Pith. sign in

REVIEW 3 major objections 7 minor 54 references

LEDGERMIND: Provenance-Constrained Multimodal Agentic Reasoning with a Structured Evidence Ledger

T0 review · 3 major / 7 minor · reviewed 2026-07-31 · grok-4.5

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

Pith's one-line read Multimodal agents become more accurate and faithful when every claim must cite a structured evidence ledger that repair cannot invent content into.

desk verdict Solid systems paper: ledger-as-state plus typed repair is a real, usable pattern; faithfulness claims need a judge-robustness check but the accuracy and ablation story already stands. read the letter →

arxiv 2607.28374 v1 pith:ILYAGCPK submitted 2026-07-30 cs.LG

Enjun Du , Hange Zhou , Chenxu Du , Siyi Liu , Zirong Chen , Ziyu Zheng , Yongqi Zhang This is my paper · ORCID
classification cs.LG
keywords multimodalagentsvisualquestionansweringtrajectoryfaithfulnessstructuredevidenceledgerprovenancephantomgroundingtypedrepairadaptiveinference
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

Final-answer scores hide how multimodal agents actually reach answers: unsupported steps, citations that look valid while smuggling in new entities, needless deep reasoning on simple questions, and free-form repair that invents fresh unsupported content. LedgerMind treats the agent trajectory as a provenance-constrained state machine whose only shared state is a Structured Evidence Ledger. Tool outputs become ledger entries; reasoning and decisions may cite only active entries; grounding is checked at entity and numeric level; and repair is limited to typed transitions that cannot add content without tool provenance. Across several multimodal benchmarks and backbone models, that design raises both answer accuracy and trajectory-level faithfulness, with a formal guarantee that repair does not amplify provenance-less claims.

What carries the argument

The Structured Evidence Ledger: a normalized trajectory state in which every tool return becomes an entry with source, type, confidence, lifecycle status, and dependencies; claims may cite only active entries; a three-layer grounding protocol checks support coverage plus entity and numeric containment; and an event-triggered engine repairs only via typed operators that preserve provenance non-amplification.

What would settle it

Run the same backbones with and without the ledger on matched trajectories: if answer accuracy rises while entity/numeric grounding rates, decision-grounding among correct answers, and chain-alignment metrics do not improve—or if typed repair still injects unsupported entities that the checks miss—the central claim fails.

Watch

Extended reading notes

Core claim

Treating a multimodal agent trajectory as a provenance-constrained state machine centered on a Structured Evidence Ledger—with entity- and numeric-level grounding and typed repair—improves both final-answer accuracy and trajectory-level faithfulness, while guaranteeing that repair cannot introduce ledger content without tool-produced provenance.

Load-bearing premise

That checking whether claimed entities and numbers appear in cited tool evidence, scored by an external model that breaks traces into a handful of atomic claims, is enough to measure whether the trajectory was truly grounded.

Editorial extensions

If this is right

  • Agent evaluation can report trajectory faithfulness (unsupported-claim rate, grounded decisions, right-for-right reasons) alongside accuracy instead of accuracy alone.
  • Citation-looking intermediate text is no longer treated as grounded unless conclusion-level entities and numbers are licensed by cited tool evidence.
  • Repair loops can be restricted to typed ledger/action transitions so free-form self-reflection cannot silently invent new provenance-less claims.
  • Simple and knowledge-heavy queries can be routed to a shallow path to avoid overwriting correct direct answers with noisy multi-step inference.
  • The same ledger state can later supply training signals that distinguish grounded from ungrounded claims at trajectory level.

Reading between the lines

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

  • If provenance is structural state rather than a prompt preference, similar ledger contracts could transfer to tool-using text agents and retrieval pipelines where citation–content mismatch is already known.
  • Long-horizon or video agents would need the ledger’s lifecycle and time-to-live rules to become first-class memory, not just per-query scratch state.
  • A learned complexity router could replace the paper’s rule-based dispatcher once task mixtures grow more heterogeneous than the evaluated benchmarks.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 7 minor

Summary. The paper proposes LedgerMind, a training-free runtime that treats multimodal agent trajectories as provenance-constrained state machines. Tool outputs are normalized into a Structured Evidence Ledger; reasoning and decision claims may cite only active entries; entity- and numeric-level containment (ECC/NCC) is checked; and repair is restricted to typed operators with a provenance non-amplification guarantee (Proposition 1). An Adaptive Dual-Path Dispatcher routes simple vs. complex queries, and an event-triggered verifier drives repair. The design targets four failure modes that final-answer accuracy obscures (unsupported intermediate claims, Phantom Grounding, over-reasoning, repair-time amplification). Experiments on VTC-Bench, MMStar, MMMU, MMMU-Pro, EMMA, MC-Search, and an in-house Hard-200 set, across six backbone MLLMs with matched tool budgets, report gains in answer accuracy and in trajectory-level metrics (UCR_reason, GDR, R4R, WDG), with ablations on MMMU-Pro and chain-alignment gains (HPS up, RD down) on MC-Search.

Significance. If the joint accuracy-and-faithfulness claim holds, the work is a useful systems contribution for multimodal agentic VQA: it makes provenance a structural runtime constraint rather than a prompting preference, names Phantom Grounding at trajectory level, and couples adaptive depth control with typed repair under a clear (if definitional) non-amplification guarantee. Strengths include multi-benchmark, multi-backbone evaluation under matched tool budgets; component ablations that isolate the ledger, typed repair, ECC/NCC, and the dispatcher; and MC-Search chain metrics that are harder to explain by answer-only post-hoc correction. Proposition 1 is correctly stated and proof-checked by operator enumeration. The main significance risk is that trajectory faithfulness is partly operationalized by a same-family external MLLM judge and by deterministic ECC/NCC over a limited alias table; if those instruments favor ledger-shaped traces, the faithfulness half of the claim is overstated even when accuracy rises.

major comments (3)
  1. [§4.3, Figure 5; §3.4; §4.1] §3.4 and §4.3 (S-RFA / Figure 5): Trajectory faithfulness (UCR_reason, GDR, R4R, WDG) is audited by a fixed Gemini-3.1-Pro judge that decomposes both baseline and LedgerMind traces into ≤10 atomic claims, while Gemini-3-Flash and Gemini-3.1-Pro are also evaluated backbones (§4.1). This creates a same-family auditor risk: the judge may preferentially credit ledger-style citation structure and leaf-evidence phrasing. The joint claim that accuracy gains come from grounded trajectories (not post-hoc rewriting) load-bears on this audit. Please add at least one of: (i) a second auditor from a different vendor family with agreement statistics, (ii) a human-labeled subset with inter-annotator agreement, or (iii) a blinded protocol that strips ledger IDs/formatting before judging. Without this, R4R/WDG and the enclosing polygons in Figure 5 remain only weakly identified.
  2. [§3.2 Eqs. (3)–(5); Appendix C; Table 3] §3.2 Eqs. (3)–(5) and Appendix C: ECC/NCC operationalize claim–evidence containment via token overlap, a small alias table, and type-aware numeric tolerance. Limitation J already notes paraphrase, coreference, and cross-lingual gaps. Because Phantom Grounding (F2) and the Hard-split ablation drop for w/o ECC/NCC (Table 3, −5.19 overall, larger on Hard) are central to the paper’s diagnostic story, please quantify false-negative/false-positive rates of ECC/NCC on a labeled claim set (including paraphrases that are still licensed by evidence and entity substitutions that are not). Otherwise it is unclear whether ECC/NCC catches F2 or mainly enforces surface form that the ledger already encourages.
  3. [§4.2; Figure 4; Appendix H] §4.2 / Appendix H (Hard-200): Hard-200 is committee-mined and partly self-constructed (RealCAR), then scored by a local LLM judge. Gains are large and uniform (Figure 4, no negative cell), which is encouraging, but the set is not a public fixed benchmark and selection uses cross-vendor failure rates that may correlate with the same failure modes LedgerMind is built to fix. For the stress-test claim, either release the full set with selection scripts and judge prompts, or demote Hard-200 to supplementary evidence and rest the main accuracy claims on the public suites (VTC-Bench, EMMA, MMMU-Pro, MC-Search), which already support a substantial part of the result.
minor comments (7)
  1. [§3.3 Proposition 1] Proposition 1 (§3.3) is correct but nearly definitional given the operator set R. In the main text, state explicitly that it guarantees provenance locality, not factual correctness of tools—this is in the proof paragraph and Limitations but should be adjacent to the proposition statement.
  2. [Appendix C] Appendix C: confidence demotion values (0.50/0.52/0.55) and σ_verify=0.6 are grid-searched on 50 held-out questions. Briefly report sensitivity of main metrics to these thresholds in the appendix so readers can see stability beyond the development set.
  3. [§4.4 Table 3] §4.4 Table 3: the dispatcher ablation shows Easy drop and stable Hard—good signature for F3—but absolute Easy/Medium/Hard definitions for MMMU-Pro should be stated in the table caption or appendix for reproducibility.
  4. [Figure 1; Figure 2] Figure 1 and Figure 2 are helpful; ensure vector text remains legible at single-column width (several labels are dense).
  5. [§2.1; Appendix B] Related work (§2.1) correctly disclaims novelty of provenance tracing in isolation; a short pointer to how the 11-field schema differs from Open Provenance Model-style records would help systems readers.
  6. [Table 1; §4.2] EMMA Coding subset shows a small regression (−0.18 vs Gemini 3.1 Pro, Table 1); the footnote explanation is fine—consider one sentence in the main text so readers do not treat it as a silent failure.
  7. [Title / running header] Typos/formatting: title casing inconsistency between running header (“A STRUCTURED EVIDENCE RUNTIME…”) and abstract name; occasional missing spaces in compound terms in the arXiv text dump. Clean for camera-ready.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: empirical gains are on external benchmarks; Proposition 1 is a correct design invariant, not a fitted or self-citing prediction.

full rationale

LedgerMind’s load-bearing empirical claims are answer accuracy and trajectory-level metrics on public benchmarks (VTC-Bench, MMStar, MMMU, MMMU-Pro, EMMA, MC-Search) plus Hard-200, compared to native backbone CoT under matched tool budgets. Those targets are not quantities fitted from the method and re-labeled as predictions. Proposition 1 (provenance non-amplification) holds because the seven typed repair operators are defined to add ledger entries only via tool invocation and the deterministic map M; the short proof enumerates operators and does not smuggle an empirical conclusion into the premises. That is a legitimate systems invariant, not self-definitional circular science. Thresholds (e.g., ECC/NCC demotion values) are chosen on a small held-out dev set and used as fixed runtime parameters, not reported as out-of-sample “predictions” of the same fitted signal. Related-work self-citations (GraphMaster, Octopus, etc.) are background and not used as uniqueness theorems that force the central result. Author-defined faithfulness metrics and a fixed LLM judge raise validity questions about measurement bias, but that is an evaluation-construct concern, not a derivation that reduces by construction to its inputs. No step matches the circularity patterns at a load-bearing level.

Assumptions & free parameters 6 free parameters · 5 assumptions · 3 invented entities

The central empirical claim rests on engineering axioms about what counts as grounded evidence and on several hand-chosen thresholds and budgets, not on fitted physical constants. The ledger ontology and the identification of containment checks with faithfulness are the main domain assumptions; free parameters are operational thresholds from a small dev grid search.

free parameters (6)
  • ECC/NCC/structural confidence demotion values (0.50 / 0.52 / 0.55) and σ_verify=0.6 = 0.50 / 0.52 / 0.55 / 0.6
    Chosen by grid search on a 50-question held-out dev set so failures fall below the verification threshold; directly gates hypothesis verification and claim influence.
  • Numeric relative tolerance δ_read = 0.15
    Allows continuous visual measurement noise; paper states stability on [0.10, 0.20], default 0.15 for Dual-Read and NCC on gauges.
  • Per-trigger repair budget M0 and initial pipeline budget M = 2
    Caps typed repair iterations and thus cost and intervention depth; set to 2 in algorithms and cost bound.
  • Default evidence confidences (perception 0.92, retrieval 0.85, dual-read agree 0.97 / disagree 0.65, empty-cite judgment = 0.92 / 0.85 / 0.97 / 0.65 / 0.55
    Hand-set priors that interact with σ_floor, σ_verify, and repair triggers; not learned from large data but affect control flow.
  • Crop-zoom factor and Dual-Read agreement threshold = z=2.5; Δv≤0.15
    2.5× crop and Δv≤0.15 agreement rule are fixed perception hyperparameters that change which values enter the ledger.
  • Decision coverage threshold β and hard confidence floor σ_floor = β=0.15; σ_floor=0.3
    Trigger unsupported-decision and confidence-drop repairs (β=0.15, σ_floor=0.3); policy-critical free knobs.
assumptions (5)
  • domain assumption Only tool-produced observations (or deterministic maps M of them) are legitimate ledger evidence; model paraphrases may not originate evidence (I1, I4).
    Load-bearing design axiom of the provenance contract; stated in §3.1 invariants and Proposition 1.
  • domain assumption Claim faithfulness is adequately captured by citation of active entries plus lexical support, entity containment (with a small alias table), and type-aware numeric match (ECC/NCC).
    Operationalizes grounding in §3.2; limitations section admits paraphrase/coreference/cross-lingual gaps.
  • ad hoc to paper A deterministic keyword/regex complexity classifier ϕ(q) can route simple vs complex queries well enough to avoid over-reasoning without a learned policy.
    Appendix D.1 rule set drives the Adaptive Dual-Path Dispatcher; authors note learned dispatchers as future work.
  • domain assumption An external frozen MLLM judge can label atomic claim grounding comparably for baseline and LedgerMind traces under one protocol.
    Symmetric Reasoning Faithfulness Audit in §4.3 depends on this for UCR/GDR/R4R/WDG.
  • standard math Standard tool-augmented agent and MLLM evaluation practice (matched tool budget, exact/partial credit rubrics, public VQA benchmarks).
    Background experimental norms; not proved in-paper but shared with the literature the paper compares to.
invented entities (3)
  • Structured Evidence Ledger (11-field trajectory state)
    purpose: Replace free-form trajectory text with canonical provenance-bearing state that all grounding, dispatch, and repair operate on.
    Central abstraction of the paper; engineering structure rather than a physical entity, with no existence claim beyond the runtime.
  • Phantom Grounding (citation-backed entity/numeric hallucination at trajectory level)
    purpose: Name and target a failure where citation structure passes but conclusion-level content is absent from cited evidence.
    Diagnostic construct instantiated from RAG citation-faithfulness ideas; validated only via the paper’s own metrics and judge protocol.
  • Three-Layer Grounding Protocol + Event-Triggered typed repair operator set R
    purpose: Enforce support coverage, ECC, NCC and restrict repair to non-amplifying state transitions.
    Method machinery introduced to implement the provenance contract; Proposition 1 refers to this finite operator set.

how reviews work

0 comments
Cite this review

Pith. "Pith review of LEDGERMIND: Provenance-Constrained Multimodal Agentic Reasoning with a Structured Evidence Ledger." pith.science (2026). https://pith.science/paper/ILYAGCPK

@misc{pith2026260728374,
  author       = {Pith},
  title        = {Pith review of: LEDGERMIND: Provenance-Constrained Multimodal Agentic Reasoning with a Structured Evidence Ledger},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ILYAGCPK}},
  note         = {Machine review of arXiv:2607.28374}
}
read the original abstract

Multimodal agents for visual question answering increasingly operate as multi-step trajectories that interleave perception, retrieval, and reasoning, yet evaluation still largely reduces to final-answer accuracy. This aggregate signal cannot tell whether a correct answer was reached through grounded evidence, language priors, or accidental error cancellation. We propose to treat a multimodal agent trajectory as a provenance-constrained state machine: tool outputs are normalized into a Structured Evidence Ledger that serves as the trajectory state, downstream reasoning and decision claims may cite only active ledger entries, grounding is checked at the entity and numeric level, and repair is realized as typed state transitions that cannot introduce content without tool-produced provenance. We instantiate this design as LedgerMind (Provenance-Constrained Multimodal Agentic Reasoning with a Structured Evidence Ledger), augmented by a Three-Layer Grounding Protocol, an Adaptive Dual-Path Dispatcher that matches reasoning depth to question complexity, and an Event-Triggered Verification-and-Repair engine with a formal provenance non-amplification guarantee. We use LedgerMind to target four recurring failure patterns that final-answer accuracy tends to obscure: unsupported intermediate reasoning, citation-backed entity hallucination (Phantom Grounding), over-reasoning on simple queries, and repair-time amplification. Experiments across multiple multimodal reasoning benchmarks and backbone MLLMs show that LedgerMind improves both answer accuracy and trajectory-level faithfulness.

Figures

Figures reproduced from arXiv: 2607.28374 by the authors.

Figure 1
Figure 1. Representative failure patterns in multimodal agentic reasoning that final-answer accuracy [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The LedgerMind framework. The Structured Evidence Ledger is the central trajectory state: tool outputs are normalized into ledger entries, reasoning and decision claims cite active entries, grounding checks claim–evidence consistency, the dispatcher controls evidence-acquisition depth, and typed repair modifies only ledger state or tool actions. 3 METHOD [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. (a) Overall accuracy on VTC-Bench (Zhu et al., 2026): LedgerMind with Gemini-3-Flash reaches 58.9%, a new state of the art. (b) Generalisation to three general VLM benchmarks (MM￾Star (Chen et al., 2024), MMMU (Yue et al., 2024), MMMU-Pro (Yue et al., 2025)): LedgerMind with GPT-4o and Gemini-3-Flash backbones outperforms 11 baselines. GPT-5.5 Gemini-3.1-Pro Gemini-3-Flash Sonnet-4.6 Opus-4.7 Kimi-K2.6 0 10 20 30 40… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Results on Hard-200 (Appendix H) across six frontier backbones. LedgerMind improves every backbone on the overall score and on each of the three sub-sources (RealCAR, TIR-Bench, BrowseComp-VL); the right heatmap (∆ = LedgerMind− baseline) contains no negative cell. Cha…
Figure 5
Figure 5. Figure 5: Symmetric Reasoning Faithfulness Audit on V*Bench and EMMA-160 across five frontier [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

54 extracted references · 7 linked inside Pith

  1. [1]

    arXiv preprint arXiv:2303.08774 , year=

    arXiv

  2. [2]

    arXiv preprint arXiv:2508.05748 , year=

    WebWatcher: Breaking New Frontiers of Vision-Language Deep Research Agent , author=. arXiv preprint arXiv:2508.05748 , year=

    arXiv

  3. [3]

    arXiv preprint arXiv:2312.11805 , year=

    Gemini: A Family of Highly Capable Multimodal Models , author=. arXiv preprint arXiv:2312.11805 , year=

    arXiv

  4. [4]

    Advances in Neural Information Processing Systems (NeurIPS) , year=

    Visual Instruction Tuning , author=. Advances in Neural Information Processing Systems (NeurIPS) , year=

  5. [5]

    Bai, Jinze and Bai, Shuai and Yang, Shusheng and Wang, Shijie and Tan, Sinan and Wang, Peng and Lin, Junyang and Zhou, Chang and Zhou, Jingren , journal=

  6. [6]

    Chen, Zhe and Wu, Jiannan and Wang, Wenhai and Su, Weijie and Chen, Guo and Xing, Sen and Zhong, Muyan and Zhang, Qinglong and Zhu, Xizhou and Lu, Lewei and Li, Bin and Luo, Ping and Lu, Tong and Qiao, Yu and Dai, Jifeng , journal=

  7. [7]

    Advances in Neural Information Processing Systems (NeurIPS) , year=

    Chain-of-Thought Prompting Elicits Reasoning in Large Language Models , author=. Advances in Neural Information Processing Systems (NeurIPS) , year=

  8. [8]

    Yao, Shunyu and Zhao, Jeffrey and Yu, Dian and Du, Nan and Shafran, Izhak and Narasimhan, Karthik and Cao, Yuan , booktitle=

Show all 54 references
  1. [9]

    Advances in Neural Information Processing Systems (NeurIPS) , year=

    Toolformer: Language Models Can Teach Themselves to Use Tools , author=. Advances in Neural Information Processing Systems (NeurIPS) , year=

  2. [10]

    Advances in Neural Information Processing Systems (NeurIPS) , year=

    Chameleon: Plug-and-Play Compositional Reasoning with Large Language Models , author=. Advances in Neural Information Processing Systems (NeurIPS) , year=

  3. [11]

    arXiv preprint arXiv:2404.18930 , year=

    Hallucination of Multimodal Large Language Models: A Survey , author=. arXiv preprint arXiv:2404.18930 , year=

    arXiv

  4. [12]

    Advances in Neural Information Processing Systems (NeurIPS) , year=

    Tree of Thoughts: Deliberate Problem Solving with Large Language Models , author=. Advances in Neural Information Processing Systems (NeurIPS) , year=

  5. [13]

    Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , year=

    Sur. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , year=

  6. [14]

    Advances in Neural Information Processing Systems (NeurIPS) , year=

    Self-Refine: Iterative Refinement with Self-Feedback , author=. Advances in Neural Information Processing Systems (NeurIPS) , year=

  7. [15]

    Advances in Neural Information Processing Systems (NeurIPS) , year=

    Reflexion: Language Agents with Verbal Reinforcement Learning , author=. Advances in Neural Information Processing Systems (NeurIPS) , year=

  8. [16]

    International Conference on Learning Representations (ICLR) , year=

    Self-Consistency Improves Chain of Thought Reasoning in Language Models , author=. International Conference on Learning Representations (ICLR) , year=

  9. [17]

    Findings of the Association for Computational Linguistics (ACL) , year=

    Chain-of-Verification Reduces Hallucination in Large Language Models , author=. Findings of the Association for Computational Linguistics (ACL) , year=

  10. [18]

    Huang, Qidong and Dong, Xiaoyi and Zhang, Pan and Wang, Bin and He, Conghui and Wang, Jiaqi and Lin, Dahua and Zhang, Weiming and Yu, Nenghai , booktitle=

  11. [19]

    International Conference on Learning Representations (ICLR) , year=

    Large Language Models Cannot Self-Correct Reasoning Yet , author=. International Conference on Learning Representations (ICLR) , year=

  12. [20]

    Retrieval-Augmented Generation for Knowledge-Intensive

    Lewis, Patrick and Perez, Ethan and Piktus, Aleksandra and Petroni, Fabio and Karpukhin, Vladimir and Goyal, Naman and K. Retrieval-Augmented Generation for Knowledge-Intensive. Advances in Neural Information Processing Systems (NeurIPS) , year =

  13. [21]

    Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP) , year =

    The Troubling Emergence of Hallucination in Large Language Models -- An Extensive Definition, Quantification, and Prescriptive Remediations , author =. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP) , year =

    2023

  14. [22]

    Yin, Shukang and Fu, Chaoyou and Zhao, Sirui and Xu, Tong and Wang, Hao and Sui, Dianbo and Shen, Yunhang and Li, Ke and Sun, Xing and Chen, Enhong , journal =

  15. [23]

    Advances in Neural Information Processing Systems (NeurIPS) , year =

    Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting , author =. Advances in Neural Information Processing Systems (NeurIPS) , year =

  16. [24]

    arXiv preprint arXiv:2307.13702 , year =

    Measuring Faithfulness in Chain-of-Thought Reasoning , author =. arXiv preprint arXiv:2307.13702 , year =

    arXiv

  17. [25]

    Faithful Chain-of-Thought Reasoning , author =. Proceedings of the 13th International Joint Conference on Natural Language Processing and the 3rd Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (IJCNLP-AACL) , year =

  18. [26]

    Findings of the Association for Computational Linguistics: EMNLP 2024 , year =

    Measuring and Improving Faithfulness of Chain-of-Thought Reasoning , author =. Findings of the Association for Computational Linguistics: EMNLP 2024 , year =

    2024

  19. [27]

    Findings of the Association for Computational Linguistics: EMNLP 2023 , year =

    Evaluating Verifiability in Generative Search Engines , author =. Findings of the Association for Computational Linguistics: EMNLP 2023 , year =

    2023

  20. [28]

    Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP) , year =

    Enabling Large Language Models to Generate Text with Citations , author =. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP) , year =

    2023

  21. [29]

    Correctness is not Faithfulness in

    M. Correctness is not Faithfulness in. arXiv preprint arXiv:2412.18004 , year =

    arXiv

  22. [30]

    International Conference on Learning Representations (ICLR) , year =

    On the Self-Verification Limitations of Large Language Models on Reasoning and Planning Tasks , author =. International Conference on Learning Representations (ICLR) , year =

  23. [31]

    Pride and Prejudice:

    Xu, Wenda and Zhu, Guanglei and Zhao, Xuandong and Pan, Liangming and Li, Lei and Wang, William Yang , booktitle =. Pride and Prejudice:

  24. [32]

    Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like

    Chen, Xingyu and Xu, Jiahao and Liang, Tian and He, Zhiwei and Pang, Jianhui and Yu, Dian and Song, Linfeng and Liu, Qiuzhi and Zhou, Mengfei and Zhang, Zhuosheng and Wang, Rui and Tu, Zhaopeng and Mi, Haitao and Yu, Dong , booktitle =. Do NOT Think That Much for 2+3=? On the ...

  25. [33]

    Don't Overthink It: Preferring Shorter Thinking Chains for Improved

    Hassid, Michael and Remez, Tal and Gehring, Jonas and Schwartz, Roy and Adi, Yossi , journal =. Don't Overthink It: Preferring Shorter Thinking Chains for Improved

  26. [34]

    doi:10.48550/arXiv.2603.15030 , url =

    Zhu, Xuanyu and Dong, Yuhao and Wang, Rundong and Shi, Yang and Wu, Zhipeng and Peng, Yinlun and Zhang, YiFan and Lou, Yihang and Zhang, Yuanxing and Liu, Ziwei and Bai, Yan and Zhou, Yuan , year =. doi:10.48550/arXiv.2603.15030 , url =. 2603.15030 , archivePrefix =

    doi

  27. [35]

    Advances in Neural Information Processing Systems , volume =

    Are We on the Right Way for Evaluating Large Vision-Language Models? , author =. Advances in Neural Information Processing Systems , volume =. 2024 , url =

    2024

  28. [36]

    2024 , doi =

    Yue, Xiang and Ni, Yuansheng and Zhang, Kai and Zheng, Tianyu and Liu, Ruoqi and Zhang, Ge and Stevens, Samuel and Jiang, Dongfu and Ren, Weiming and Sun, Yuxuan and Wei, Cong and Yu, Botao and Yuan, Ruibin and Sun, Renliang and Yin, Ming and Zheng, Boyuan and Yang, Zhenzhu an...

    2024

  29. [37]

    International Conference on Learning Representations (ICLR) , year=

    MC-Search: Evaluating and Enhancing Multimodal Agentic Search with Structured Long Reasoning Chains , author=. International Conference on Learning Representations (ICLR) , year=

  30. [38]

    Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages =

    Yue, Xiang and Zheng, Tianyu and Ni, Yuansheng and Wang, Yubo and Zhang, Kai and Tong, Shengbang and Sun, Yuxuan and Yu, Botao and Zhang, Ge and Sun, Huan and Su, Yu and Chen, Wenhu and Neubig, Graham , editor =. Proceedings of the 63rd Annual Meeting of the Association for Co...

    2025 doi

  31. [39]

    Hao, Yunzhuo and Gu, Jiawei and Wang, Huichen Will and Li, Linjie and Yang, Zhengyuan and Wang, Lijuan and Cheng, Yu , booktitle =. Can. 2025 , editor =

    2025

  32. [40]

    An Empirical Study of Reasoning Length and Correctness in

    Han, Yuyang and others , journal =. An Empirical Study of Reasoning Length and Correctness in

  33. [41]

    Transactions on Machine Learning Research (TMLR) , year =

    Stop Overthinking: A Survey on Efficient Reasoning for Large Language Models , author =. Transactions on Machine Learning Research (TMLR) , year =

  34. [42]

    Jiang, Dongzhi and Zhang, Renrui and Guo, Ziyu and Li, Yanmin and Qi, Yu and Chen, Xinyu and Wang, Liuhui and Jin, Jianhan and Guo, Claire and Yan, Shen and Zhang, Bo and Fu, Chaoyou and Zhang, Peng and Li, Xiangyu and Luo, Gao and Li, Hongsheng , journal =

  35. [43]

    Li, Mingxiang and Zhong, Jiaying and Zhao, Shengqiong and Zhang, Haoquan and Lin, Shuhuai and Lai, Yuxin and Chen, Wenhu , journal=

  36. [44]

    Database Theory---ICDT 2001 , series =

    Why and Where: A Characterization of Data Provenance , author =. Database Theory---ICDT 2001 , series =. 2001 , doi =

    2001

  37. [45]

    Future Generation Computer Systems , volume =

    The Open Provenance Model Core Specification (v1.1) , author =. Future Generation Computer Systems , volume =. 2011 , doi =

    2011

  38. [46]

    Dapper, a Large-Scale Distributed Systems Tracing Infrastructure , author =

  39. [47]

    2026 , eprint =

    Claim-Level Rubric Rewards for Video Caption Reinforcement Learning , author =. 2026 , eprint =

    2026

  40. [48]

    2601.08276 , archiveprefix =

    Yao, Zhiyuan and Xu, Zishan and Guo, Yifu and Han, Zhiguang and Yang, Cheng and Zhang, Shuo and Zhang, Weinan and Zeng, Xingshan and Liu, Weiwen , year =. 2601.08276 , archiveprefix =

    arXiv

  41. [49]

    2601.06966 , archiveprefix =

    Bian, Haonan and Yao, Zhiyuan and Hu, Sen and Xu, Zishan and Zhang, Shaolei and Guo, Yifu and Yang, Ziliang and Han, Xueran and Wang, Huacan and Chen, Ronghao , year =. 2601.06966 , archiveprefix =

  42. [50]

    2025 , eprint =

    Octopus: Agentic Multimodal Reasoning with Six-Capability Orchestration , author =. 2025 , eprint =

    2025

  43. [51]

    2508.02085 , archiveprefix =

    Lin, Jiaye and Guo, Yifu and Han, Yuzhen and Hu, Sen and Ni, Ziyi and Wang, Licheng and Chen, Mingguang and Jiang, Daxin and Jiao, Binxing and Hu, Chen and Wang, Huacan , year =. 2508.02085 , archiveprefix =

  44. [52]

    2511.16077 , archiveprefix =

    Xu, Zishan and Guo, Yifu and Lu, Yuquan and Yang, Fengyu and Li, Junxin , year =. 2511.16077 , archiveprefix =

  45. [53]

    2025 , eprint =

    Decoupling Continual Semantic Segmentation , author =. 2025 , eprint =

    2025

  46. [54]

    Du, Enjun and Li, Xunkai and Jin, Tian and Zhang, Zhihan and Li, Rong-Hua and Wang, Guoren , booktitle =

Pith tools

Reviewed July 31, 2026 · model on record in the stance chip above.