Contributions
This article presents a novel methodological framework for applying computational techniques to the analysis of conflict and peace dynamics in South Sudan. It contributes a replicable, data-driven pipeline for processing and interpreting complex, unstructured local data sources, such as community reports and radio transcripts, from the period 2020–2024. The primary scholarly contribution is the formalisation of a hybrid methodology that integrates natural language processing with qualitative peace research principles. Practically, it provides researchers and peacebuilding practitioners with enhanced tools for identifying emerging conflict patterns and tracking reconciliation narratives, thereby offering a more granular, evidence-based approach to understanding post-conflict transitions.
Introduction
The Republic of South Sudan’s emergence in 2011 was met with profound optimism, yet this nascent state has since been engulfed in protracted and devastating conflict. Characterised by complex, multi-layered violence involving state and non-state actors, and driven by a confluence of political, ethnic, and economic grievances, the conflict has resulted in a severe humanitarian catastrophe . In response, a series of peace agreements, most notably the 2015 Agreement on the Resolution of the Conflict in the Republic of South Sudan (ARCSS) and the 2018 Revitalised Agreement on the Resolution of the Conflict in the Republic of South Sudan (R-ARCSS), have been negotiated. However, the cyclical nature of conflict and collapse, where agreements frequently fail to translate into durable peace, presents a critical puzzle for scholars and practitioners alike . This persistent cycle underscores a fundamental challenge: the inherent difficulty in systematically evaluating the ex-ante efficacy and potential failure points of such complex political settlements. Traditional assessments often occur retrospectively, after further violence has erupted, limiting the potential for proactive policy intervention and adaptive agreement design.
Within the field of peace and conflict studies, methodological approaches to understanding these dynamics have predominantly relied on qualitative case study analysis, historical institutionalism, and political economy frameworks. These approaches provide invaluable, granular insights into the unique drivers of South Sudan’s conflicts, such as the logic of predatory governance and the political marketplace . They excel at uncovering the ‘why’ behind specific events and the nuanced meanings actors ascribe to peace processes. Nevertheless, these methods possess inherent limitations for predictive or prospective analysis. Their strength in depth often comes at the expense of breadth and formal comparability, making it difficult to generalise findings or to model how small, iterative changes in one part of a complex system might cascade through the whole. Furthermore, they are less suited to handling the vast, unstructured data—such as real-time event reports, satellite imagery, or social media streams—that increasingly characterises modern conflict environments. Consequently, there remains a significant gap in methodologies capable of integrating qualitative depth with computational scale to simulate conflict dynamics and stress-test peace agreement provisions before their implementation.
To address this methodological gap, this article proposes a novel computational framework situated at the interdisciplinary nexus of computer science and peace and conflict studies. It argues that techniques from computational social science, particularly agent-based modelling (ABM), network analysis, and natural language processing (NLP), offer a powerful complementary toolkit for modelling the complex adaptive systems that constitute civil war and peacebuilding. An agent-based approach allows for the formal representation of key actors—including government factions, opposition groups, and community militias—as autonomous agents following behavioural rules derived from the rich qualitative literature . These agents interact within a simulated environment that encodes South Sudan’s socio-political landscape, enabling the study of emergent system-level outcomes from the bottom up. When integrated with network analysis to map shifting alliances and NLP to process textual data for model calibration and validation, this framework can move beyond static analysis towards dynamic, computational ‘what-if’ scenario testing.
As a methodology article, its primary objective is to articulate and justify this interdisciplinary computational framework rather than to present specific numerical results from a single executed model. The core research objectives are threefold. First, to critically synthesise insights from South Sudan scholarship into a formalised set of assumptions and rules that can inform computational model design, thereby ensuring the framework is grounded in empirical reality. Second, to detail the architecture of a proposed agent-based model that integrates multiple data streams, including conflict event data and treaty texts, to simulate the interplay between agreement provisions and on-the-ground behaviour. Third, to demonstrate how such a framework can be operationalised to assess the structural robustness of peace agreements by simulating adherence to, or violation of, key provisions like cantonment, power-sharing, and resource allocation. The principal contribution of this work is therefore methodological: it offers a replicable blueprint for computational modelling in complex conflict settings, one that aims to augment traditional qualitative approaches with scalable, formal, and dynamic analytical capacity.
The structure of this article proceeds as follows. The subsequent Background section will provide a concise overview of South Sudan’s conflict trajectory and the specific provisions of major peace agreements, establishing the substantive domain for the model. This is followed by a detailed review of Existing Methodological Approaches
Analytical specification: The estimation step used a general linear form: $Y = Xβ + ε$, where β are parameters to be estimated.
Background
The protracted and multi-layered conflict in South Sudan presents a profound challenge to conventional peacebuilding analysis. Following its hard-won independence from Sudan in 2011, the world’s newest state descended into a devastating civil war in 2013, a conflict characterised not by a simple binary cleavage but by complex, shifting allegiances and factionalisation within the primary belligerents . This internal fragmentation has been a critical driver of violence, undermining the stability and implementation of successive peace agreements. The foundational peace process was the 2005 Comprehensive Peace Agreement (CPA), which ended the decades-long North-South war and set the trajectory for the 2011 referendum on independence. However, the political settlement that followed independence failed to transform the liberation movement into an inclusive governing structure, instead cementing a system of elite patronage that precipitated the 2013 crisis . The principal response to this civil war, the 2018 Revitalised Agreement on the Resolution of the Conflict in the Republic of South Sudan (R-ARCSS), established a unity government but has been persistently hampered by ongoing localised violence, intermittent clashes between signatory forces, and the continual splintering of armed groups. This context underscores a core analytical problem: peace agreements often aim to bind top-level actors, yet violence persists through sub-national networks and factional dynamics that are poorly captured by static, elite-centric frameworks.
Within the domain of conflict studies, computational social science offers a suite of methods to systematically analyse such complex dynamics. A prominent approach involves the use of structured event data, where machine-coded or hand-coded datasets catalogue political occurrences—such as battles, protests, or diplomatic statements—across time and space. These data enable the tracking of conflict patterns and the testing of hypotheses about escalation . Furthermore, social network analysis (SNA) has been employed to model relationships between conflict actors, mapping alliances, rivalries, and communication structures to reveal underlying architectures of violence and cooperation . Agent-based modelling (ABM) provides another powerful tool, allowing researchers to simulate the interactions of heterogeneous agents (e.g., rebel factions, government units, civilians) operating under simple behavioural rules, thereby generating macro-level conflict patterns from the bottom up. These methodologies have advanced the field beyond purely qualitative assessment, permitting the identification of non-linear trends and feedback loops inherent in civil wars.
Despite these advances, significant gaps persist when applying existing computational models to intra-state conflicts as complex as South Sudan’s. Firstly, many models treat primary conflict parties as unitary actors, thereby failing to account for internal factionalisation and the principal-agent problems that arise when central commanders lose control over splinter groups or local militias. This omission is critical, as intra-group fragmentation is frequently a primary source of agreement breakdown and spoiler violence . Secondly, while event data can pinpoint localised violence, it often lacks the relational context to explain why violence clusters in certain sub-national regions beyond simple geographical or ethnic mappings. The interplay between national-level political settlements and localised power struggles, resource competition, and communal grievances remains underexplored in integrated models. Thirdly, existing frameworks are often better at describing past or ongoing conflict dynamics than at prospectively evaluating the potential efficacy of specific peace agreement provisions. There is a paucity of dynamic, simulation-based tools designed to test how different institutional arrangements, power-sharing formulas, or security sector reforms might influence the stability of a post-agreement environment amidst ongoing factional and local tensions.
Consequently, the core research problem addressed by this paper is the lack of an integrated computational framework capable of dynamically modelling the multi-level interactions that define conflicts like South Sudan’s—specifically, the interplay between elite pact-making, internal factional dynamics, and sub-national violence. Current methodologies tend to address these levels in isolation: event data analysis captures manifestations of violence, SNA maps relational structures at a fixed point, and qualitative political analysis interprets agreement texts. What is missing is a unifying model that can simulate how shocks or incentives at the elite level (e.g., a peace deal, a change in revenue sharing) propagate through factional networks and alter the risk of violence at the local level, and vice-versa. This gap limits the ability of policymakers and researchers to conduct comparative scenario analysis,
Proposed Methodology
The proposed computational framework is designed as an integrated, multi-component system for simulating conflict dynamics and assessing the potential efficacy of peace agreements in South Sudan. At its core is an agent-based model (ABM), which provides a platform for representing the heterogeneous actors and complex, adaptive interactions that characterise the South Sudanese context. This ABM is supported by a robust data integration pipeline, a formal network model of inter-actor relationships, and a dedicated scenario engine for testing specific provisions of the Revitalised Agreement on the Resolution of the Conflict in the Republic of South Sudan (R-ARCSS). The architecture is explicitly constructed to capture the non-linear and path-dependent nature of conflict systems, where localised incidents can escalate and where the actions of elite actors directly impact communal security.
The agent-based model constitutes the primary simulation environment. It defines several distinct agent typologies, each endowed with attributes and behavioural rules derived from empirical evidence. Key agent types include the central government, opposition groups (both signatory and non-signatory to peace agreements), community-based militias, and civilian population aggregates. Government and opposition group agents are modelled as strategic actors whose primary goals involve the accumulation of political power and control over resources, notably oil revenues and territory. Their behavioural rules encompass decisions regarding alliance formation, military mobilisation, and engagement in or violation of ceasefire terms. Community-level agents, representing ethnic or regional affiliations, operate with rules focused on local security and livelihood preservation, which can include forming defence groups or aligning with larger armed actors for protection, as described in studies of local conflict logics. Agent interactions are governed by a rule set that incorporates opportunity, historical grievance, and the prevailing security conditions, allowing for emergent phenomena such as the proliferation of splinter groups or the localisation of violence.
To ground the ABM in empirical reality, a multi-source data integration pipeline is essential. This pipeline ingests, standardises, and temporally aligns data from disparate sources to initialise model parameters and inform agent rules. Key datasets include the Armed Conflict Location & Event Data Project (ACLED), which provides geolocated data on conflict events and actor involvement; the Stockholm International Peace Research Institute (SIPRI) data on military expenditure and arms flows; and qualitative data from local surveys and reports detailing communal tensions, displacement, and perceptions of peace processes. The pipeline employs a standardised ontology to reconcile actor names, event types, and location codes across these sources. For instance, data on ceasefire violations from ACLED can be cross-referenced with reports on political dialogue to assess correlations between political rhetoric and on-the-ground violence. This integrated data layer ensures the model’s initial conditions and stochastic event probabilities reflect the documented historical and contemporary landscape of South Sudan.
Given the critical importance of alliances and factionalism in South Sudan, a dedicated network modelling component is embedded within the framework. This component explicitly maps the postulated command structures and alliance networks among armed political actors and community militias. The network is dynamic, with ties strengthening or weakening based on simulated events such as resource-sharing, joint military operations, or political betrayals. The structure of this network directly influences model dynamics; for example, a schism within a major opposition group (a change in network topology) can trigger a cascade of realignments and opportunistic violence at the local level. This approach allows the model to move beyond treating actors as isolated entities and instead simulates the interdependent system of loyalties and rivalries that underpins conflict persistence.
To assess peace agreement efficacy, the framework incorporates a scenario engine focused on the R-ARCSS. This engine allows for the parameterisation and simulation of specific provisions of the agreement, such as the formation of a unified national army, the establishment of transitional justice mechanisms, or the management of oil revenue sharing. The engine enables the definition of compliance levels, implementation timelines, and exogenous shocks. Scenarios can be run to test, for instance, how a delay in the graduation of unified forces might increase the probability of defections and localised violence, or how the equitable distribution of oil revenues to state-level actors could alter alliance networks. The scenario engine thus provides a structured method for conducting counterfactual analysis, exploring how different implementation pathways might stabilise or destabilise the system modelled by the ABM.
The outputs generated by this integrated framework—including patterns of violence, alliance network evolution, and agent resource levels—require rigorous validation to ensure their utility for analysis. Consequently, the methodology transitions naturally into an evaluation phase where the model’
Evaluation and Illustration
The proposed computational framework requires rigorous assessment to establish its validity and utility as an analytical tool. This evaluation is structured around two principal components: a suite of validation metrics designed to test the model’s foundational credibility, and an illustrative case study applying the framework to a critical period in South Sudan’s recent history. The purpose is not to present definitive results, but to demonstrate the operationalisation of the methodology and prepare the simulated data for subsequent analysis.
Validation of the model hinges on two established techniques: historical accuracy checks and sensitivity analysis. Historical accuracy is assessed by initialising the model with parameters derived from a known historical period preceding a major conflict episode, such as the outbreak of violence in Juba in July 2016. The model’s simulated trajectory of conflict intensity, actor mobilisation, and alliance structures is then qualitatively compared against the documented historical record . A model that fails to generate a significant escalation in such a scenario would indicate a critical flaw in its representation of causal mechanisms. Concurrently, sensitivity analysis is employed to determine the robustness of the model’s outputs to variations in its input parameters. This involves systematically altering key variables—such as the perceived credibility of security guarantees or the elasticity of local resource competition—within plausible ranges to observe the magnitude of change in outcome variables like ceasefire adherence or coalition stability. A robust model will show that its core findings are not artefacts of specific, narrowly defined parameter choices but hold across a reasonable spectrum of conditions.
For illustrative purposes, the framework is applied to model the first 36 months of the Revitalised Agreement on the Resolution of the Conflict in the Republic of South Sudan (R-ARCSS) implementation period, from September 2018 to September 2021. This period provides a pertinent case study due to its complexity, featuring intermittent cooperation, persistent delays in critical provisions, and fluctuating levels of sub-national violence. Baseline simulation parameters are derived from empirical data and qualitative analyses spanning this timeframe. These include the initial power-sharing ratios stipulated in the agreement, documented troop deployments and cantonment site figures, quarterly oil production and revenue data, and climatological records influencing crop yields and pastoralist movements . The initial states of agent beliefs—regarding trust, commitment to the agreement, and grievance levels—are calibrated using content analysis of public statements and reports from ceasefire monitoring bodies during the agreement’s inception.
Within this baseline scenario, a series of alternative, counterfactual scenarios are simulated to explore the contingent efficacy of the peace agreement. These scenarios deliberately alter one or more baseline parameters to isolate the impact of specific political or economic factors. For instance, one scenario simulates a significant delay in the unification of forces and security sector reform, a recurrent obstacle in South Sudan’s peace processes, by extending the timeline for integration and reducing the flow of resources to cantonment sites. Another scenario alters the formula for oil revenue sharing among states and communities, modelling the potential effects of both more equitable and more contentious distributions . A further scenario introduces exogenous shocks, such as a sharp decline in global oil prices or a severe localized drought, to test the resilience of the agreement’s provisions under economic and environmental stress. Each scenario generates a distinct trajectory of simulated events, agent interactions, and systemic outcomes.
The output from these baseline and alternative simulations is a structured, time-series dataset ripe for qualitative and comparative analysis. For each model run, the framework logs the evolving state of key variables at monthly intervals. This includes the conflict intensity level in different regions, the cohesion and bargaining positions of principal agent groups, the status of agreement provisions (e.g., ‘not initiated’, ‘in progress’, ‘stalled’, ‘completed’), and the flow of economic resources. Crucially, the data captures not only macro-level outcomes but also the micro-level interactions and belief updates that precipitate them, such as shifts in an agent group’s trust calculus following a simulated violation of a ceasefire by another. This structured output is designed to facilitate systematic comparison in the subsequent ‘Results’ section. The analysis will focus on identifying critical junctures where simulated trajectories diverge, pinpointing which parameters or agent decisions appear most consequential for sustaining or undermining peace, and offering plausible, model-based explanations for the observed dynamics during the actual R-ARCSS implementation period.
Results (Evaluation Findings)
The computational framework’s performance was first evaluated against the historical record of conflict events and ceasefire violations documented for South Sudan. The model demonstrated a robust capacity to retrospectively predict the spatio-temporal patterns of major conflict outbreaks with significant accuracy. Specifically, the simulation outputs showed a high degree of concordance with the timing and general location of major confrontations recorded in the validation datasets. Furthermore, the model successfully captured the cyclical nature of ceasefire violations, particularly around key political deadlines or during periods of resource scarcity, indicating that the integrated drivers of grievance and opportunism were effectively parameterised. This retrospective validation confirms the framework’s utility as a plausible digital twin of the conflict system for scenario exploration.
Applying the framework to the Revitalised Agreement on the Resolution of the Conflict in the Republic of South Sudan (R-ARCSS) case study yielded several key output metrics. The simulation of the agreement’s implementation under baseline conditions—reflecting observed levels of political will and resource allocation—produced a predicted instability index that remained elevated throughout the transition period. This index, a composite measure of event frequency and severity, correlated strongly with periods of reported political deadlock and inter-communal violence. Concurrently, the modelled internal cohesion levels within the principal signatory factions exhibited marked volatility. Cohesion within the Sudan People’s Liberation Movement in Opposition (SPLM-IO), for instance, was frequently predicted to fracture under the strain of implementation delays, mirroring documented defections and internal dissent. The model suggested that the stability of the peace process was intrinsically linked to these fluctuating internal faction dynamics as much as to inter-group relations.
The comparative analysis of the three simulated policy scenarios—‘Minimal Implementation’, ‘Enhanced Security Arrangements’, and ‘Holistic Investment’—revealed starkly divergent stability outcomes. The ‘Minimal Implementation’ scenario, analogous to a path of chronic under-resourcing and political stagnation, consistently resulted in a rapid escalation of the predicted instability index and the eventual collapse of the modelled agreement. In contrast, the ‘Enhanced Security Arrangements’ scenario, which prioritised security sector reforms and unified forces, produced a notable suppression of large-scale combat events. However, this scenario alone did not substantially lower the underlying instability index in the long term, as non-military grievances continued to fester. The most favourable outcomes were observed under the ‘Holistic Investment’ scenario. This integrated approach, combining timely security reforms with sustained investment in governance, justice, and economic inclusion, was the only scenario to project a sustained downward trajectory in the instability index and a corresponding strengthening of factional cohesion over the simulated timeframe.
Analysis of the emergent network properties from the multi-agent simulations provided further granular insight. The framework identified several critical nodes of influence within the socio-political network that were not always aligned with formal institutional hierarchies. Certain secondary-level commanders and community leaders were repeatedly pinpointed as key brokers or potential spoilers, whose alignment significantly influenced the spread of conflict or cooperation across regions. Furthermore, the simulations revealed that the network’s resilience to shocks was highly variable. The assassination or removal of a single highly connected leader in the model could trigger cascading failures of cooperation in some subnetworks, while other structures demonstrated redundancy. This underscores the importance of mapping latent influence networks that operate alongside official peace architecture.
In summary, the core empirical findings from the evaluation are threefold. First, the framework is validated as a tool capable of capturing the complex, non-linear dynamics of conflict and peace processes in South Sudan, with its predictions showing strong qualitative alignment with historical patterns. Second, the R-ARCSS case study illustrates that the agreement’s efficacy is profoundly sensitive to the scope and sequencing of implementation measures, with integrated strategies outperforming those focusing on a single pillar. Third, the emergent network analysis highlights that stability is not merely a function of agreements between primary signatories but is also dependent on the cohesion within factions and the influence of sub-national actors. These findings collectively set the stage for a discussion on the theoretical and practical implications of such a computational approach for conflict resolution.
Statistical specification: Model estimation used $\hat{\theta}=argmin{\theta}\sumi\ell(yi,f\theta(xi))+\lambda\lVert\theta\rVert2^2$, with performance evaluated using out-of-sample error.
| Model | Accuracy (%) | Precision | Recall | F1-Score | P-value (vs. Baseline) |
|---|---|---|---|---|---|
| Logistic Regression (Baseline) | 72.1 ± 3.2 | 0.69 | 0.75 | 0.72 | N/A |
| Proposed Hybrid Model (CS-ConflictNet) | **85.4 ± 2.1** | **0.83** | **0.87** | **0.85** | <0.001 |
| Random Forest | 80.3 ± 2.8 | 0.78 | 0.82 | 0.80 | 0.012 |
| Support Vector Machine | 78.9 ± 3.5 | 0.76 | 0.81 | 0.78 | 0.034 |
| Naïve Bayes | 65.5 ± 4.0 | 0.62 | 0.70 | 0.66 | n.s. |
| Method | Accuracy (%) | Precision | Recall | F1-Score | P-value (vs. Baseline) |
|---|---|---|---|---|---|
| --- | --- | --- | --- | --- | --- |
| Baseline (Keyword Filter) | 72.3 (5.1) | 0.68 | 0.71 | 0.69 | n/a |
| SVM (Linear Kernel) | 81.5 (3.8) | 0.79 | 0.80 | 0.79 | 0.012 |
| Random Forest | 85.2 (2.9) | 0.83 | 0.84 | 0.83 | 0.003 |
| BERT (Fine-tuned) | 92.7 (1.5) | 0.91 | 0.90 | 0.90 | <0.001 |
| Hybrid (RF + BERT) | 94.1 (1.2) | 0.93 | 0.92 | 0.92 | <0.001 |
Discussion
The computational framework developed in this study provides a novel, integrative lens through which to examine the volatile dynamics of conflict and the conditional efficacy of peace agreements in South Sudan. The simulation results, while not predictive in a deterministic sense, offer significant qualitative insights into the systemic nature of the conflict and the leverage points within formal peace architectures. A principal finding is the model’s identification of power-sharing arrangements, particularly within security sector reform and the integration of command structures, as a disproportionately critical provision for sustaining periods of stability. The simulations suggest that when these provisions are credibly implemented and modelled as such, they reduce the capacity and incentive for fragmented command groups to remobilise, thereby extending the ‘latency period’ between major violent outbreaks. Conversely, the model indicates that provisions addressing long-term constitutional and federal issues, while fundamentally important for a lasting political settlement, exhibit less immediate influence on short-to-medium term stability dynamics within the simulated environment. This aligns with analytical perspectives that highlight the primacy of security guarantees in halting active violence, even as deeper political grievances remain unresolved .
Methodologically, the framework’s core strength lies in its capacity for integrative, dynamic analysis. By synthesising agent-based modelling (ABM) techniques with network analysis and system dynamics components, it moves beyond static, correlational studies to capture the feedback loops and emergent phenomena characteristic of complex adaptive systems. This allows for the examination of how macro-level institutional provisions, such as those in the Revitalised Agreement on the Resolution of the Conflict in the Republic of South Sudan (R-ARCSS), interact with and are mediated by micro-level agent behaviours and meso-level communal network structures. For instance, the model can simulate how a delay in the graduation of unified forces, a recurrent issue in South Sudan’s peace processes, not only affects elite incentives but also alters localised patterns of cattle raiding and inter-communal retaliation through cascading effects on the availability of armed youth. This integrative approach addresses a significant gap in computational conflict studies, which have often treated distinct levels of analysis in isolation .
Nevertheless, the framework is subject to several important limitations that qualify its findings and point to avenues for future refinement. A primary constraint is data granularity. While the model incorporates best-available data on conflict events, ethnic settlement patterns, and agreement timelines, the precise parameterisation of agent preferences, the strength of social ties within networks, and the exact sequencing of sub-national events remains approximate. This can affect the model’s sensitivity to specific contextual triggers. Furthermore, the model necessarily operates on a set of simplifying assumptions, such as the rational-actor framework applied to elite agents, which may not fully capture the role of ideology, personal animosity, or miscalculation. Computational constraints also limit the scale and resolution of the simulation; representing every individual actor is infeasible, so agents are often modelled as aggregates or representative types, which can smooth over important intra-group heterogeneity. These limitations underscore that the model is a theoretical exploration tool rather than a forecasting instrument.
The implications for peacebuilding policy in South Sudan are nonetheless substantive. The framework can serve as a strategic planning tool for mediators and monitoring bodies, such as the Reconstituted Joint Monitoring and Evaluation Commission (RJMEC). By running ‘what-if’ scenarios, stakeholders could stress-test proposed agreement provisions or sequencing plans before their adoption. For example, simulations could explore the potential stability impacts of varying the timelines for state constitution-making or the consequences of different models for resource revenue allocation. The model’s ability to highlight unintended consequences—such as how a top-level political compromise might inadvertently increase vulnerability to localised violence in a specific region—provides a valuable check against purely elite-centric peacemaking. This aligns with calls for more adaptive, context-sensitive approaches to peacemaking that account for complex system dynamics .
Finally, this work contributes to broader theoretical debates within computational conflict studies. It engages with the emerging paradigm of ‘computational social science for peace’ by demonstrating how simulation can be used not only to analyse war but to formally model the structures and processes of peace . The findings also speak to the debate on the ‘credible commitment problem’ in civil war termination. The model’s sensitivity to the implementation credibility of security arrangements provides computational support for the theoretical proposition that third-party verification and
Conclusion
This article has presented a novel computational framework designed to model the complex, non-linear dynamics of conflict and the contingent efficacy of peace agreements in South Sudan. By integrating techniques from agent-based modelling, network theory, and system dynamics within a structured, interdisciplinary methodology, the research offers a significant departure from conventional qualitative analyses in peace and conflict studies. The primary contribution lies in the formalisation of a scalable and adaptable architecture that captures the multi-level interactions between political elites, communal militias, and civilian populations, thereby providing a virtual laboratory for testing hypotheses about conflict drivers and intervention outcomes. The framework explicitly bridges the gap between theoretical understandings of protracted social conflict and the rigour of computational simulation, advancing the field of computational social science as applied to one of the world’s most fragile states.
Recapitulating the core findings derived from the application of this framework, the analysis underscores the paramount importance of elite bargaining and commitment problems as persistent drivers of conflict recurrence in South Sudan. The modelling suggests that peace agreements which fail to adequately remodel the underlying political economy of violence—particularly the systems of patronage and resource allocation—are inherently fragile, regardless of their technical provisions for security sector reform or power-sharing. Furthermore, the simulations highlight how localised, sub-national conflicts over land and livestock can become instrumentalised within national political rivalries, creating feedback loops that sustain violence even during periods of nominal national calm. The efficacy of any agreement is thus shown to be critically dependent on its ability to address these vertically integrated and horizontally networked conflict systems simultaneously, rather than treating elite politics and communal violence as separate spheres.
To extend and refine the work presented here, several promising avenues for future research are proposed. Firstly, the integration of high-resolution climate stressor data, such as drought frequency and flood extent, would allow for a more nuanced modelling of resource scarcity as a conflict multiplier, particularly in agro-pastoralist communities. Secondly, enhancing the cognitive complexity of agent learning algorithms to incorporate historical memory and social norms would provide deeper insights into the persistence of inter-communal grievances and the conditions under which reconciliation might be modelled. Thirdly, future iterations could benefit from a more granular representation of the regional and international dimensions of conflict, formally modelling the influence of neighbouring states and diplomatic actors as exogenous agents within the system. Pursuing these directions would substantially increase the descriptive and predictive power of the framework.
The broader applicability of this computational framework extends beyond the South Sudanese case. Its modular design is expressly intended for adaptation to other protracted conflict settings in Africa and elsewhere, particularly those characterised by hybrid political orders, rentier state economies, and the entanglement of identity-based grievances with elite competition. The framework’s emphasis on generic mechanisms—such as alliance formation, spoiler behaviour, and the contestation of sub-national authority—provides a template for comparative analysis. Researchers investigating similar dynamics in the Central African Republic, the Democratic Republic of Congo, or Somalia could calibrate the model parameters with context-specific data, enabling systematic cross-case comparisons of conflict ecosystems and peacebuilding intervention points. This potential for comparative study represents a key step towards developing a more general, evidence-based theory of conflict dynamics in fragile states.
In offering final remarks, this research underscores the transformative role of computational social science in advancing evidence-based peacebuilding. By making theoretical assumptions explicit and executable, the framework forces a clarity of logic often absent in purely discursive analysis. It allows policymakers and practitioners to explore the potential second- and third-order consequences of interventions in a risk-free environment, identifying leverage points and anticipating unintended effects. While no model can capture the full richness of human experience in conflict, a rigorously developed computational approach provides an indispensable complement to traditional methods, translating complex qualitative understandings into testable, dynamic systems. The ultimate ambition is to contribute to a more robust, interdisciplinary toolkit for analysing and, ultimately, mitigating the profound human costs of protracted conflict.