pith. machine review for the scientific record. sign in

arxiv: 2604.10989 · v1 · submitted 2026-04-13 · 💻 cs.AI

Recognition: unknown

MAFIG: Multi-agent Driven Formal Instruction Generation Framework

Shixing Zhao , Zheng Si , Pengpeng Ouyang , Zhengqing Hu , Wanqi Zhu , Dong Chen , Yibo Guo , Mingliang Xu
Authors on Pith no claims yet

Pith reviewed 2026-05-10 15:51 UTC · model grok-4.3

classification 💻 cs.AI
keywords multi-agent systemsemergency schedulinglarge language modelsknowledge distillationformal instructionslocal decision makingscheduling robustness
0
0 comments X

The pith

MAFIG confines emergency decisions in scheduling systems to affected local modules and generates formal instructions via multi-agent distillation for rapid repair.

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

The paper introduces MAFIG to address emergency-induced failures in scheduling systems that traditional rule-based or full-rescheduling methods struggle to handle due to unpredictability. It restricts the decision scope to local functional modules, deploys a Perception Agent and an Emergency Decision Agent to produce formal instructions, and applies span-focused loss-driven local distillation to move capability from heavy cloud models to fast local ones. This combination aims to cut inference latency while retaining effectiveness. Tests across Port, Warehousing, and Deck datasets report success rates of 98.49%, 94.97%, and 97.50% with average times of 0.33 s, 0.23 s, and 0.19 s. The approach seeks to improve system robustness by avoiding lengthy global contexts and predefined emergency catalogs.

Core claim

MAFIG demonstrates that limiting emergency response to local functional modules, combined with multi-agent generation of formal instructions and span-focused distillation from cloud LLMs to lightweight models, repairs scheduling logic rapidly without requiring full system rescheduling or anticipation of every disruption, as shown by the reported success rates and processing times on three scheduling datasets.

What carries the argument

The MAFIG framework, which uses a Perception Agent and an Emergency Decision Agent to generate formal instructions for affected local scheduling modules, supported by span-focused loss-driven local distillation (SFL) that transfers decision capability while lowering latency.

If this is right

  • Scheduling systems become more robust to diverse, unforeseen emergencies without exhaustive rule sets or global recomputation.
  • Local agents with distilled models deliver sub-second responses while keeping decision quality close to full cloud models.
  • Formal instructions enable verifiable, machine-readable fixes that integrate directly into existing scheduling engines.
  • The method scales to multiple concurrent emergencies by handling each within its own local scope.

Where Pith is reading between the lines

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

  • The local-scope design could lower the communication overhead in distributed scheduling platforms where global state sharing is costly.
  • Formal instructions might allow safety-critical domains to add automated verification steps before applying agent-generated repairs.
  • Extending the same agent-plus-distillation pattern to other real-time control problems, such as traffic signal adjustment during incidents, appears straightforward.
  • If the formal language is kept simple, human operators could review or override instructions with minimal training.

Load-bearing premise

Isolating decisions to only the local modules directly hit by an emergency, together with formal instruction generation, is enough to restore correct scheduling behavior without creating inconsistencies that require broader system knowledge.

What would settle it

A test case in which an emergency in one module produces a dependency conflict that cannot be resolved by local changes alone, causing the repaired schedule to fail validation or propagate errors.

Figures

Figures reproduced from arXiv: 2604.10989 by Dong Chen, Mingliang Xu, Pengpeng Ouyang, Shixing Zhao, Wanqi Zhu, Yibo Guo, Zhengqing Hu, Zheng Si.

Figure 1
Figure 1. Figure 1: Accident types in Mediterranean port areas. they often significantly disrupt the overall scheduling plan [3]. As illustrated in [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Architecture of MAFIG for emergency decision in scheduling systems. The framework consists of the Perception Agent, the Emergency Decision Agent and the Atomic Function Library. It supports semantic parsing, impact analysis, affected function localization and atomic function revision, which enables rapid recovery of scheduling logic under emergency situations [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of the EvalPort evaluation dataset. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Overview of the EvalWare evaluation dataset. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Overview of the EvalDeck evaluation datasets. However, its average processing times still reach 27.44 s and 40.75 s. These results indicate that Large Language Models still suffer from the pronounced Latency-Quality Tradeoff in scheduling tasks. Moreover, large-scale models are generally difficult to deploy directly in local scheduling systems and must therefore be accessed through cloud APIs in practical … view at source ↗
Figure 6
Figure 6. Figure 6: Performance comparison of SFL and LoRA under different training set sizes across tasks. Emergency description: In the aircraft carrier deck scheduling system, the position of hydraulic vehicle No. 2 is adjusted to (0, 1), maintenance vehicle No. 5 becomes unavailable due to a failure, oxygen-supply vehicle No. 3 becomes unavailable due to a failure, an explosion occurs in the area covering positions (8, 5)… view at source ↗
Figure 7
Figure 7. Figure 7: Case study of MAFIG in the deck scheduling scenario under concurrent emergency situations. unchanged contextual code still accounts for the majority of the sequence. Consequently, LoRA tends to disperse parameter updates over irrelevant background code, which weakens learning of the core modification logic. As scenario complexity increases, the advantage of SFL becomes more pronounced. In the warehouse sce… view at source ↗
read the original abstract

Emergency situations in scheduling systems often trigger local functional failures that undermine system stability and even cause system collapse. Existing methods primarily rely on robust scheduling or reactive scheduling, handling emergencies through predefined rules or rescheduling strategies. However, the diversity and unpredictability of real-world emergencies make them difficult to anticipate, which limits the adaptability of these methods in complex scenarios. Recent studies have shown that Large Language Models (LLMs) possess strong potential for complex scheduling tasks because of their extensive prior knowledge and strong reasoning capabilities. Nevertheless, the high inference latency of LLMs and the lengthy contextual information of scheduling systems significantly hinder their application for emergency handling. To mitigate these issues, we propose the Multi-agent Driven Formal Instruction Generation Framework (MAFIG). The framework constrains the decision scope to local functional modules affected by emergency situations and repairs scheduling logic rapidly by generating formal instructions. MAFIG contains a Perception Agent and an Emergency Decision Agent, which mitigates the adverse impact of lengthy system contexts on emergency decision-making. We further introduce span-focused loss-driven local distillation mechanism (SFL) to transfer the decision-making capability of powerful Cloud Large Language Models (C-LLMs) to lightweight local models, reducing inference latency while preserving decision-making effectiveness. Experiments in the Port, Warehousing, and Deck scheduling datasets show success rates of 98.49\%, 94.97\%, and 97.50\%, with average processing times of 0.33 s, 0.23 s, and 0.19 s. These results demonstrate that MAFIG effectively mitigates the impact of emergencies and improves the robustness and adaptability of scheduling systems.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript proposes the Multi-agent Driven Formal Instruction Generation Framework (MAFIG) for handling unpredictable emergencies in scheduling systems. It constrains decision-making to local functional modules via a Perception Agent and Emergency Decision Agent, repairs logic through generated formal instructions, and applies span-focused loss-driven local distillation (SFL) to transfer capabilities from cloud LLMs to lightweight local models for reduced latency. Experiments on Port, Warehousing, and Deck scheduling datasets report success rates of 98.49%, 94.97%, and 97.50% with average processing times of 0.33 s, 0.23 s, and 0.19 s, claiming improved robustness and adaptability compared to traditional robust or reactive scheduling approaches.

Significance. If the results can be substantiated with baselines, global consistency metrics, and component ablations, MAFIG could provide a practical method for real-time emergency response in logistics and operations research by combining modular LLM reasoning with efficient local execution. The SFL distillation mechanism specifically addresses a key deployment barrier for LLMs in latency-sensitive scheduling, representing a concrete engineering contribution.

major comments (3)
  1. [Abstract] Abstract: The reported success rates of 98.49%, 94.97%, and 97.50% are presented without defining the success criterion, reporting dataset sizes or test-set statistics, or providing any baseline comparisons to the robust scheduling or reactive rescheduling methods discussed in the introduction. This prevents evaluation of whether the numbers support the central claim of improved adaptability.
  2. [Framework Design] Framework description: The core design isolates repairs to local modules affected by an emergency and uses formal instructions for logic repair, yet no analysis, feasibility metrics, or counter-example checks are supplied to confirm that such local repairs avoid global inconsistencies (e.g., resource conflicts or cascading delays across coupled flows). This assumption is load-bearing for the robustness claim.
  3. [Experiments] Experiments section: No ablation studies, variance statistics, or worst-case latency figures are reported for the multi-agent components or SFL mechanism. Average processing times alone do not establish real-time suitability or isolate the contribution of each proposed element to the observed success rates.
minor comments (2)
  1. [Abstract] The term 'formal instructions' is introduced without specifying the target formal language, syntax, or verification procedure, which would aid reproducibility and allow readers to assess the claimed precision of the generated repairs.
  2. [Framework Design] A high-level diagram showing data flow among the Perception Agent, Emergency Decision Agent, SFL distillation, and the underlying scheduling system would improve clarity of the overall architecture.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the thoughtful and constructive report. The comments highlight important areas where additional clarity, analysis, and experimental rigor will strengthen the manuscript. We address each major comment below and commit to a major revision that incorporates the requested elements.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The reported success rates of 98.49%, 94.97%, and 97.50% are presented without defining the success criterion, reporting dataset sizes or test-set statistics, or providing any baseline comparisons to the robust scheduling or reactive rescheduling methods discussed in the introduction. This prevents evaluation of whether the numbers support the central claim of improved adaptability.

    Authors: We agree that the abstract as written does not supply these details. In the revised manuscript we will expand the abstract to (i) define the success criterion as the local repair of the affected scheduling module that restores feasible operation without immediate system collapse, (ii) report the number of emergency scenarios and test-set sizes for each dataset, and (iii) include concise baseline comparisons against the robust-scheduling and reactive-rescheduling approaches referenced in the introduction. These additions will be backed by the quantitative results already obtained in the experiments section. revision: yes

  2. Referee: [Framework Design] Framework description: The core design isolates repairs to local modules affected by an emergency and uses formal instructions for logic repair, yet no analysis, feasibility metrics, or counter-example checks are supplied to confirm that such local repairs avoid global inconsistencies (e.g., resource conflicts or cascading delays across coupled flows). This assumption is load-bearing for the robustness claim.

    Authors: The referee correctly identifies that the manuscript provides no explicit verification of global consistency after local repairs. While the Perception and Emergency Decision Agents are intentionally scoped to affected modules, we did not include supporting analysis. We will add a dedicated subsection that presents (a) feasibility metrics quantifying the scope of each repair, (b) discussion of potential resource conflicts with illustrative examples drawn from the three scheduling domains, and (c) counter-example checks where feasible. Any cases where exhaustive global verification remains intractable will be acknowledged as a limitation with suggested directions for future work. revision: yes

  3. Referee: [Experiments] Experiments section: No ablation studies, variance statistics, or worst-case latency figures are reported for the multi-agent components or SFL mechanism. Average processing times alone do not establish real-time suitability or isolate the contribution of each proposed element to the observed success rates.

    Authors: We concur that the current experimental presentation is insufficient to isolate component contributions or to substantiate real-time claims. The revised experiments section will include: (1) ablation studies that systematically disable or replace the Perception Agent, Emergency Decision Agent, and the span-focused local distillation (SFL) mechanism; (2) standard-deviation and variance statistics for both success rates and processing times across repeated trials; and (3) worst-case latency figures in addition to the reported averages. These results will be presented in new tables and figures that directly address the referee’s concerns. revision: yes

Circularity Check

0 steps flagged

No circularity: framework description relies on experiments, not derivations or self-referential fits

full rationale

The paper describes the MAFIG framework, its agents, and a distillation mechanism (SFL) but contains no equations, derivations, or mathematical predictions. Central claims rest on reported success rates and latencies from three scheduling datasets. No load-bearing self-citations, fitted parameters renamed as predictions, or ansatzes smuggled via prior work are present. The design choices (local module isolation, formal instructions) are presented as engineering decisions justified by experimental outcomes rather than reducing to their own inputs by construction. This is a standard non-circular empirical paper.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The framework rests on the domain assumption that LLMs can reason effectively about scheduling when context is constrained, plus the practical claim that formal instructions suffice for local repairs; no free parameters or new physical entities are introduced.

axioms (1)
  • domain assumption Large Language Models possess strong potential for complex scheduling tasks because of their extensive prior knowledge and strong reasoning capabilities
    Invoked in the abstract to justify the use of LLMs for emergency handling.
invented entities (1)
  • MAFIG framework (Perception Agent + Emergency Decision Agent + SFL) no independent evidence
    purpose: To constrain LLM decision scope and transfer capability to lightweight models for low-latency emergency repair
    Newly proposed system components whose effectiveness is asserted via the reported experiments.

pith-pipeline@v0.9.0 · 5607 in / 1306 out tokens · 75156 ms · 2026-05-10T15:51:44.298945+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

41 extracted references · 10 canonical work pages · 5 internal anchors

  1. [1]

    G.Chen,J.Zhang,M.Ning,W.Cui,M.Ma,Taskschedulinginreal- time industrial scenarios, Comput. Ind. Eng. (2023) 109372

    2023

  2. [2]

    Agnetis, J.-C

    A. Agnetis, J.-C. Billaut, M. Pinedo, D. Shabtay, Fifty years of researchinscheduling—Theoryandapplications,EuropeanJournal of Operational Research (2025) 367–393

    2025

  3. [3]

    Ouelhadj, S

    D. Ouelhadj, S. Petrovic, A survey of dynamic scheduling in manu- facturing systems, Journal of Scheduling (2009) 417–431

    2009

  4. [4]

    J. Lu, W. Li, J. Guo, X. Ding, Z. Tang, T. Wang, W. Jia, Hybrid learning for cold-start-aware microservice scheduling in dynamic edgeenvironments,IEEETransactionsonMobileComputing(2025) 1–16

    2025

  5. [5]

    L. Liu, Z. Xu, X. Qu, A reconfigurable architecture for industrial control systems: Overview and challenges, Machines (2024) 793

    2024

  6. [6]

    J. M. Framinan, R. Leisten, R. Ruiz, Architecture of manufacturing scheduling systems: Literature review and an integrated proposal, European Journal of Operational Research 205 (2010) 237–246

    2010

  7. [7]

    N. M. Sadeh, D. W. Hildum, T. J. Laliberty, J. McA’Nulty, D. Kjenstad, A. Tseng, A blackboard architecture for integrating pro- cess planning and production scheduling, Concurrent Engineering: Research and Applications 6 (1998) 88–100

    1998

  8. [8]

    S. F. Smith, Reactive scheduling systems, Intelligent Scheduling Systems, Kluwer Academic Publishers (1994) 155–192

    1994

  9. [9]

    Roman, E

    M.Marino,L.Cavallaro,E.Castro,R.E.Musumeci,M.Martignoni, F. Roman, E. Foti, Analysis on a database of ship accidents in port areas, Data in Brief (2023) 109127

    2023

  10. [10]

    Ghaleb, H

    M. Ghaleb, H. Zolfagharinia, S. Taghipour, Real-time production scheduling in the Industry-4.0 context: Addressing uncertainties in job arrivals and machine breakdowns, Computers & Operations Re- search (2020) 105031

    2020

  11. [11]

    Herroelen, R

    W. Herroelen, R. Leus, Project scheduling under uncertainty: Survey and research potentials, European Journal of Operational Research (2005) 289–306

    2005

  12. [12]

    G.E.Vieira,J.W.Herrmann,E.Lin,Adaptiveproductionreschedul- ing for managing unforeseen emergency situations, Proceedings of the2003IEEEInternationalConferenceonRoboticsandAutomation (2003) 4011–4016

    2003

  13. [13]

    Llms can schedule,

    H. Abgaryan, G. Harutyunyan, T. Cazenave, LLMs can schedule, arXiv preprint arXiv:2408.06993, 2024

    work page arXiv 2024

  14. [14]

    M. Tang, C. Bian, L. Yang, X. Zhong, Key-concept thinking prompt- ing for improved reasoning in large language models,Neurocomput- ing656 (2025) 130986

    2025

  15. [15]

    X.Li,X.Zhou,J.Li,B.Fan,Retrieval-augmentedLLM-drivenmulti- agent optimization framework for intelligent manufacturing schedul- ing, in:Proceedings of the IEEE International Conference on High Performance Computing and Communications, 2025

    2025

  16. [16]

    7338–7346

    D.Chen,S.Zhang,F.Gao,Y.Zhuang,S.Tang,Q.Liu,M.Xu,Logic distillation: learning from code function by function for decision- makingtasks,in:ProceedingsoftheThirty-FourthInternationalJoint Conference on Artificial Intelligence, 2025, pp. 7338–7346

    2025

  17. [17]

    S.Brahmachary,S.M.Joshi,A.Panda,K.Koneripalli,A.K.Sagotra, H.Patel,A.Sharma,A.D.Jagtap,K.Kalyanaraman,Largelanguage model-based evolutionary optimizer: Reasoning with elitism,Neuro- computing622 (2025) 129272

    2025

  18. [18]

    T. B. Brown, B. Mann, N. Ryder, et al., Language models are few- shot learners, Advances in Neural Information Processing Systems 33 (2020) 1877–1901

    2020

  19. [19]

    J. Wei, X. Wang, D. Schuurmans, et al., Chain-of-thought prompting elicitsreasoninginlargelanguagemodels,AdvancesinNeuralInfor- mation Processing Systems 35 (2022) 24824–24837

    2022

  20. [20]

    Kojima, S

    T. Kojima, S. Gu, M. Reid, Y. Matsuo, Y. Iwasawa, Large language models are zero-shot reasoners, Advances in Neural Information Processing Systems 35 (2022) 22199–22213

    2022

  21. [21]

    D. Chen, F. Gao, S. Zhang, Y. Zhuang, S. Tang, Q. Liu, H. Wang, X. Yang, M. Xu, Improving large models with small models: Lower costs and better performance, Neural Netw. (2025) 108276

    2025

  22. [22]

    Demeulemeester, W

    E. Demeulemeester, W. Herroelen, Robust Project Scheduling, Found. Trends Technol. Inf. Oper. Manag. 3(3-4) (2009) 201–376

    2009

  23. [23]

    Portoleau, C

    T. Portoleau, C. Artigues, R. Guillaume, Robust Predictive-Reactive Scheduling: An Information-Based Decision Tree Model, in: Infor- mation Processing and Management of Uncertainty in Knowledge- Based Systems, CCIS 1239, Springer, Cham, 2020, pp. 479–492

    2020

  24. [24]

    Herroelen, R

    W. Herroelen, R. Leus, Robust and reactive project scheduling: A reviewandclassificationofprocedures,Int.J.Prod.Res.42(8)(2004) 1599–1620

    2004

  25. [25]

    G. Chai, J. Cao, W. Huang, J. Guo, Optimized traffic emergency resource scheduling using time varying rescue route travel time, Neurocomputing275 (2018) 1567–1575

    2018

  26. [26]

    Jędrzejowicz, E

    P. Jędrzejowicz, E. Ratajczak-Ropel, Reinforcement Learning strate- giesforA-TeamsolvingtheResource-ConstrainedProjectScheduling Problem,Neurocomputing146 (2014) 301–307

    2014

  27. [27]

    J. Wen, D. Liu, Y. Xie, Y. Ren, J. Wang, Y. Xia, P. Zhu, AcuGPT- Agent: An LLM-powered intelligent system for acupuncture-based infertility treatment, Neurocomputing 652 (2025) 131116

    2025

  28. [28]

    D. Chen, Y. Zhuang, S. Zhang, J. Liu, S. Dong, S. Tang, Data shunt: Collaboration of small and large models for lower costs and better performance, in:Proceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 10, 2024, pp. 11249–11257

    2024

  29. [29]

    D.Chen,Z.Hu,P.Fan,Y.Zhuang,Y.Li,Q.Liu,X.Jiang,M.Xu,Kka: Improving vision anomaly detection through anomaly-related knowl- edge from large language models, arXiv preprint arXiv:2502.14880, Shixing Zhao et al.:Preprint submitted to ElsevierPage 12 of 13 Multi-agent Driven Formal Instruction Generation Framework 2025

    work page arXiv 2025

  30. [30]

    Huang, J

    W. Huang, J. Pan, Z. Wang, Y. Liu, Y. Wang, S. Shen, J. Hu, Enhancing multimodal large language models with efficient feature alignment and processing using state space models,Neurocomputing 665 (2026) 132152

    2026

  31. [31]

    M. Chen, J. Tworek, H. Jun, Q. Yuan, H. P. de Oliveira Pinto, J. Kaplan, H. Edwards, Y. Burda, N. Joseph, G. Brockman, et al., Evaluating Large Language Models Trained on Code, arXiv preprint arXiv:2107.03374 (2021)

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  32. [32]

    B.Rozière,J.Gehring,F.Gloeckle,S.Sootla,I.Gat,X.E.Tan,Y.Adi, J. Liu, R. Sauvestre, T. Remez, et al., Code Llama: Open Foundation Models for Code, arXiv preprint arXiv:2308.12950 (2023)

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  33. [33]

    Abgaryan, T

    H. Abgaryan, T. Cazenave, A. Harutyunyan, Starjob: Dataset for LLM-driven Job Shop Scheduling, arXiv preprint arXiv:2503.01877 (2025)

    work page arXiv 2025

  34. [34]

    J.An,H.Cai,Y.Zhao,X.Gui,X.He,X.Jin,JSHM:Adynamicflex- ible job-shop scheduling method with human-machine collaboration, Neurocomputing666 (2026) 132213

    2026

  35. [35]

    S. Cao, Y. Yuan, ReflecSched: Solving Dynamic Flexible Job-Shop SchedulingviaLLM-PoweredHierarchicalReflection,arXivpreprint arXiv:2508.01724 (2025)

    work page arXiv 2025

  36. [36]

    Agarwal, Y

    V. Agarwal, Y. Pei, S. Alamir, X. Liu, CodeMirage: Hallucina- tions in Code Generated by Large Language Models, arXiv preprint arXiv:2408.08333 (2024)

    work page arXiv 2024

  37. [37]

    Rodrigues, A

    F. Rodrigues, A. Agra, Berth allocation and quay crane assign- ment/scheduling problem under uncertainty: A survey, Eur. J. Oper. Res. 303(2) (2022) 501–524

    2022

  38. [38]

    X. Wang, J. Liu, X. Su, H. Peng, X. Zhao, C. Lu, A review on carrier aircraftdispatchpathplanningandcontrolondeck,Chin.J.Aeronaut. 33(12) (2020) 3039–3057

    2020

  39. [39]

    Qwen Team, Qwen3 technical report, arXiv preprint arXiv:2505.09388, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  40. [40]

    DeepSeek-AI, DeepSeek-V3.2: Pushing the frontier of open large language models, arXiv preprint arXiv:2512.02556, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  41. [41]

    A. Zeng, B. Liu, R. Zheng, B. Zhang, F. Du, Z. Lu, Z. Lai, T. Ni, C. Shen, Y. Ding, et al., ChatGLM: A family of large lan- guage models from GLM-130B to GLM-4 all tools, arXiv preprint arXiv:2406.12793, 2024. Shixing Zhao et al.:Preprint submitted to ElsevierPage 13 of 13

    work page internal anchor Pith review arXiv 2024