More Context, Same Budget: Dual-Bounded Relational Recall Beyond Top-K Retrieval

Thomson D. Nguy

cs.IRAug 19, 2026

More Context, Same Budget: Dual-Bounded Relational Recall Beyond Top-K Retrieval

Authors: Thomson D. Nguy

Organizations: Radiant Institute for Manifold Studies (RIMS)

Abstract

More context does not require a larger retrieval budget. Under the same ceiling, a retrieval system can recover more of the evidence a question requires by following relationships between evidence that flat top-k ranking leaves behind. We test that proposition with Dual-Bounded Relational Recall (DBRR), which allocates a fixed retrieval budget between relevance-selected seeds and bounded graph-adjacent context, against matched flat top-k retrieval using the same relevance-ranking stage and the same maximum number of retrieval units and tokens. The outcome is complete recovery of the official HotpotQA supporting-evidence set for each question. Across 7,405 FullWiki questions, the Primary DBRR allocation increased complete supporting-evidence recovery by 23.8 percentage points over its matched flat baseline (paired risk difference 0.2377; question-level bootstrap 95% interval 0.2269 to 0.2489). It improved 1,952 questions, tied on 5,261, and harmed 192. Bridge questions drove the effect, with a 28.7-point increase; comparison questions showed a smaller 4.2-point difference. In a prespecified, evaluation-only diagnostic population, real relationships also outperformed random-neighbor and degree-preserving shuffled-graph controls. The result is straightforward: under the same context budget, complete-evidence retrieval depends not only on which items rank highest, but on how context is allocated around them. Relational allocation recovered complete evidence sets that flat top-k retrieval left incomplete.

Figures & tables

Figure 1: Same ceiling, different allocation. A common relevance-ranked candidate list feeds two routes under matched maximum object and token ceilings. The flat route continues down the ranking. The relational route reserves part of the allowance for bounded context adjacent to selected evidence. Both terminate at the same allowed context ceiling; held allocation mechanics are not shown.

Figure 2: Primary complete-support recovery contrast. The first-read message is that a matched relational allocation produced a large net gain, while 192 questions still moved in the wrong direction. The figure reports n=7,405 , absolute recovery rates 58.3% and 34.5%, paired risk difference 0.2377, 95% interval [0.2269, 0.2489], the +0.03 practical threshold, and improve/tie/harm counts 1,952/5,261/192.

Figure 3: Prespecified question-type strata. The relational advantage is concentrated in bridge questions. Bridge and comparison estimates appear on the same probability-difference axis with their intervals and adjusted dispositions; the comparison contrast is labeled INCONCLUSIVE under the practical rule.

Figure 4: Full-population controls and conditional relationship diagnostic. Panel A reports real-versus-random and real-versus-shuffled differences across all 7,405 questions. Panel B reports the prespecified gold-derived, evaluation-only conditional diagnostic ( n=3,562 ). Panel B is not a deployment estimate and does not replace Figure 2.

Public allocation	n	Paired risk difference	95% interval	Improve / tie / harm	Disposition
Primary	7,405	0.237677	[0.226874, 0.248886]	1,952 / 5,261 / 192	WIN
Sensitivity A	7,405	0.221202	[0.209858, 0.232681]	1,930 / 5,183 / 292	WIN
Sensitivity B	7,405	0.205537	[0.194058, 0.217016]	1,871 / 5,185 / 349	WIN

Table 5

Evidence Retrieval

Evidence retrieval finds and organizes source-grounded information to answer questions, verify claims, and support decisions.

See all 161 papers

What Survives Into Context: A Diagnostic for Budget-Constrained Multi-Hop RAG and When Submodular Evidence Packing Improves It
Jul 1, 2026Ananto Nayan BalaMulti-Hop QA Long-Context QA
Ahsanullah University of Science and Technology
EviReform: Evidence-Guided Query Reformulation for Multi-Hop Graph Retrieval
Aug 13, 2026Xinlong Xu, Yoshua Y. LiMulti-Hop QA Graph-Based Retrieval
Nanjing University of Information Science and Technology · Meituan
Evidence Interfaces Shape How Retrieval-Augmented Readers Use Support
Jul 19, 2026Junchi Liao, Jiawen Deng, Fuji RenMulti-Hop QA Learning to Rank

Jul 1, 2026J/K move · Enter open · S save

What Survives Into Context: A Diagnostic for Budget-Constrained Multi-Hop RAG and When Submodular Evidence Packing Improves It

Ananto Nayan Bala

Ahsanullah University of Science and Technology

Multi-Hop QA Long-Context QA RAG Evaluation

Open paper AskPDF arXiv

Retrieval-augmented generation (RAG) under a fixed reader-context budget forces a selection problem: of the evidence retrieved, only a fraction can be shown to the reader. We argue that document recall -- the standard retrieval metric -- is the wrong quantity to optimize in this regime, and we make two contributions. First, as a general contribution, we introduce answer-in-context, a diagnostic that measures whether a gold answer survives as a contiguous span in the packed reader context (not the retrieved set). It predicts answer F1 better than recall (r=0.39-0.55 vs. about 0.31), separates answer quality roughly five-fold (0.60 vs. 0.12 on HotpotQA), and carries information beyond retrieval: it adds Delta R squared=0.17 over recall and shows a 4.6x EM gap even among questions where all gold was retrieved. We also confirm it interventionally: on 2WikiMultiHopQA a packing change that raises coverage but not answer-in-context yields no accuracy gain. Second, as a conditional contribution, we cast reader-context construction as budgeted monotone submodular maximization and build a packer that jointly optimizes relevance, query coverage, representativeness, and diversity. On HotpotQA with a 160-token budget and a 3B reader it beats a strong focused heuristic, MMR, and naive packing -- by up to +5.1 F1 at equal-or-lower token cost, across three seeds. Crucially, we map the scope of this win honestly: it requires the conjunction of (i) multi-hop complementary structure, (ii) retrieval that surfaces the evidence, (iii) a binding but not extreme budget, and (iv) a reader weak enough that evidence density, not reading capacity, is the bottleneck. A quantization-controlled reader-scale ladder (3B to 7B to 14B) shows the edge over the heuristic is absorbed by 7B and significantly reverses by 14B, while the diagnostic explains every boundary with a single variable.

Abstract

Evidence Retrieval

Explore similar work

What Survives Into Context: A Diagnostic for Budget-Constrained Multi-Hop RAG and When Submodular Evidence Packing Improves It

EviReform: Evidence-Guided Query Reformulation for Multi-Hop Graph Retrieval

Evidence Interfaces Shape How Retrieval-Augmented Readers Use Support