CDF-RAG: Causal Dynamic Feedback for Adaptive Retrieval-Augmented Generation

Amir M. Rahmani; Elahe Khatibi; Ziyu Wang

REVIEW 5 major objections 5 minor 2 cited by

CDF-RAG: Causal Dynamic Feedback for Adaptive Retrieval-Augmented Generation

T0 review · 5 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A dynamic feedback loop that refines queries with reinforcement learning, retrieves over a causal graph as well as text passages, and verifies generated answers against causal paths yields the highest accuracy across four…

desk verdict A plausible RAG integration whose central accuracy claim is undermined by a missing train/test split and a circular causal metric. read the letter →

arxiv 2504.12560 v1 pith:TPHHUFJO submitted 2025-04-17 cs.CL

Elahe Khatibi , Ziyu Wang , Amir M. Rahmani This is my paper · ORCID

classification cs.CL

keywords retrieval-augmentedgenerationcausalreasoningqueryrefinementreinforcementlearninggraphretrievalhallucinationdetectionquestionanswering

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Retrieval-augmented generation often retrieves topically relevant but causally shallow evidence. CDF-RAG claims to fix this by turning retrieval into a closed loop: a reinforcement-learned policy rewrites the user's question, a dual-path retriever pulls both vector-similar passages and directed causal paths from a prebuilt graph, and a causal check verifies the answer before it is delivered. The paper reports that this loop outperforms standard and query-refining RAG methods on four question-answering benchmarks spanning commonsense, adversarial, and biomedical reasoning, across several language-model backbones. If the results hold, causal structure could become a standard component of retrieval rather than an optional add-on.

What carries the argument

The load-bearing mechanism is the causal dynamic feedback loop. A PPO-trained policy maps the query embedding to one of three refinement actions—expand, simplify, or decompose—then retrieval runs over two paths at once: dense semantic search over passage embeddings and traversal of a directed causal graph built from verified cause-effect triples. The retrieved union is rewritten and fed to the generator, and a causal consistency score checks whether the generated claims are entailed by the retrieved causal paths; if the score falls below a threshold, generation is retried under stricter grounding constraints. A separate hallucination score triggers a similar correction. This loop is what the paper claims converts correlation-based retrieval into causality-aware generation.

What would settle it

Rebuild the causal graph and the query-refinement training data from sources that exclude every evaluation question and its answer, then rerun the four-benchmark comparison; if the accuracy and causal-coverage margins collapse, the gains came from the graph encoding answer-related information rather than from the feedback loop.

Watch

Extended reading notes

Core claim

The paper's central discovery claim is that injecting causal structure into every stage of a RAG pipeline—query rewriting, retrieval, and answer verification—produces consistently higher accuracy than retrieval based on semantic similarity alone. Concretely, CDF-RAG reports the highest accuracy on all four evaluated benchmarks with four different LLM backbones, with gains such as 16 percent over the strongest baseline on the biomedical dataset when using GPT-4. It also reports higher causal retrieval coverage, higher groundedness, and lower hallucination rates, attributing the gains to the feedback loop rather than to any single module.

Load-bearing premise

The causal metrics assume that the causal graph built with help from the same LLM that later checks the answers is an independent source of causal truth.

Editorial extensions

If this is right

Across all four datasets and all four tested backbones, CDF-RAG reports the highest accuracy among the compared RAG systems, including on biomedical and adversarial questions.
Causal retrieval coverage reaches near-perfect levels on the biomedical benchmark (1.00 with GPT-4), suggesting the graph traversal routinely surfaces query-aligned cause-effect paths.
The stepwise ablation shows each added component—RL refinement, causal graph, rewriter, hallucination correction—improves F1 and groundedness while lowering hallucination rate from 0.18 to 0.07.
Because the final answer is checked against retrieved causal paths before delivery, generation is steered toward evidence-backed causal claims rather than plausible co-occurrence text.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

If the causal graph is built from the same benchmarks used for evaluation, part of the measured gain may be graph exposure rather than new causal reasoning; a transfer test with an externally built graph would separate the two.
The framework amplifies whatever causal coverage the graph has; it does not itself discover missing causal links, so its guarantees are upper-bounded by graph quality.
A natural extension is to apply the same loop to open-domain RAG where no domain graph exists, using the vector retriever alone as a fallback; the comparison would show when graph construction pays for itself.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

You should know two things about this paper. First, the architecture is a legitimate integration of known pieces: RL-based query refinement, dual-path retrieval over a causal graph and a vector store, and a hallucination-correction loop. The authors ablate each component and release code, which is more than many papers in this area do. Second, the empirical claims as presented are not trustworthy. The evaluation leaks information from the test benchmarks into training and graph construction, and the main causal metric measures agreement with GPT-4 rather than with independent ground truth.

The novelty is modest but real. The closed feedback loop—query refinement influencing retrieval, retrieval influencing verification, verification triggering regeneration—is a sensible engineering contribution over single-shot causal RAG or static refinement methods. If the numbers were solid, this would be a useful framework for medical QA and other domains where causal coherence matters. The ablations show monotonic gains as modules are added, which is a good sign that the pieces do something.

The soft spots are load-bearing. Appendix A.1 builds the causal graph from the four evaluation datasets, Appendix A.3 trains the RL agent by sampling queries from three of them, and Appendix B.1 fine-tunes all non-GPT-4 backbones on a multi-task dataset derived from the same sources. Appendix B.4 evaluates on roughly 2,200 queries across those four datasets without ever stating that the evaluation queries are disjoint from the training data. That is a textbook leakage risk, and it directly undermines the 'state-of-the-art accuracy across all datasets and models' claim. The causal metric is also circular: the gold-standard graph for CRC is constructed using GPT-4 verification, and the system's own graph is verified by GPT-4, so the metric mostly measures agreement with GPT-4's judgments. Hallucination rate is defined as answers the LLM verifier flags, which is again the system evaluating itself. There are no error bars or significance tests anywhere, and key pieces like the entailment indicator and the claim set Y are never operationalized.

These are not minor presentation issues. They go to the validity of the central claim. The framework itself is coherent and could work, but the current evidence cannot distinguish real gains from benchmark overfitting and circular scoring.

Who should read this? Someone working on causal RAG or query refinement might read it for architectural ideas, with the numbers treated as entirely unverified. I would not cite the results in my own work yet.

My recommendation: send it to peer review, but with a clear expectation of major revision. The reviewers should demand a proper train/test split, an independent causal gold standard, and significance testing. If the authors can supply those, the paper has a chance. If not, it should not be accepted.

Referee Report

5 major / 5 minor

Summary. The paper proposes CDF-RAG, a retrieval-augmented generation framework that adds a causal knowledge graph, reinforcement-learning-based query refinement, dual-path semantic/causal retrieval, response verification against causal structures, and hallucination correction. The authors evaluate CDF-RAG on CosmosQA, MedQA, MedMCQA, and AdversarialQA with GPT-4, LLaMA 3-8B, Mistral, and Flan-T5, reporting state-of-the-art accuracy, causal retrieval coverage, and hallucination metrics.

Significance. If its empirical claims were supported, CDF-RAG would be a useful contribution to causality-aware RAG: the integration of RL-driven query rewriting, structured causal graph traversal, and verifier-based correction is a coherent architectural idea, and the public code release is a positive step. However, the current evidence does not support the stated conclusions because the evaluation protocol appears to train and construct knowledge resources on the same benchmarks used for testing, and the causal metrics are circular. The significance of the claimed improvements therefore cannot be assessed from the reported results.

major comments (5)

[Appendix A.1, A.3, B.1, B.4] No train/test split is reported anywhere. Appendix A.1 states that the causal graph and query-refinement prompts are constructed from CosmosQA, MedQA, MedMCQA, and AdversarialQA; Appendix A.3 trains the PPO query-refinement agent by sampling queries from MedQA, CosmosQA, and AdversarialQA; Appendix B.1 fine-tunes every non-GPT-4 backbone on the same multi-task dataset; and Appendix B.4 reports evaluation on approximately 2,200 queries from these four benchmarks without asserting that they are disjoint from the training or graph-construction data. As a result, the graph, the RL policy, and the fine-tuned generators may all encode information from the evaluation sets, making the accuracy and causal-coverage gains in Tables 1 and 2 uninterpretable as generalization results.
[Section 4.2 vs. Section 2.1] The Causal Retrieval Coverage metric is circular. Section 4.2 defines CRC by checking retrieved items against a 'gold-standard causal graph constructed using GPT-4 verification,' while Section 2.1 builds the system's own causal graph using the same GPT-4 verification procedure. There is no independent reference, so CRC measures agreement with GPT-4 rather than causal coverage with respect to an external ground truth. The reported perfect CRC score of 1.00 on MedMCQA therefore does not constitute evidence of causal retrieval quality unless an independently constructed graph is provided.
[Sections 2.4 and 2.5] The formal definitions of the reported metrics are underspecified. The causal consistency score Scausal uses an indicator I(C→E ⊨ ŷ) with no operational definition of entailment between a causal triple and a generated response; the hallucination score Shallucination depends on an unexplained claim extraction function Y and a set intersection K∩Y; and the groundedness metric in Section 4.2 is described only as 'span-level alignment.' Without these definitions, the metrics cannot be reproduced or verified, and the claims based on them are not falsifiable as stated.
[Table 1 and Section 4.1] The headline claim that 'across all datasets and models, CDF-RAG achieves the highest accuracy scores' is supported only by point estimates. No standard errors, confidence intervals, significance tests, or per-cell evaluation sizes are reported for Table 1 or Tables 4 and 5, and the total of roughly 2,200 queries in Appendix B.4 is not tied to the per-method and per-dataset results. The observed margins, such as the 16% accuracy gain over Gym-RAG on MedMCQA, need statistical support before they can be interpreted as robust improvements.
[Appendix B.4 and Section 4.3] Hallucination Rate is defined as the percentage of responses flagged by the LLM verifier, which is the same mechanism used for hallucination correction within CDF-RAG. This makes HR a measure of consistency with the system's own verifier rather than an objective measure of hallucination. The 'lowest HR' claims in Tables 4 and 5 require an external or human-annotated hallucination benchmark to be meaningful.

minor comments (5)

[References] Several references use placeholder citations such as '(et al., 2020)', '(et al., 2025)', and '(et al., 2024)' instead of proper author names; these need to be completed.
[Table 2] Table 2 omits the MedQA rows for both CRC and Context Relevance, even though the text and Figure 3 report MedQA-specific results; the table should be completed or the omission should be explained.
[Figure 3] The sentence 'The improvement is especially notable with GPT-4 and GPT-4 Mistral LLaMA 3-8B Flan-T5' is garbled and should be rewritten to list the model names correctly.
[Introduction and Contributions] The bullet point 'We introduceCDF-RAG' is missing a space, and the abstract's statement that existing RAG frameworks 'primarily rely on semantic similarity and correlation-driven retrieval' is an overgeneralization that should be qualified in light of existing structured and causality-aware RAG work.
[Appendix A.4] The prompt templates appear more than once, with the Simplification Prompt Template repeated in slightly different wording; please consolidate these into a single canonical version.

Circularity Check

3 steps flagged · score 7.0 of 10

CDF-RAG's headline gains are in-sample and self-referential: the agent and backbones are trained on the evaluation benchmarks, CRC's gold standard is GPT-4-built like the system's own graph, and Hallucination Rate is defined by the system's own verifier.

fitted input called prediction [Appendix A.1, A.3, B.1; evaluation in Appendix B.4]
"Each dataset is used to extract causally relevant triples and generate query refinement prompts across decomposition, expansion, and simplification modes. ... The training curriculum covers diverse domains by sampling queries from MedQA, CosmosQA, and AdversarialQA. ... Across all experiments, CDF-RAG was evaluated on approximately 2,200 queries spanning four benchmark datasets—CosmosQA, MedQA, MedMCQA, and AdversarialQA—across multiple LLM backbones."

The query-refinement RL agent, the LoRA fine-tuning of the non-GPT-4 backbones, and the causal graph are all constructed from the same four benchmarks used for final evaluation, with no train/test split stated anywhere in the paper. The reported accuracy and CRC numbers are therefore in-sample fits to the evaluation distribution rather than predictions on held-out queries. The claimed state-of-the-art superiority is forced by optimizing on the same data that is later measured, not demonstrated as out-of-sample generalization.
self definitional [Section 2.1 vs Section 4.2]
"To ensure logical validity, extracted causal pairs are verified by GPT-4 before being encoded into G as directed triples (C,E, relation). ... For each query, CRC is computed by checking whether the retrieved items match entries in a gold-standard causal graph constructed using GPT-4 verification."

The same GPT-4 verification that validates the system's causal graph also defines the gold-standard graph against which Causal Retrieval Coverage is measured. CRC therefore measures agreement between the system's retrieval graph and GPT-4's causal judgments—the very oracle used to build that graph—rather than agreement with an independent causal ground truth. As a reported evidence of causal correctness, the metric is self-referential by construction.

1 more flagged steps

self definitional [Section 2.5 and Section 4.3]
"If Shallucination > δ, where δ is a predefined threshold, the system applies knowledge rewriting. ... HR denotes the percentage of responses flagged as hallucinated by the LLM verifier."

Hallucination Rate is defined as the percentage of responses flagged by the same LLM verifier whose flags trigger the correction and rewriting loop. Once the hallucination-correction stage is added, outputs are regenerated until the verifier no longer flags them, so the reported HR drop from 0.18 to 0.07 is partly by construction. The metric is a self-report of the system's own detector, not an independent measure of factual accuracy.

full rationale

The paper's central claim—state-of-the-art accuracy and causal correctness across all four benchmarks—is undercut by three evaluation circularities. First, no train/test split is described: Appendix A.1 builds the causal graph and query-refinement data from the same four datasets later evaluated, Appendix A.3 trains the PPO query-refinement agent on MedQA, CosmosQA, and AdversarialQA, and Appendix B.1 fine-tunes LLaMA 3-8B, Mistral, and Flan-T5 on the same multi-task dataset. The results in Tables 1, 2, 4, and 5 are therefore in-sample measurements, not held-out predictions. Second, the causal retrieval metric is self-referential: CRC's gold-standard graph is 'constructed using GPT-4 verification,' and the system's own Neo4j graph also contains pairs 'verified by GPT-4 before being encoded into G,' so high CRC mainly shows concordance with GPT-4 rather than with an independent causal reference. Third, the reported hallucination rate is defined as the percentage flagged by the LLM verifier that also triggers the correction module; adding that module lowers HR because the system regenerates until the verifier is satisfied. The Limitations section acknowledges reliance on structured graphs and GPT-based validation overhead but does not address these methodological circularities. Because the advertised gains are not separated from their training data or their verification oracles, the score is 7.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claim depends on LLM-generated causal graphs, undisclosed thresholds, and graph construction from the evaluation corpora. None of these are grounded in independent evidence, and one of them, the GPT-4-built gold standard, creates a circular evaluation.

free parameters (4)

Reward weights λ1, λ2, λ3, λ4 = not reported
Tune the RL query refinement objective in Section 2.2; values are not given, so the policy is underspecified.
Causal consistency threshold τ = not reported
Triggers fallback generation in Section 2.4 if Scausal < τ; value is not stated.
Hallucination threshold δ = not reported
Triggers knowledge rewriting in Section 2.5 if Shallucination > δ; value is not stated.
Retrieval top-k = not reported
Dual-path retrieval returns top-k passages in Section 2.3; k is never specified.

assumptions (5)

domain assumption UniCausal BERT classifier correctly detects cause-effect pairs in the four benchmark datasets.
Section 2.1 builds the graph on UniCausal output; no precision or recall on these corpora is reported.
domain assumption GPT-4 verification yields correct and unbiased causal judgments for pair validation and for the gold-standard CRC graphs.
Sections 2.1 and 4.2; cited literature (Chi et al., 2024; Jin et al., 2023) questions LLM causal inference, so this is load-bearing.
ad hoc to paper Causal graphs and query refinement training examples built from the four benchmarks do not leak test-set answers.
Appendix A.1 states graphs and prompts are extracted from the evaluation datasets; no train/test split is specified.
domain assumption The causal entailment indicator I(C→E |= y) and claim sets K, Y are computable and meaningful as stated.
Sections 2.4 and 2.5 present formulas without an operational algorithm or validation.
domain assumption The RL reward components can be measured at training time without oracle access and correlate with final answer accuracy.
Section 2.2 and Appendix A.3 define rewards from retrieval coverage, causal depth, context relevance, and hallucination penalty; no evidence of correlation is provided.

how reviews work

0 comments

Cite this review

Pith. "Pith review of CDF-RAG: Causal Dynamic Feedback for Adaptive Retrieval-Augmented Generation." pith.science (2026). https://pith.science/paper/TPHHUFJO

@misc{pith2026250412560,
  author       = {Pith},
  title        = {Pith review of: CDF-RAG: Causal Dynamic Feedback for Adaptive Retrieval-Augmented Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TPHHUFJO}},
  note         = {Machine review of arXiv:2504.12560}
}

read the original abstract

Retrieval-Augmented Generation (RAG) has significantly enhanced large language models (LLMs) in knowledge-intensive tasks by incorporating external knowledge retrieval. However, existing RAG frameworks primarily rely on semantic similarity and correlation-driven retrieval, limiting their ability to distinguish true causal relationships from spurious associations. This results in responses that may be factually grounded but fail to establish cause-and-effect mechanisms, leading to incomplete or misleading insights. To address this issue, we introduce Causal Dynamic Feedback for Adaptive Retrieval-Augmented Generation (CDF-RAG), a framework designed to improve causal consistency, factual accuracy, and explainability in generative reasoning. CDF-RAG iteratively refines queries, retrieves structured causal graphs, and enables multi-hop causal reasoning across interconnected knowledge sources. Additionally, it validates responses against causal pathways, ensuring logically coherent and factually grounded outputs. We evaluate CDF-RAG on four diverse datasets, demonstrating its ability to improve response accuracy and causal correctness over existing RAG-based methods. Our code is publicly available at https://github.com/ elakhatibi/CDF-RAG.

Figures

Figures reproduced from arXiv: 2504.12560 by the authors.

**Figure 1.** Rethinking Retrieval-Augmented Generation (RAG). (a) Traditional RAG pipelines rely on static queries and keyword- or similarity-based retrieval, often retrieving topically related but causally irrelevant content, which can result in hallucinated or incoherent outputs. (b) CDF-RAG addresses these limitations through reinforcement learning-based query refinement, dual-path retrieval combining semantic vector search … view at source ↗

**Figure 2.** Overview of CDF-RAG Framework. (a) The CDF-RAG pipeline refines user queries (LLM + RL), retrieves structured causal and unstructured textual knowledge, applies knowledge rewriting, and ensures factual consistency through causal verification. (b) The PPO-trained query refinement agent optimizes retrieval coverage and causal consistency. causal claims. The verification process computes a causal consistency score: Sca… view at source ↗

**Figure 3.** Groundedness comparison of different meth [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

Figures from the paper (1 more)

**Figure 4.** Figure 4: Ablation study of CDF-RAG across incremental stages. Left: performance metrics including CRC, SRS, [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

PerCaM-Health: Personalized Dynamic Causal Graphs for Healthcare Reasoning
cs.LG 2026-05 unverdicted novelty 6.0 of 10

PerCaM-Health learns evolving personalized dynamic causal graphs from longitudinal health data to enable more reliable patient-level counterfactual queries than cohort or per-patient baselines.
Enhancing LLMs through human feedback: a journey towards self-improvement
cs.IR 2026-07 unverdicted novelty 4.0 of 10

An auxiliary feedback RAG continuously ingests classified human feedback to iteratively raise a primary RAG system’s answer accuracy and relevance.

Reference graph

Works this paper leans on

17 extracted references · 14 canonical work pages · cited by 2 Pith papers

[1]

Is the causal relationship valid and sup- ported? (Answer: True/False)

work page
[2]

arXiv preprint arXiv:2412.19437

Deepseek-v3 technical report. arXiv preprint arXiv:2412.19437. Hang Luo, Jian Zhang, and Chujun Li. 2025. Causal graphs meet thoughts: Enhancing complex rea- soning in graph-augmented llms. arXiv preprint arXiv:2501.14892. OpenAI. 2023. Gpt-4 technical report. https:// openai.com/research/gpt-4. Accessed: 2025- 03-27. Ankit Pal, Logesh Kumar Umapathi, and...

arXiv 2025
[3]

High blood pressure causes stroke

Redeep: Detecting hallucination in retrieval- augmented generation via mechanistic interpretabil- ity. arXiv preprint arXiv:2410.11414. Fiona Anting Tan, Xinyu Zuo, and See-Kiong Ng. 2023. Unicausal: Unified benchmark and repository for causal text mining. In International Conference on Big Data Analytics and Knowledge Discovery, pages 248–262. Springer. ...

arXiv 2023
[4]

{ARG0}" causes

Estimate your confidence in the answer (High / Medium / Low) Output Format: Correctness: {True / False} Refined Causal Statement: "{ARG0}" causes "{ARG1}" or "{ARG1}" causes "{ARG0}" Confidence: {High / Medium / Low} Explanation: {Short justification grounded in domain knowledge} A.3 Reinforcement Learning for Query Refinement. To dynamically optimize que...

work page
[5]

If the direction is incorrect, provide the corrected direction

work page
[6]

Provide a one-sentence explanation justi- fying your decision

work page
[10]

Refine the following query for better causal retrieval

Ensure that the simplified version can be interpreted and answered indepen- dently. Guidelines: • Use precise language that avoids un- necessary technical or abstract phras- ing. • Do not generate multiple sub- questions. • Keep the simplified question to a single line of text. • Preserve the core meaning of the origi- nal question. Here is your task: • P...

work page
[11]

Identify any ambiguity, compound phrasing, or indirect constructs in the input question

work page

Show all 17 references

[12]

Reformulate the question as a concise, direct, and self-contained single ques- tion
[13]

Guidelines: • Use precise language that avoids un- necessary technical or abstract phras- ing

Ensure that the simplified version can be interpreted and answered indepen- dently. Guidelines: • Use precise language that avoids un- necessary technical or abstract phras- ing. • Do not generate multiple sub- questions. • Keep the simplified question to a single line of text...
[14]

Analyze the multihop question to iden- tify its underlying causal or semantic components
[15]

Reformulate the question into a list of 2–4 clear, concise, self-contained sub- questions that can be independently an- swered
[16]

Break this question into sub-questions for causal reasoning

Maintain logical flow between sub- questions (i.e., each one should build toward answering the original ques- tion). Guidelines: • Avoid repeating the same phrasing across sub-questions. • Each sub-question should be answer- able on its own. • Use one line per sub-question, an...
[17]

Identify missing context or implicit as- sumptions in the question
[18]

Reformulate the question to explic- itly mention key entities, causal mech- anisms, or domains relevant to the query
[19]

Make the question more specific for causal reasoning

Ensure the expanded question guides a more targeted and informative answer. Guidelines: • Use a single line for the expanded ques- tion. • Avoid changing the core topic, but add specificity or scope. • Preserve the original intent, while mak- ing the question more complete or ...

2019
[2024]

arXiv preprint arXiv:2410.04585

Reasoning-enhanced healthcare predictions with knowledge graph community retrieval. arXiv preprint arXiv:2410.04585. Bowen Jin, Jinsung Yoon, Jiawei Han, and Sercan O Arik. 2024. Long-context llms meet rag: Overcom- ing challenges for long inputs in rag. In The Thir- teenth In...

2024 arXiv

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.

[1] [1]

Is the causal relationship valid and sup- ported? (Answer: True/False)

work page

[2] [2]

arXiv preprint arXiv:2412.19437

Deepseek-v3 technical report. arXiv preprint arXiv:2412.19437. Hang Luo, Jian Zhang, and Chujun Li. 2025. Causal graphs meet thoughts: Enhancing complex rea- soning in graph-augmented llms. arXiv preprint arXiv:2501.14892. OpenAI. 2023. Gpt-4 technical report. https:// openai.com/research/gpt-4. Accessed: 2025- 03-27. Ankit Pal, Logesh Kumar Umapathi, and...

arXiv 2025

[3] [3]

High blood pressure causes stroke

Redeep: Detecting hallucination in retrieval- augmented generation via mechanistic interpretabil- ity. arXiv preprint arXiv:2410.11414. Fiona Anting Tan, Xinyu Zuo, and See-Kiong Ng. 2023. Unicausal: Unified benchmark and repository for causal text mining. In International Conference on Big Data Analytics and Knowledge Discovery, pages 248–262. Springer. ...

arXiv 2023

[4] [4]

{ARG0}" causes

Estimate your confidence in the answer (High / Medium / Low) Output Format: Correctness: {True / False} Refined Causal Statement: "{ARG0}" causes "{ARG1}" or "{ARG1}" causes "{ARG0}" Confidence: {High / Medium / Low} Explanation: {Short justification grounded in domain knowledge} A.3 Reinforcement Learning for Query Refinement. To dynamically optimize que...

work page

[5] [5]

If the direction is incorrect, provide the corrected direction

work page

[6] [6]

Provide a one-sentence explanation justi- fying your decision

work page

[7] [10]

Refine the following query for better causal retrieval

Ensure that the simplified version can be interpreted and answered indepen- dently. Guidelines: • Use precise language that avoids un- necessary technical or abstract phras- ing. • Do not generate multiple sub- questions. • Keep the simplified question to a single line of text. • Preserve the core meaning of the origi- nal question. Here is your task: • P...

work page

[8] [11]

Identify any ambiguity, compound phrasing, or indirect constructs in the input question

work page