Bridging Static and Agentic RAG for Taiwanese Historical Question Answering

Kai-Hsin Chen; Wei-Yu Chen; Xuanjun Chen; Jyh-Shing Roger Jang

cs.CLSep 19, 2026

Bridging Static and Agentic RAG for Taiwanese Historical Question Answering

Authors: Kai-Hsin Chen, Wei-Yu Chen, Xuanjun Chen, Jyh-Shing Roger Jang

Organizations: Graduate Institute of Networking and Multimedia, National Taiwan University · Dept. of Computer Science & Information Engineering, National Taiwan University · Graduate Institute of Communication Engineering, National Taiwan University

Abstract

Agentic retrieval-augmented generation (RAG) enables language models to adapt retrieval based on previously retrieved evidence, but it remains unclear whether such adaptive orchestration consistently outperforms well-designed static pipelines. We conduct a controlled comparison of agentic and static RAG for Taiwanese historical question answering, sharing the same generator and hybrid retrieval backend. Despite similar aggregate performance, the two pipelines differ on 70.83% of questions, with their advantages largely canceling out when averaged. An oracle that selects the better response per question improves the composite score by 0.2417 over the better individual pipeline, revealing substantial headroom for question-level selection. We therefore introduce a post-hoc selector that compares the two responses and their cited evidence, significantly outperforming either individual pipeline and recovering 60.34% of the oracle headroom. These results show that aggregate comparisons can obscure meaningful question-level differences between retrieval strategies, suggesting that exploiting their complementarity may be more fruitful than seeking a universally superior pipeline.

Figures & tables

Figure 1: Comparison of Static and Agentic RAG under a shared hybrid retrieval backend. Static RAG follows a predefined multi-query retrieval procedure and provides the top 15 passages to the generator, whereas Agentic RAG adaptively invokes the same retrieval backend through an MCP tool and receives the top five passages per call.

Figure 2: Post-hoc answer selection framework. Agentic and Static RAG independently generate answers with cited evidence, which are compared by an LLM-based selector to determine the preferred answer or a tie.


	Correctness	Completeness	Faithfulness	Relevance	Composite

Static RAG	4.1750	3.9833	4.5417	3.7750	4.1188
Agentic RAG	4.2917	4.1167	4.4833	3.9500	4.2104
$\Delta$ (A $-$ S)	+0.1167	+0.1333	$-0.0583$	+0.1750	+0.0917
Wilcoxon $p$	0.2927	0.4204	0.4304	0.0331	0.6543

Table 1: Generation quality of Agentic and Static RAG on the Minqing benchmark (

n=120

\Delta

denotes Agentic minus Static;

p

-values are from two-sided Wilcoxon signed-rank tests.

Figure 3: Question-level comparison of Agentic and Static RAG on Minqing: paired composite scores (left) and Agentic-minus-Static score differences (right).


Strategy	Composite	Recovery (%)
Static RAG	4.1188	-37.93
Agentic RAG	4.2104	0.00
Random selection	4.1646	-18.97
Length heuristic	4.1917	-7.76
Post-hoc selector	4.3563	60.34

Table 2: Performance of post-hoc selection strategies on the Minqing benchmark. Recovery denotes the percentage of oracle headroom recovered relative to the better individual pipeline; negative values indicate performance below this baseline.

Figure 4: Generation quality across evaluation dimensions for Static RAG, Agentic RAG, and the post-hoc selector on the Minqing benchmark.


Tie Handling	Final Score	Recovery (%)	vs. Agentic
Practical handling rules
Tie $\rightarrow$ Agentic	4.3479	56.90	+0.1375
Tie $\rightarrow$ Static	4.3646	63.79	+0.1542
Expected Value (Main)	4.3563	60.34	+0.1458
Sensitivity bounds

Table 3: Sensitivity of selector performance to tie handling.

Appendix figures & tables1 asset

Supplementary material from the paper’s appendix.

Appendix


Question Type	Agentic	Static	$p$
Fill-in-the-blank ( $n=52$ )	0.9808	0.9615	1.0000
Multiple-choice ( $n=50$ )	0.9400	0.8800	0.4531
Overall ( $n=102$ )	0.9608	0.9216	0.3438

Appendix

Table 4: Objective QA accuracy on TFCAD.

Retrieval-Augmented Generation

Retrieval-augmented generation connects language models to external knowledge so answers can use current, domain-specific evidence.

See all 792 papers

Forgotten History or Test-of-Time? Retrospect and Prospect on RAG from an IR Perspective
Aug 9, 2026Xiaoyan Zhao, Yujie Cai, Yang Zhang +2Retrieval-Augmented Generation Question Answering
National University of Singapore · Georgetown University
Dissecting Agentic RAG: A Component Ablation for Multi-Hop QA with a Local 7B Model
Jun 19, 2026Sheroz ShaikhMulti-Hop QA Cross-Encoder Reranking
Independent Researcher
One Size Does Not Fit All! Dynamic Retriever and Generator Selection for RAG
Sep 15, 2026Neeraj Anand, Payel Santra, Partha Basuchowdhuri +2Retrieval-Augmented Generation Adaptive RAG
Media and Data Science Research Lab Adobe Systems, Noida, India · IACS Kolkata, India · University of Glasgow Glasgow, United Kingdom

Aug 9, 2026J/K move · Enter open · S save

Forgotten History or Test-of-Time? Retrospect and Prospect on RAG from an IR Perspective

Xiaoyan Zhao, Yujie Cai, Yang Zhang, Grace Hui Yang+1

National University of Singapore · Georgetown University

Retrieval-Augmented Generation Question Answering Agentic RAG Information Retrieval

Open paper AskPDF arXiv

Retrieval-Augmented Generation (RAG) is widely regarded as a novel paradigm born from the limitations of large language models (LLMs)--a mechanism to ground their outputs in external knowledge. This view, however, is incomplete when considered within a broader historical context. In this paper, we argue that the core ideas underlying RAG are not new: foundational concepts such as integrating retrieval and language generation, knowledge augmentation, answer verification, and iterative query (or prompt) refinement had already been studied and instantiated in information retrieval (IR) and question answering (QA) research dating back to the early 2000s, well before the emergence of LLMs. We make this case by systematically tracing the intellectual lineage of modern RAG and Agentic RAG back to their classical IR and QA antecedents, and examining why this continuity has gone under-recognized -- a consequence of community fragmentation, shifting terminology, and the recency bias endemic to fast-moving fields. Rather than treating LLMs as the origin point of retrieval-augmented intelligence, we propose viewing them as a new interface layer atop a decades-old QA architecture. This reframing is not merely historical: by situating RAG within the longer trajectory of IR research, we surface underutilized prior work -- on user modeling, answer validation, and query refinement -- that can directly inform next-generation RAG design, reducing unintentional rediscovery and fostering genuine cross-community integration.

Abstract

Retrieval-Augmented Generation

Explore similar work

Forgotten History or Test-of-Time? Retrospect and Prospect on RAG from an IR Perspective

Dissecting Agentic RAG: A Component Ablation for Multi-Hop QA with a Local 7B Model

One Size Does Not Fit All! Dynamic Retriever and Generator Selection for RAG