PaperlayerView exact revision on arXiv

Source-linked AI summary

Understanding Issues, Causes and Solutions in Open-Source LLM-based Multi-Agent Systems

Asad Ur Rehman, Syed Mohammad Kashif, Ruiyin Li, Peng Liang, Zengyang Li, Arif Ali Khan

arXiv:2610.00905v1cs.SEcs.AI

TL;DR

Empirical evidence on challenges in developing and using open-source LLM-based MAS, their causes, and their solutions is limited. This study analyzes filtered GitHub issues from open-source MAS projects and finds orchestration and execution issues most common, with workflow, tool integration, and memory among the frequent causes. It derives practical implications for users, developers, and researchers from these findings.

  • Problem

    Empirical studies of challenges in developing and maintaining open-source LLM-based MAS, including their causes and solutions, remain limited.

    Source: §1 · p. 1

  • Method

    The study analyzed 944 filtered issues from 22,848 closed GitHub issues across qualifying open-source LLM-based MAS projects, extracting issues, causes, and solutions.

    Source: §7 · p. 28

  • Results

    Orchestration & Execution Issue is the most common issue, Workflow Problem, Tool Integration Problem, and Memory Problem are frequent causes, and Optimize Workflow and Adjust Configuration are predominant solutions.

    Source: §1 · p. 2

  • Takeaways & Limitations

    Users can improve handoffs and execution controls, developers can strengthen tool integration, and researchers can investigate diagnosis and reliable memory management.

    Source: §7 · p. 29

    The dataset may not fully represent all practices in developing and using LLM-based MAS.

    Source: §5 · p. 26

Abstract

from arXiv · show

With the advancement of LLM-based multi-agent systems (MAS), an increasing number of opensource projects are adopting multi-agent architectures as the foundation of their core functionality. Although research and practice on MAS have attracted considerable attention, limited studies have explored the challenges faced by practitioners of open-source LLM-based MAS, the causes of these challenges, and potential solutions. To address this gap,we conducted an empirical study to understand the issues that practitioners encounter when developing and using open-source LLM-based MAS, the possible causes of these issues, and potential solutions. We collected 22,848 closed issues from 21 open-source LLM-basedMASand applied a mixed automated and manual filtering approach to reduce the dataset to 944 issues related to LLM-based MAS.We then analyzed these issues to understand the frequent issues encountered by practitioners, their underlying causes, and potential solutions. Our study results show that (1) Orchestration & Execution Issue is the most common issue faced by practitioners, (2) Workflow Problem, Tool Integration Problem, and Memory Problem are identified as the most frequent causes of the issues, and (3) Optimize Workflow is the predominant solution to the issues. Based on the study results, we derive empirically grounded implications for practitioners and researchers aimed at improving orchestration, tool integration, and memory mechanisms in LLM-based MAS.

1. Introduction

This study addresses the lack of empirical evidence on practitioners’ challenges in open-source LLM-based MAS by analyzing GitHub issues. It identifies common issues, causes, and solutions, and organizes these findings into taxonomies and mappings.

1. Introduction

  • Empirical evidence on real-world development and maintenance challenges in LLM-based MAS remains limited despite growing adoption and the complexity of agent coordination, planning, and interaction.The study frames these challenges as not yet fully understood.Source: §1 · p. 1
  • The study collected 22,848 closed issues from relevant GitHub repositories and analyzed issue-tracking data to observe practitioner-reported issues, causes, and solutions.The analysis focused on projects whose documentation explicitly describes them as MAS projects.Source: §1 · pp. 1–2
  • A motivating issue reports a manager agent passing structured dictionaries where the schema expects strings, prompting a proposed JSON message format.The example illustrates a type mismatch in multi-agent coordination.Source: §1 · p. 2
  • Orchestration & Execution Issue is the most common practitioner issue; Workflow Problem, Tool Integration Problem, and Memory Problem are the most frequent causes.Optimize Workflow and Adjust Configuration are the predominant solutions.Source: §1 · p. 2
  • The authors organize issues into a two-level taxonomy, create taxonomies of causes and solutions, and map issues to their causes and associated solutions.These mappings are intended to clarify how issues are triggered and addressed in practice.Source: §1 · p. 2

2. Methodology

The study uses a staged GitHub-issue methodology to investigate practitioner-reported MAS issues, their causes, and potential solutions. It combines automated and manual filtering with manual extraction and iterative qualitative analysis.

2. Methodology

  • The methodology proceeds through research questions, data collection, filtering and labeling, extraction, and analysis.Figure 2 presents an overview of this research process.Source: §2 · p. 2, Figure 2 · p. 3

2.1. Research Questions

  • The study asks what issues practitioners face in MAS, what causes them, and what solutions address them.These three research questions guide the methodology.Source: §2.1 · pp. 2–3

2.2. Data Collection

  • The researchers selected 21 actively maintained open-source LLM-based MAS projects and collected 22,848 closed issues from their repositories.Project selection included checks for actual multi-agent functionality and criteria concerning project quality, activity, and popularity.Source: §2.2 · pp. 3–4

2.3.1. Pilot Data Filtering

  • In a 500-issue pilot, automated and manual filtering agreed on 452 cases, yielding a 90.4% agreement rate.The pilot also found that keyword-based automation could miss context-dependent issues and flag terminology that did not describe a practitioner-reported MAS problem.Source: §2.3.1 · p. 5

2.3. Data Filtering and Labeling

  • Mixed automated and manual filtering reduced 22,848 closed issues to 944 practitioner-reported MAS issues.Automated filtering identified 2,335 candidates, which were manually reviewed before the final set was selected.Source: §2.3 · p. 4, §2.3.2 · p. 5

2.4. Data Extraction

  • Manual extraction mapped issue, cause, and solution data items to RQ1–RQ3 using issue titles, descriptions, and discussions.The pilot extraction established rules to record the first issue described while retaining all identified causes and solutions for it.Source: §2.4 · p. 6, §2.4.1 · p. 6, Table 5 · p. 6

2.4.2. Formal Data Extraction

  • For 16 reports containing multiple problems, the researchers extracted only the first reported problem and its corresponding causes and solutions.Some closed reports were too vague to identify causes or solutions, while inactivity, low priority, or complexity could also make extraction difficult.Source: §2.4.2 · p. 6

2.5. Data Analysis

  • The researchers used open coding and constant comparison to build and refine categories, with author review and discussion resolving disagreements.The final analysis resolved 29 issue-category, 24 cause-category, and 22 solution-category conflicts.Source: §2.5 · pp. 6–7

3. Results and Interpretation

The results present taxonomies of practitioner-reported issues, their causes, and solutions, then map issues to causes and solutions. The mappings include only closed issues for which the corresponding cause or solution was extracted.

3. Results and Interpretation

  • The results introduce two-stage taxonomies for issues, causes, and solutions, with categories and types describing practitioner-reported problems, underlying reasons, and resolution approaches.The section also presents issue–cause and issue–solution category mappings.Source: §3 · p. 7
  • Issue–cause and issue–solution mappings include only closed issues for which the corresponding cause or solution was extracted.Source: §3 · p. 7
  • Figure 4 presents the issue taxonomy, while Figure 3 shows the distribution of issue-closure durations.Source: §3 · p. 7

3.1. Category of Issues (RQ1)

Orchestration & Execution is the largest issue category, followed by Memory & State Management. Across these categories, practitioners report failures in runtime coordination and in preserving or propagating state across agents.

3.1.1. Orchestration & Execution Issue

  • Orchestration problems include failed task handoffs, incorrect sequencing, loops, termination failures, premature stops, resource conflicts, and failed agent selection.Agent design, context loss, and concurrency problems also disrupt information flow, context propagation, or shared-state handling across agents.Source: §3.1.1 · p. 7

3.1. Category of Issues (RQ1)

  • 22.0% of reported issues were Orchestration & Execution, ahead of Memory & State Management (14.7%) and Feature Request (13.2%).Figure 4 presents the taxonomy of issues extracted from the dataset.Source: §3.1 · p. 7, Figure 4 · p. 8

3.1.1. Orchestration & Execution Issue

  • Coordination & Handoff Issue (62) and Control Flow Errors (40) were the most frequent orchestration types, involving task-transfer failures and unstable execution ordering.The authors interpret the pattern as reflecting weak coordination between agents rather than limitations in individual agents.Source: §3.1.1 · p. 9

3.1.2. Memory & State Management Issue

  • Memory & State Management issues (14.7%) concern storing, updating, and propagating memory, state, or intermediate information across agents.Reported types include context loss, state inconsistency, persistence and retrieval failures, context overflow, and state loss.Source: §3.1.2 · p. 9
  • Context Loss (42) and State Inconsistency (36) were the most frequent memory issues, indicating unreliable state propagation across agents.The authors identify a gap between designed memory abstractions and their behavior in execution; other reported types include retrieval failure, context overflow, and state loss.Source: §3.1.2 · p. 9

3.1.3. Feature Request

Feature requests make up 13.2% of reported issues and focus on extending framework capabilities. Implementation and tool issues, respectively, involve flawed runtime behavior and problems integrating, configuring, or using external tools.

3.1.3. Feature Request

  • 13.2% of reported issues were Feature Requests, with Extensibility & Customization (57) and Communication & Data Flow (27) the most common types.Requests also covered monitoring and debugging, framework limitations, and performance and scalability.Source: §3.1.3 · pp. 9–10

3.1.4. Implementation Issue

  • Implementation Issues (10.6%) arise from incorrect or incomplete implementation; Execution Logic Error (29) and Correctness Issue (20) were the most frequent types.The authors describe a gap between intended design and implementation, including runtime flaws, dependency conflicts, and parameter misconfiguration.Source: §3.1.4 · pp. 10–11
  • Implementation problems span execution logic, correctness, dependency conflicts, parameter configuration, missing attributes or parameters, class handling, and agent execution strategy.For example, an implementation may enforce sequential processing when parallel execution is intended.Source: §3.1.4 · pp. 10–11

3.1.5. Tool Issue

  • Tool Issues (9.2%) involve incorrect integration, configuration, or use of external tools; Tool Execution Failure (41) and Tool Invocation Error (28) were most common.The authors characterize the main pattern as runtime breakdowns and incorrect tool selection in multi-agent settings.Source: §3.1.5 · p. 11
  • Tool problems also include incorrect tool selection and parameter-schema mismatches that can cause validation failures before execution.The tool may be selected correctly while its inputs fail to match the expected structure, type, or format.Source: §3.1.5 · p. 11

3.1.6. Integration Issue

Integration issues concern component interaction and compatibility, while role issues involve role assignment or labeling. Environment and configuration problems can prevent agents from initializing, communicating, or executing correctly.

3.1.6. Integration Issue

  • Integration Issues (6.7%) concern interaction and compatibility across system components; API Integration Failure (36) and Parameter Misconfiguration (8) were the most common types.The authors identify incompatibilities and configuration mismatches between external components and the MAS as the primary pattern.Source: §3.1.6 · pp. 11–12
  • Integration problems include API and framework integration failures, parameter misconfiguration, dependency conflicts, and interface incompatibility.A workflow may work in a direct environment but fail through an external interface such as a WebSocket or IO stream.Source: §3.1.6 · pp. 11–12

3.1.7. Role Issue

  • Role Issues (3.8%) involve unclear or incorrect role assignment, including execution-time assignment errors and role labels in logs that diverge from actual behavior.The two reported types are Role Assignment Error (30) and Role Misalignment (6).Source: §3.1.7 · p. 12

3.1.8. Environment & Configuration Issue

  • Environment & Configuration Issues concern setup, dependencies, and runtime configuration that can prevent agents from initializing or executing correctly.Dependency Conflict (9) and Parameter Configuration Error (9) were the most common types; the category also includes environment misconfiguration and API integration failure.Source: §3.1.8 · pp. 12–13
  • Environment-related failures include incompatible package versions, malformed parameter overrides, deployment constraints, and API lifecycle handling.These constraints can prevent agents from initializing, communicating, or executing correctly.Source: §3.1.8 · pp. 12–13

3.1.9. Output Issue

Output problems concern completeness, formatting, and reliability, while the adjacent categories cover documentation, communication, performance, and model constraints in LLM-based MAS.

3.1.9. Output Issue

  • Incomplete responses (12) and output inaccuracy (10) are the most common output issues, alongside inconsistent results that reduce usability and reliability.Incomplete responses may hide intermediate worker outputs; output inaccuracy includes unintended formatting artifacts or structural tokens.Source: §3.1.9 · p. 13

3.1.10. Documentation Issue

  • Documentation issues include unclear or outdated descriptions and missing guidance, with 14 unclear-documentation and 12 missing-documentation issues.Reported examples include obsolete tool prefixes and missing setup examples.Source: §3.1.10 · p. 14

3.1.11. Communication Issue

  • Communication issues center on malformed message formats (12) and connection failures (6), alongside role-assignment and message-transmission problems.These problems disrupt information exchange and interaction between agents.Source: §3.1.11 · p. 14

3.1.12. Performance Issue

  • Performance issues comprise inefficient execution and response delay, with 9 cases of each affecting system efficiency and responsiveness.Examples include unnecessary agents continuing after an output is generated and latency between callbacks and tool invocation.Source: §3.1.12 · p. 14

3.1.13. Model Issue

  • Model issues involve message-format mismatches (12) and API compatibility problems (3), including provider-specific role-sequencing requirements.An example reports an API rejecting sequential multi-agent messages because user and assistant roles must alternate.Source: §3.1.13 · p. 15

3.1.14. Compatibility Issue

Compatibility, error-handling, task-management, and security issues arise from mismatches or failures in external interfaces, execution recovery, task flow, and safeguards.

3.1.14. Compatibility Issue

  • Compatibility issues involve interface type mismatches (10) and API compatibility issues (3), including violations of external schemas and provider-specific role rules.Such violations can cause integration breakdowns despite correct system design.Source: §3.1.14 · p. 15

3.1.15. Error Handling Issue

  • Error-handling issues include unhandled runtime exceptions (5) and recovery failures (7), where prior state may persist after an error.Examples include an unassigned variable causing an exception and a guardrail exception failing to remove a prior response from session history.Source: §3.1.15 · p. 15

3.1.16. Task Management Issue

  • Task-management issues involve allocation failures (7) and duplicated tasks (4), disrupting task distribution or repeating agent execution.A reported example describes agents being re-invoked after completing their first cycle.Source: §3.1.16 · p. 16

3.1.17. Security Issue

  • Security issues include validation problems (2) and access-control problems (1), associated with unsafe input processing and unauthorized access to user-specific resources.Examples include unsafe deserialization and agents returning data belonging to other users.Source: §3.1.17 · p. 16

3.2. Category of Causes (RQ2)

The study categorizes causes reported for open-source LLM-based MAS issues, maps them to issue categories, and finds workflow, tool integration, and memory causes especially prominent.

3.2. Category of Causes (RQ2)

  • Researchers identified 540 causes in 57.2% of 944 issues across 14 categories; Workflow Problem (29.4%), Tool Integration Problem (11.7%), and Memory Problem (11.1%) were most frequent.An issue could have multiple causes, and some issues had no identifiable cause.Source: §3.2 · p. 16, §3.2.1 · p. 16

3.2.1. Results

  • Workflow, tool-integration, and memory causes concern task coordination and execution, failures in tool integration or invocation, and storing or retrieving context across interactions.Workflow Problem includes context-management, coordination, handoff, orchestration, loop, and state errors; Tool Integration Problem includes MCP, parameter, execution, and schema problems; Memory Problem includes persistence, management, and state problems.Source: §3.2.1 · pp. 16–17
  • Other cause categories cover message exchange, setup and dependencies, documentation, component integration, and agent creation or runtime failures.The taxonomy describes communication, configuration, dependency, documentation, integration, and agent-lifecycle problems.Source: §3.2.1 · p. 18
  • Performance and database causes include latency, timeouts, scalability, and execution problems, as well as database setup, retrieval, or schema failures.The performance taxonomy also includes parallel-execution problems.Source: §3.2.1 · pp. 18–19
  • Infrastructure, frontend, and security causes concern deployment or file handling, output display, and vulnerabilities involving prompt injection or insecure deserialization.Examples include network policies blocking access to external providers and truncated streamed outputs.Source: §3.2.1 · p. 19

3.2.2. Causes to Issues Mapping

  • Orchestration & Execution Issue is the largest issue category; 54.8% have identified causes, most often Workflow Problem (65), Agent Lifecycle Problem (10), or Integration Problem (9).Across issue categories, identified-cause rates range from 13.6% for Feature Request to 87.0% for Implementation Issue.Source: §3.2.2 · p. 19

3.2.3. Interpretation

  • Overall, causes cluster around workflow execution, tool use, and memory handling, while other categories also affect system stability, usability, scalability, and security.The interpretation links workflow failures to state handling and outdated context, tool-integration problems to tool access, and memory problems to context retention and retrieval.Source: §3.2.3 · pp. 20–21

3.2.2. Causes to Issues Mapping

  • Workflow Problem accounts for 57.0% of Orchestration & Execution Issue causes, while Memory Problem explains 51.2% of Memory & State Management cases and Tool Integration Problem 35.9% of Tool Issues.Integration Problem accounts for 37.7% of Integration Issues.Source: §3.2.3 · p. 21

3.3. Category of Solutions (RQ3)

Practitioners use a range of fixes for open-source LLM-based MAS issues, with workflow optimization the most common. The mapping shows that solutions vary by issue category.

3.3.1. Results

  • Optimize Workflow accounts for 26.0% of solutions, followed by Adjust Configuration (15.0%), Optimize Tool Integration (11.9%), and Optimize Infrastructure (11.0%).The study identified 427 solutions across 10 categories; they address 45.2% of the 944 issues.Source: §3.3.1 · p. 21, §3.3 · p. 21, Table 8 · p. 22
  • Workflow fixes cover orchestration, state management, loops, parallelism, and sequential execution, while configuration fixes adjust prompts, runtime parameters, dependencies, and environments.Examples include changing runtime controls to prevent unintended code execution and reducing transfer latency between agents.Source: §3.3.1 · p. 21
  • Tool integration fixes target tool execution, registration, connectivity, and compatibility; infrastructure fixes address database compatibility and memory initialization, synchronization, and persistence.The study also describes error-handling fixes for error propagation, fallback, handoffs, and exceptions, and communication fixes for message handling and inter-agent exchange.Source: §3.3.1 · pp. 21–22
  • Other solution categories include dependency updates, agent initialization, documentation, frontend improvements, and error handling.Agent initialization covers imports, agent creation and registration, and connections to required components; documentation fixes add or clarify guidance, and frontend fixes improve interaction and output display.Source: §3.3.1 · pp. 22–23

3.3.3. Interpretation

  • Common solution categories address workflow, configuration, tools, and infrastructure; the interpretation links their frequency to execution flow, system settings, external tools, and reliability.Optimize Communication is also described as important because inter-agent message exchange is essential for task completion.Source: §3.3.3 · p. 24

3.3.2. Solutions to Issues Mapping

  • For Orchestration & Execution Issue, the largest issue category, Optimize Workflow addresses 48.1% of cases; Documentation Issue has the highest resolution rate and Output Issue the lowest.The prose reports 76.0% for Documentation Issue and 24.1% for Output Issue.Source: §3.3.2 · p. 23, §3.3.3 · p. 24

3.3.3. Interpretation

  • The solution-to-issue mapping shows different leading fixes across categories, including Optimize Tool Integration for Tool Issues and Optimize Infrastructure for Memory & State Management Issues.Optimize Workflow is the predominant solution for Orchestration & Execution Issue; the mapping also identifies category-specific fixes for other issue types.Source: §3.3.2 · p. 23, §3.3.3 · p. 24

4. Implications

The implications translate the study's findings into suggestions for users, developers, and researchers of open-source LLM-based MAS. They emphasize workflow coordination, tool integration, memory, and reproducibility.

4.1. Implications for Users of LLM-based MAS

  • Users can specify handoff prompts and execution controls to address Orchestration & Execution Issues, which rank first among 17 issue categories at 22.0%.Suggested controls include retry limits, timeouts, and logging to reduce infinite loops, premature termination, and unexpected execution behavior.Source: §4.1 · p. 24
  • Users of MAS development frameworks should monitor workflows, manage agent memory and state, verify tool use, and configure explicit communication mechanisms.Orchestration & Execution, Memory & State Management, Tool, and Integration Issues account for 22.0%, 14.7%, 9.2%, and 6.7% of issues, respectively.Source: §4.1 · p. 25

4.2. Implications for Developers of LLM-based MAS

  • Developers need foundational knowledge of workflow design, tool integration, and memory management; their corresponding causes account for 29.4%, 11.7%, and 11.1% of identified causes.The section highlights agent coordination and handoffs, tool registration and lifecycle, and session persistence and state consistency.Source: §4.2 · p. 25
  • Developers can strengthen tool integration with schema validation, tool-level timeouts, availability checks, connection management, and retries for failed calls.The section identifies deadlocks, resource contention, and schema mismatches as typical failure conditions.Source: §4.2 · p. 25

4.3. Implications for Researchers of LLM-based MAS

  • Researchers are urged to investigate orchestration frameworks, workflow verification, and runtime monitoring because Workflow Problem accounts for 57.0% of causes associated with Orchestration & Execution Issues.The proposed directions include detecting coordination anomalies such as infinite loops and premature termination.Source: §4.3 · pp. 25–26
  • Researchers could explore failure-reproduction techniques and reliable memory mechanisms, including context propagation, state synchronization, and context compression.The study reports that 27 of 128 unsolved issues involve difficulty reproducing problems; Memory & State Management Issues account for 51.1% of issues caused by Memory Problem.Source: §4.3 · p. 26

5. Threats to Validity

The study identifies possible validity threats in manual analysis and source selection, while describing steps taken to support reliability.

Construct validity

  • Manual filtering, coding, comparison, and extraction may introduce subjective interpretation, although pilot studies and author discussions were used to reduce this risk.The first author primarily conducted open coding and constant comparison, so the risk could not be eliminated entirely.Source: §5 · p. 26

External validity

  • The dataset's reliance on closed GitHub issues may not represent all LLM-based MAS development and use practices.Relevant information may instead reside outside GitHub or in undocumented settings.Source: §5 · p. 26

Reliability

  • A pilot labeling phase produced Cohen’s Kappa of 0.82, and the authors report regular discussions to resolve interpretation differences.The dataset was made publicly available for independent replication and validation.Source: §5 · p. 27

6. Related Work

The related work reviews repository mining, LLM-based MAS for software engineering, and studies of MAS challenges. It positions this study as addressing limited evidence about practitioner issues, causes, and solutions.

6.1. Mining Software Repositories for Exploring Issues, Causes, and Solutions

  • Prior repository-mining studies examine practitioner-reported issues, causes, and solutions in microservices, performance problems, and LLM-based projects.The reviewed work includes taxonomies of issue categories and root causes, alongside findings about common problems and remedies.Source: §6.1 · p. 27

6.2. LLM-based Multi-Agent Systems for Software Engineering

  • Prior LLM-based MAS work describes software-development frameworks such as ChatDev, MetaGPT, AutoGen, and AgileCoder, and studies design patterns and quality attributes.These studies report framework capabilities and benchmark results; the review notes a need to understand deployment challenges more deeply.Source: §6.2 · pp. 27–28

6.3. Exploration of Issues, Causes, and Solutions in LLM-based MAS

  • Studies of MAS challenges use repository data, online discussions, and execution traces to identify issues and failure patterns, but comprehensive evidence on practitioner issues, causes, and solutions remains limited.The reviewed studies include analyses of framework maintenance, developer discussions, and execution-trace failure modes.Source: §6.3 · p. 28

6.4. Conclusive Summary

  • The review concludes that prior MAS studies provide limited understanding of real-world practitioner problems, their causes, and adopted solutions.This study analyzes closed GitHub issues to derive taxonomies of issues, underlying causes, and solutions; Table 10 compares empirical studies on LLM-based MAS.Source: §6.4 · p. 28, Table 10 · p. 28

7. Conclusions

The study analyzed 944 issues from open-source LLM-based MAS, identifying the most common issue, frequent causes, and predominant solution. It offers practical implications for users, developers, and researchers.

7. Conclusions

  • Orchestration & Execution Issue was the most common issue; Workflow Problem, Tool Integration Problem, and Memory Problem were the most frequent causes, while Optimize Workflow was the predominant solution.The analysis followed filtering 22,848 closed issues down to 944 issues related to LLM-based MAS.Source: §7 · p. 28
  • Users can address orchestration issues with explicit agent handoffs and execution controls, and refine prompts for more reliable responses.Suggested controls include retry limits, timeouts, and monitoring mechanisms.Source: §7 · p. 29
  • Developers can improve tool integration reliability through schema validation, execution controls, and tool connection management.Source: §7 · p. 29
  • Researchers can investigate reproducing and diagnosing unsolved issues and develop memory mechanisms for context propagation, state synchronization, and retrieval.Source: §7 · p. 29

CRediT authorship contribution statement

The statement assigns conceptualization, methodology, investigation, data curation, analysis, supervision, and writing tasks among six authors.

CRediT authorship contribution statement

Related papers

Loading 2610.00905v1…