Behavioral Controllability of Agentic Models for Information Extraction: From Fixed Workflows to Reflective Agents

Hongwei Feng; Lujia Zhang; Xingzhou Chen

REVIEW 3 major objections 3 minor 13 references

Reflective agents change extraction process far more than final records.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 22:29 UTC pith:W7HZ6QTN

load-bearing objection Honest but thin: the process-centric framing is the real contribution, while the one-run, no-gold-standard evidence can't support the coverage-increase claim. the 3 major comments →

arxiv 2607.15715 v2 pith:W7HZ6QTN submitted 2026-07-17 cs.AI

Behavioral Controllability of Agentic Models for Information Extraction: From Fixed Workflows to Reflective Agents

Lujia Zhang , Xingzhou Chen , Hongwei Feng This is my paper

classification cs.AI

keywords behavioral controllabilityLLM agentsinformation extractionreflectionmemoryretryReActdataset mention detection

verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper compares a fixed LLM workflow to reflective agent variants on the task of extracting dataset mentions from conference papers. It finds that adding reflection, retry, and memory produces a modest increase in extracted records—158 records from the workflow versus 165–168 from the agents—but a dramatic increase in observable process activity: agent logs contain thousands of reflection and retry events absent from the workflow. The paper argues that this process-level change, not the record count, is the primary effect of agentic components, and it introduces 'behavioral controllability' as an evaluation target. The value of the work is in reframing how agentic extraction systems should be assessed and tuned.

Core claim

On a NeurIPS 2024 corpus of up to 50 papers, the fixed workflow S0 produces 158 dataset records from 42 papers, the rule-reflection agent S1a produces 165 records from 43 papers, and the LLM-reflection agent S1b produces 168 records from 43 papers. Link grounding remains low and non-monotonic across all conditions (below 20%). The agent logs are about seven times larger than the workflow log (50,189 vs 7,084 lines), containing thousands of reflection, observation, and retry markers. The paper concludes that reflection and memory produce clear behavioral changes and modest coverage gains, while the absence of a human-annotated gold standard prevents claims about precision or recall. The main

What carries the argument

The central mechanism is the 'behavioral controllability' framework, operationalized through an AgentController that runs an Observe–Think–Act–Reflect–Learn–Adjust loop over tool actions. Reflection assigns quality scores to each action result, triggering bounded retries when scores fall below a threshold; memory stores short- and long-term experiences that can be injected into later prompts. The harness logs every action, reflection, retry, memory access, and run manifest, making process behavior measurable. The S2 design extends this with twelve atomic tools and dynamic tool selection, but its numerical evaluation is left to future work.

Load-bearing premise

Record count and field-completeness are treated as indicators of extraction quality without a human-annotated gold standard, so the additional records produced by the agents could be false positives; if they are, the claimed coverage increase is uninterpretable.

What would settle it

Take a stratified random sample of (paper, dataset) pairs from the S0 and S1b outputs, have annotators verify whether each dataset is genuinely mentioned in the paper and whether the record describes a dataset rather than a method or task. If the precision of agent-only records (those absent from S0) is at or below the precision of S0 records, the coverage gain claimed by the paper collapses.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

If agentic extraction is judged by process observability, then retry budgets, quality thresholds, and memory injection become explicit tuning knobs for cost versus coverage.
Rule-based reflection appears to capture most of the output gain; LLM reflection's measurable benefit is richer diagnostics, not more records, so it should be used selectively.
The persistent sub-20% link rate indicates that URL grounding needs dedicated verification tools, not just a field inside a broad extraction prompt.
The pairwise overlap results (136 shared pairs between S0 and S1a, 161 between S1a and S1b) suggest that agent variants mostly agree with each other but diverge from the workflow baseline, implying the harness changes which datasets get surfaced.
A human-annotated audit protocol, already outlined in the paper, is the necessary next step to convert coverage claims into accuracy claims.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The 165–168 vs 158 record difference may be within run-to-run noise, given single runs per condition; repeated runs with the same prompts could either shrink or amplify the reported gap.
The behavioral-controllability lens could generalize beyond dataset extraction to other scholarly IE tasks (e.g., method or contribution extraction), where trace-level debugging matters more than raw counts.
If the S2 design's dedicated link-search and evidence-verification tools were tested, a concrete prediction is that link rate would rise above 20% while record count stays roughly flat—this is a testable extension of the paper's failure-mode analysis.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

Honest but thin: the process-centric framing is the real contribution, while the one-run, no-gold-standard evidence can't support the coverage-increase claim.

read the letter

Here's my take. The paper is a clear, honest comparison of a fixed extraction workflow against ReAct-style agent variants with reflection and memory on the task of extracting dataset mentions from NeurIPS 2024 PDFs. What's genuinely useful is the framing: instead of only reporting record counts, the authors log tool calls, retries, reflection events, and memory injections, and they treat process observability as a first-class evaluation target. That is the right instinct, and the case studies from the logs are illustrative. The paper is also transparent about its limits: no gold standard, single-run results, and an explicit statement that the outcome metrics are not precision/recall.

The empirical core, though, is too thin to carry the main claim. The headline result—158 records for S0, 165 for S1a, 168 for S1b—comes from exactly one run per condition. LLM extraction is stochastic; a 10-record difference out of ~160 is safely inside run-to-run noise. The stress-test note is right: without repeated runs or confidence intervals, the 'agents increase coverage' conclusion in the abstract and Section 6 is not supported. The paper's caveat that the analysis is 'descriptive' softens this, but it does not fix the mismatch between that language and the causal-sounding 'increase.'

Two more soft spots. First, S1a and S1b share a single cumulative agent log—Table 2 says so explicitly—so the claim that LLM reflection changes process behavior more than rule reflection, the only S1a/S1b difference, cannot be evaluated from these data. The authors admit this, but it means the process-level comparison collapses. Second, S2, which the abstract and contributions present as a headline optimized agent, was never run. It is a design proposal, not a result. And without a gold standard, the record-count gains could be mostly false positives; the paper says this, but it still leans on record-count as 'coverage.'

So what do we have? A useful pilot and a sensible evaluation framework, written by people who know the literature and are honest about what they can and cannot claim. It is not a solid empirical contribution as is. It deserves a serious referee only if the authors can add repeated runs, a human-validated sample, and fix the shared-log problem. As it stands, I would not cite the numeric results, but I would recommend the framework to anyone building agent harnesses for extraction. Send it out, but expect major revision.

Referee Report

3 major / 3 minor

Summary. The paper compares a fixed LLM extraction workflow (S0) with two reflective agent variants (S1a with rule reflection, S1b with rule plus LLM reflection) on a NeurIPS 2024 dataset-mention extraction task, and additionally specifies an optimized S2 agent design with twelve atomic tools and dynamic planning. The authors argue for a process-centric evaluation of 'behavioral controllability'—observability, configurability, reproducibility, and comparability—rather than outcome metrics alone. The reported June 6 runs show S0 producing 158 records, S1a 165, and S1b 168, with link rates below 20%; the agent logs are roughly seven times larger than the workflow log and contain reflection, retry, and memory signals. The paper concludes that agentic mechanisms increase extraction coverage and observability, but that the outcome gains are modest, and that S2 should be evaluated in follow-up work.

Significance. If the empirical claims were adequately supported, the paper would make a useful contribution by proposing a concrete process-level evaluation framework for LLM-based information-extraction agents, with a controlled corpus, shared output schema, explicit configuration parameters, and reproducible run manifests. The authors are commendably explicit about their limitations: they acknowledge the absence of a complete human-annotated gold standard, the single-run nature of the results, and the fact that S2 has no completed numerical evaluation. These are real strengths in transparency. However, the central comparative claims—that agentic components 'increase extraction coverage' and that reflection changes process behavior in measurable ways—currently rest on evidence that is too thin: one stochastic run per condition, no verification of whether additional records are true positives, and a shared agent log that prevents clean per-condition process-metric attribution. The process-observability claims are largely supported, but the comparative outcome claims are not yet established.

major comments (3)

[§5 Evaluation Metrics / §6 Overall System Comparison] The central coverage-increase claim rests on a single run per condition. Section 5 states that 'the current evaluation reports single-run descriptive results rather than repeated-run statistical estimates.' Table 1 shows S0=158, S1a=165, S1b=168 records; the S0-to-S1b gap is 10 records, about 6%. Given stochastic LLM decoding, PDF-parsing side effects, retry timing, and API variability, this gap is well within plausible run-to-run noise. Section 6 nevertheless concludes that 'the agent harness increases extraction coverage under the same paper and per-paper output budget.' To support this load-bearing comparative claim, the paper needs repeated runs, confidence intervals, or a paired statistical test across runs; without such evidence the ordering S0 vs S1a vs S1b is not empirically established.
[§6 Table 2] The outcome interpretation is unverifiable without a gold standard. The paper states in Section 5: 'We do not assume a complete human-annotated ground truth for all dataset mentions... the current evaluation treats record count, field completeness, and pairwise differences as task-completion indicators rather than definitive precision or recall.' Appendix B shows substantial symmetric differences (e.g., 22 S0-only vs 29 S1a-only pairs). If the extra records produced by S1a/S1b are mostly false positives—for example, method or benchmark names misclassified as datasets—the claimed coverage increase does not represent a task-completion improvement. The manual audit protocol is deferred to a 'follow-up submission version,' but a sample audit of condition-specific records is needed in the present paper before claiming that agentic components 'increase extraction coverage' as a quality-relevan
[§6 Table 2] Table 2 cannot support per-condition process differences between S1a and S1b. The caption and text explain that S1a and S1b append to a shared agent log, so the reflection, observation, and retry mention counts (1,967; 1,388; 2,222) are identical by construction and represent cumulative activity through the S1b run. Consequently, RQ2—'Do reflection and memory produce measurable changes in tool use, retries, error recovery'—is not answerable for the rule-reflection vs LLM-reflection contrast from this table. Only the outcome counts (165 vs 168) and qualitative case studies differentiate the conditions. The logs must be separated per condition or regenerated so that process metrics are attributable to each variant; otherwise the claimed process-level comparison of S1a and S1b is an artifact of the logging design.

minor comments (3)

[§6 Table 2] The caption should explicitly warn that the S1a and S1b rows are cumulative and identical by construction. Currently, a reader may misread them as independent per-condition measurements; the text does explain this, but the table itself is misleading without a prominent note.
[Figure 3] Figure 3 appears corrupted in the manuscript: the axis label reads '6 -40 100 160' and there are repeated 't' markers. The figure should be regenerated so that the plotted values and axis labels are legible.
[Abstract] The abstract says 'observable and controllable improvements' over fixed workflows, but the empirical evidence mainly supports observable behavioral differences, not demonstrated improvements in extraction quality. Suggest rephrasing to 'observable and controllable differences' or 'changes in behavior' to match the evidence actually presented.

Circularity Check

0 steps flagged

No circularity found: the paper is an empirical trace comparison with explicit validity limitations, not a parameter-fit or self-citation chain.

full rationale

The paper's central claims are empirical comparisons between a fixed workflow and agentic variants on a shared corpus, measured from archived JSONL outputs and logs. There is no fitted parameter renamed as a prediction, no uniqueness theorem imported from the authors' prior work, and no ansatz smuggled in via citation. The only self-citation (Chen et al. 2025) supplies the model backend and is not load-bearing for the derivation. The paper explicitly disclaims a gold standard and single-run inferential statistics: 'the current evaluation reports single-run descriptive results rather than repeated-run statistical estimates' (Section 5) and 'We do not assume a complete human-annotated ground truth...' (Section 5). These are validity threats, not circularity. The shared-log issue in Table 2 means process metrics for S1a and S1b are identical because both conditions append to the same log; the paper acknowledges this and does not use Table 2 to claim a per-condition process difference. The 'controllability' framing is definitional—controllability is operationalized through observability, configurability, reproducibility, and comparability—but it is presented as an evaluation framework rather than as a derived prediction. Thus no load-bearing step reduces to its own inputs. Score 0 reflects that the derivation chain is self-contained and non-circular, while the acknowledged limitations are about reliability and external validity, not circularity.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 2 invented entities

The paper contributes a small empirical comparison whose conclusions are heavily moderated by domain assumptions (no gold standard, single runs, single corpus, single model). The only ad hoc knobs are standard engineering controls (retry budget, quality threshold, output caps), all listed transparently in Table 3. The invented entities are a conceptual framework and an unimplemented architecture, neither of which has independent falsifiable evidence.

free parameters (4)

QUALITY_RETRY_THRESHOLD = 0.6
Ad hoc threshold determining when rule reflection triggers a retry. Global, not fitted to data, but affects all retry behavior and thus the process/outcome comparison.
MAX_RETRIES = 2
Hand-chosen retry budget that bounds the agent's recovery attempts; a different budget would change record counts and trace statistics.
MAX_DATASETS_PER_PAPER = 5
Per-paper output cap that compresses high-density papers and directly shapes aggregate record counts and the S0-vs-agent differences.
MAX_PAPERS = 50
Corpus size cap; the entire study is a 50-paper sample chosen by availability, not representativeness.

axioms (4)

domain assumption The June 6 run package is a reliable basis for comparing conditions.
All empirical conclusions rest on single-run descriptive results, acknowledged in Section 5 as 'single-run descriptive results rather than repeated-run statistical estimates'.
domain assumption Record count, field completeness, and pairwise differences are meaningful task-completion indicators without a gold standard.
Explicitly stated in Section 5: absent a human-labeled gold standard, coverage indicators are used instead of precision or recall. This is the load-bearing epistemic premise.
domain assumption openPangu-Embedded-7B is a representative LLM for agentic extraction.
All runs use a single 7B model served via vLLM; the paper itself lists transfer to other model families as a threat to validity (Section 7).
domain assumption PDF text extraction captures the evidence needed to find datasets.
The task treats dataset evidence as present in the parsed text window; the authors list PDF parsing quality and extracted-text coverage as limitations (Section 7).

invented entities (2)

Behavioral controllability (observability, configurability, reproducibility, comparability) no independent evidence
purpose: Conceptual evaluation framework proposed as a first-class target for agent evaluation.
A framing construct, not a measurable entity; it has no falsifiable handle outside the paper's own metrics.
S2 optimized agent with twelve atomic tools no independent evidence
purpose: Proposed solution design for the failure modes observed in S1 (link grounding, section-level retrieval).
S2 is specified but never run; it produces no data and no predictions that could be tested against the world, only design conjectures.

pith-pipeline@v1.3.0-alltime-deepseek · 11555 in / 7901 out tokens · 47827 ms · 2026-08-01T22:29:37.752713+00:00 · methodology

0 comments

read the original abstract

Large language model (LLM) agents are increasingly used for complex information-extraction tasks, yet it remains unclear whether agentic components such as reflection and memory lead to observable and controllable improvements over fixed LLM workflows. We study this question through conference-paper dataset extraction, where a system must identify datasets mentioned in scholarly PDFs and produce structured records. We compare a fixed workflow baseline with reflective agent variants and specify an optimized agent condition (S2) that extends the same task with richer PDF tools and dynamic tool selection. Our evaluation emphasizes process-level behavior--including tool execution, retries, reflection, memory use, runtime, and failure recovery--while treating extraction coverage and field completeness as secondary outcome measures. The paper characterizes when agentic mechanisms change system behavior, whether these changes improve task completion, and how the observed failure modes motivate an optimized agent design under the same evaluation harness.

Figures

Figures reproduced from arXiv: 2607.15715 by Hongwei Feng, Lujia Zhang, Xingzhou Chen.

**Figure 2.** Figure 2: System overview. S0 uses a fixed extraction sequence, S1 wraps the same task with observable agent actions, reflec [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗

**Figure 3.** Figure 3: Outcome trends across the completed June 6 conditions. Agentic variants increase record count and datasets per paper [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

discussion (0)

Reference graph

Works this paper leans on

13 extracted references · 3 linked inside Pith

[1]

International Conference on Learning Representations , year =

Yao, Shunyu and Zhao, Jeffrey and Yu, Dian and Du, Nan and Shafran, Izhak and Narasimhan, Karthik and Cao, Yuan , title =. International Conference on Learning Representations , year =
[2]

Toolformer: Language Models Can Teach Themselves to Use Tools , booktitle =

Schick, Timo and Dwivedi-Yu, Jane and Dess. Toolformer: Language Models Can Teach Themselves to Use Tools , booktitle =
[3]

Advances in Neural Information Processing Systems , volume =

Shinn, Noah and Cassano, Federico and Gopinath, Ashwin and Narasimhan, Karthik and Yao, Shunyu , title =. Advances in Neural Information Processing Systems , volume =
[4]

Advances in Neural Information Processing Systems , volume =

Madaan, Aman and Tandon, Niket and Gupta, Prakhar and Hallinan, Skyler and Gao, Luyu and Wiegreffe, Sarah and Alon, Uri and Dziri, Nouha and Prabhumoye, Shrimai and Yang, Yiming and Gupta, Shashank and Majumder, Bodhisattwa Prasad and Hermann, Katherine and Welleck, Sean and Yazdanbakhsh, Amir and Clark, Peter , title =. Advances in Neural Information Pro...
[5]

International Conference on Learning Representations , year =

Huang, Jie and Chen, Xinyun and Mishra, Swaroop and Zheng, Huaixiu Steven and Yu, Adams Wei and Song, Xinying and Zhou, Denny , title =. International Conference on Learning Representations , year =
[6]

and Cai, Carrie J

Park, Joon Sung and O'Brien, Joseph C. and Cai, Carrie J. and Morris, Meredith Ringel and Liang, Percy and Bernstein, Michael S. , title =. Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology , pages =
[7]

and Stoica, Ion and Gonzalez, Joseph E

Packer, Charles and Wooders, Sarah and Lin, Kevin and Fang, Vivian and Patil, Shishir G. and Stoica, Ion and Gonzalez, Joseph E. , title =. arXiv preprint arXiv:2310.08560 , year =

Pith/arXiv arXiv
[8]

International Conference on Learning Representations , year =

Liu, Xiao and Yu, Hao and Zhang, Hanchen and Xu, Yifan and Lei, Xuanyu and Lai, Hanyu and Gu, Yu and Ding, Hangliang and Men, Kaiwen and Yang, Kejuan and Zhang, Shudan and Deng, Xiang and Zeng, Aohan and Du, Zhengxiao and Zhang, Chenhui and Shen, Sheng and Zhang, Tianjun and Su, Yu and Sun, Huan and Huang, Minlie and Dong, Yuxiao and Tang, Jie , title =. ...
[9]

and Zhu, Hao and Zhou, Xuhui and Lo, Robert and Sridhar, Abishek and Cheng, Xianyi and Ou, Tianyue and Bisk, Yonatan and Fried, Daniel and Alon, Uri and Neubig, Graham , title =

Zhou, Shuyan and Xu, Frank F. and Zhu, Hao and Zhou, Xuhui and Lo, Robert and Sridhar, Abishek and Cheng, Xianyi and Ou, Tianyue and Bisk, Yonatan and Fried, Daniel and Alon, Uri and Neubig, Graham , title =. arXiv preprint arXiv:2307.13854 , year =

Pith/arXiv arXiv
[10]

, title =

Lo, Kyle and Wang, Lucy Lu and Neumann, Mark and Kinney, Rodney and Weld, Daniel S. , title =. Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics , pages =
[11]

Transactions of the Association for Computational Linguistics , volume =

Pan, Huitong and Zhang, Qi and Dragut, Eduard and Caragea, Cornelia and Latecki, Longin Jan , title =. Transactions of the Association for Computational Linguistics , volume =. 2023 , doi =

2023
[12]

Data , volume =

Heddes, Jenny and Meerdink, Pim and Pieters, Miguel and Marx, Maarten , title =. Data , volume =. 2021 , doi =

2021
[13]

arXiv preprint arXiv:2505.22375 , year=

Pangu embedded: An efficient dual-system llm reasoner with metacognition , author=. arXiv preprint arXiv:2505.22375 , year=

Pith/arXiv arXiv

[1] [1]

International Conference on Learning Representations , year =

Yao, Shunyu and Zhao, Jeffrey and Yu, Dian and Du, Nan and Shafran, Izhak and Narasimhan, Karthik and Cao, Yuan , title =. International Conference on Learning Representations , year =

[2] [2]

Toolformer: Language Models Can Teach Themselves to Use Tools , booktitle =

Schick, Timo and Dwivedi-Yu, Jane and Dess. Toolformer: Language Models Can Teach Themselves to Use Tools , booktitle =

[3] [3]

Advances in Neural Information Processing Systems , volume =

Shinn, Noah and Cassano, Federico and Gopinath, Ashwin and Narasimhan, Karthik and Yao, Shunyu , title =. Advances in Neural Information Processing Systems , volume =

[4] [4]

Advances in Neural Information Processing Systems , volume =

Madaan, Aman and Tandon, Niket and Gupta, Prakhar and Hallinan, Skyler and Gao, Luyu and Wiegreffe, Sarah and Alon, Uri and Dziri, Nouha and Prabhumoye, Shrimai and Yang, Yiming and Gupta, Shashank and Majumder, Bodhisattwa Prasad and Hermann, Katherine and Welleck, Sean and Yazdanbakhsh, Amir and Clark, Peter , title =. Advances in Neural Information Pro...

[5] [5]

International Conference on Learning Representations , year =

Huang, Jie and Chen, Xinyun and Mishra, Swaroop and Zheng, Huaixiu Steven and Yu, Adams Wei and Song, Xinying and Zhou, Denny , title =. International Conference on Learning Representations , year =

[6] [6]

and Cai, Carrie J

Park, Joon Sung and O'Brien, Joseph C. and Cai, Carrie J. and Morris, Meredith Ringel and Liang, Percy and Bernstein, Michael S. , title =. Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology , pages =

[7] [7]

and Stoica, Ion and Gonzalez, Joseph E

Packer, Charles and Wooders, Sarah and Lin, Kevin and Fang, Vivian and Patil, Shishir G. and Stoica, Ion and Gonzalez, Joseph E. , title =. arXiv preprint arXiv:2310.08560 , year =

Pith/arXiv arXiv

[8] [8]

International Conference on Learning Representations , year =

Liu, Xiao and Yu, Hao and Zhang, Hanchen and Xu, Yifan and Lei, Xuanyu and Lai, Hanyu and Gu, Yu and Ding, Hangliang and Men, Kaiwen and Yang, Kejuan and Zhang, Shudan and Deng, Xiang and Zeng, Aohan and Du, Zhengxiao and Zhang, Chenhui and Shen, Sheng and Zhang, Tianjun and Su, Yu and Sun, Huan and Huang, Minlie and Dong, Yuxiao and Tang, Jie , title =. ...

[9] [9]

and Zhu, Hao and Zhou, Xuhui and Lo, Robert and Sridhar, Abishek and Cheng, Xianyi and Ou, Tianyue and Bisk, Yonatan and Fried, Daniel and Alon, Uri and Neubig, Graham , title =

Zhou, Shuyan and Xu, Frank F. and Zhu, Hao and Zhou, Xuhui and Lo, Robert and Sridhar, Abishek and Cheng, Xianyi and Ou, Tianyue and Bisk, Yonatan and Fried, Daniel and Alon, Uri and Neubig, Graham , title =. arXiv preprint arXiv:2307.13854 , year =

Pith/arXiv arXiv

[10] [10]

, title =

Lo, Kyle and Wang, Lucy Lu and Neumann, Mark and Kinney, Rodney and Weld, Daniel S. , title =. Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics , pages =

[11] [11]

Transactions of the Association for Computational Linguistics , volume =

Pan, Huitong and Zhang, Qi and Dragut, Eduard and Caragea, Cornelia and Latecki, Longin Jan , title =. Transactions of the Association for Computational Linguistics , volume =. 2023 , doi =

2023

[12] [12]

Data , volume =

Heddes, Jenny and Meerdink, Pim and Pieters, Miguel and Marx, Maarten , title =. Data , volume =. 2021 , doi =

2021

[13] [13]

arXiv preprint arXiv:2505.22375 , year=

Pangu embedded: An efficient dual-system llm reasoner with metacognition , author=. arXiv preprint arXiv:2505.22375 , year=

Pith/arXiv arXiv