511c123d8cd9b9e9dc5ee9e0e5c90","name":"R Varun","image":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/qyrus.com\/qapi\/#\/schema\/person\/image\/","url":"https:\/\/secure.gravatar.com\/avatar\/62344175a96575918f882055650fdf8d3c6c18886a2248ce250f7cd05e3ca866?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/62344175a96575918f882055650fdf8d3c6c18886a2248ce250f7cd05e3ca866?s=96&d=mm&r=g","caption":"R Varun"},"url":"https:\/\/qyrus.com\/qapi\/author\/rvarunqyrus-com\/"}]}},"_links":{"self":[{"href":"https:\/\/qyrus.com\/qapi\/wp-json\/wp\/v2\/posts\/6634","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/qyrus.com\/qapi\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/qyrus.com\/qapi\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/qyrus.com\/qapi\/wp-json\/wp\/v2\/users\/9"}],"replies":[{"embeddable":true,"href":"https:\/\/qyrus.com\/qapi\/wp-json\/wp\/v2\/comments?post=6634"}],"version-history":[{"count":4,"href":"https:\/\/qyrus.com\/qapi\/wp-json\/wp\/v2\/posts\/6634\/revisions"}],"predecessor-version":[{"id":6643,"href":"https:\/\/qyrus.com\/qapi\/wp-json\/wp\/v2\/posts\/6634\/revisions\/6643"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/qyrus.com\/qapi\/wp-json\/wp\/v2\/media\/6658"}],"wp:attachment":[{"href":"https:\/\/qyrus.com\/qapi\/wp-json\/wp\/v2\/media?parent=6634"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/qyrus.com\/qapi\/wp-json\/wp\/v2\/categories?post=6634"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/qyrus.com\/qapi\/wp-json\/wp\/v2\/tags?post=6634"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}