Sage Journals: Discover world-class research

Abstract

Background

Improving awareness of Technical Debt (TD) has been demonstrated in prior work using the serious game TDSIM. Assessing how effectively TDSIM supports epistemic change is essential for its continued development. Examining participant play strategies offers a way to quantify a serious game’s impact on shifting player epistemic states, while also revealing opportunities for refinement. Yet the literature on serious game player behavior remains largely silent on strategy assessment, a gap our methodology directly addresses.

Method

The serious game TD-SIM was developed in prior work using real-world players. We analyze prior participant game data for actions taken. Combinations of actions define the concept of agent archetypes. Bayesian analysis is used to calculate probabilities of real-world play conditioned on agent archetypes. Principal Component Analysis (PCA) is used to classify conditional probabilities to which K-Means clustering is applied. Clusters of probability data are the strategies players of TD-SIM use, referred to as player archetypes.

Results

Our method identifies four player archetypes, characterizing playing strategies as Gaming, Managers, Learning, or Testing. Player archetypes inform and quantify behaviors of players, providing insight into improvement opportunities.

Conclusion

Measuring player archetypes is complex, but not impossible. We evaluate player archetypes of TD-SIM quantifying design improvements. Identifying and modelling participant playing strategies is a novel enhancement to general serious game evaluation. The benefit of our approach is that it allows us to characterize a player’s behavior in a quantifiable manner, which we can measure against intended purpose.

Keywords

technical debt Markov chains simulation/gaming serious games epistemic state

Remote Operations Centres (ROCs) operate industrial equipment several hundred kilometres from a location to deliver nationally and commercially important resources and services in remote and often harsh environments (Lakshmanan et al., 2019). These ROCs orchestrate human decision makers and multiple software systems from numerous vendors. All software systems have Technical Debt (TD), the long-term impact of shortcuts and compromises taken during the software development lifecycle (Besker et al., 2019; Kruchten et al., 2019). Managing technical debt within ROCs is complicated by interactions between software systems and decision makers, producing adaptive and emergent behaviours characteristic of Complex Adaptive Systems (CAS) (Bekebrede et al., 2015; Gould et al., 2024).

Our use case is a ROC owned by an Australian mining company with export revenue of $37 billion AUD (US$ 25bn) in 2025. Significant technical debt issues can affect company profitability and State and Federal budgets (Government of Western Australia, 2025; Ramasubbu & Kemerer, 2021). However, the potential risk of technical debt is not visible to senior management decision makers.

Gould et al. (2024) developed the serious game TD-SIM based on a real-world ROC case study to address CAS complexities and improve awareness of technical debt. Improved situational awareness is a pre-cursor to improved decision making (Wehrle et al., 2022) and thus by extension more effective technical debt management. Evaluating TD-SIM’s effectiveness for different player strategies, which we call player archetypes, is essential for its ongoing refinement as a technical debt management tool. This paper presents a method to enhance serious game evaluation, with targeted insight into TD-SIM’s effectiveness in fostering epistemic change in technical debt awareness, by measuring player archetypes.

Background

Serious games, simulation games, and gaming simulations concern human participants playing a role in an artificial setting (Meijer, 2009). These formats abstract real-world scenarios for learning and are collectively referred to as Serious Games (SGs) (Hauge et al., 2012; Kriz, 2003; E. Leigh & Spindler, 2005). Evaluating serious games concerns measuring, validating and verifying serious game outcomes (Mayer et al., 2014; Petri & von Wangenheim, 2016). It is an integral part of serious game development (E. E. Leigh & Levesque, 2024; van den Hoogen et al., 2014). The primary form of serious game evaluation when focusing on player experience is achieved using pre/post-game questionnaires (Catalano et al., 2014), which suffer from issues of subjectivity (Loh et al., 2015).

Objective measures of play behaviour are possible by analysing in-game player data. (Loh et al., 2007). This can be used to assess if the strategy a player employs is in line with, or is exploiting an aberration of, a serious game’s abstraction of reality (Klabbers, 2009) such as gaming-the-game (Frank, 2012). By quantifying player strategies TD-SIM’s efficacy in promoting epistemic change in technical debt awareness can be measured, thereby quantifying opportunities to improve game design. Of the many perspectives in learning science on epistemic change (Arthars et al., 2024), we adopt that of Shaffer (2006), focusing on pedagogical goals for developing domain specific expert capabilities, to preserve relevance across serious game contexts.

Serious games research has largely examined strategy through pedagogical lenses (Bellotti et al., 2011; Cheng et al., 2015). Westera (2017) proposes a computational model of player strategy, but it has not been applied to an actual serious game. This leaves a gap in evaluating player strategy empirically. We address this by introducing a method for measuring strategies used by real-world players of TD-SIM (participants), analysing in situ game data within a Bayesian framework to quantify distinct player archetypes.

Overview of TD-SIM

A mid-game view of TD-SIM developed by Gould et al. (2024) is shown in Figure 1. TD-SIM was a single player serious game hosted on a stand-alone laptop, written in the Python language using the PyGame package. TD-SIM was designed to support an epistemic state change in industrial supply chain knowledge of technical debt. The bolded items in the following section correspond to the game design elements depicted in Figure 1.

Figure 1.

TD-SIM mid-game Showing Waste, Spanner and Cog Movements.

TD-SIM players move randomly created red Spanners or yellow Cogs from a Factory to either W/House (Warehouse) or Dispatch locations, and green Waste Squares to the Waste location. Players complete as many turns as possible in six minutes. The objective was to create in-game cash each turn (Turn $) dispatching products, whilst meeting turn Objectives. At random intervals throughout a turn players were presented options to upgrade the game using cash, which incurred technical debt if chosen. The more cash spent upgrading the game, the less technical debt incurred in line with real-world experience (Cunningham, 1992). A technical debt pay-down option (remove technical debt) was provided if technical debt exists.

Balancing technical debt creation, removal, and cash generation, players explore the impact of their decisions (management). Strategies employed by players were central to their learning experience. For instance, players whose strategy was to proactively minimise technical debt creation whilst generating as much cash as possible act in line with TD-SIM’s intent. A less appropriate strategy (from the game designer’s perspective) was for players to perform well by achieving a high cash outcome with minimal technical debt management, signalling that TD-SIM was being gamed. Alternatively, when players’ choices reflected prolonged engagement with basic mechanics rather than task-relevant strategy, TD-SIM was imposing unnecessary cognitive load, making it harder to master. Evidence of these later behaviours indicates limitations in TD-SIM’s ability to foster intended epistemic change in technical debt awareness, whilst providing insight into game design improvement opportunities.

Methodology

Research Question

We address two questions: (RQ1) How to measure strategies used by simulation participants? and (RQ2) What insights do such strategies reveal?

Using statistical analysis within a Bayesian framework applied to prior TD-SIM data, our methodology identifies, classifies, and clusters individual strategies into player archetypes. These archetypes provide insight into player behaviour and inform the game’s effectiveness in supporting epistemic change. First, we outline our Bayesian framework, followed by a description of our methodology.

Bayesian Framework Overview

Our model calculates which predefined play strategies, our agent archetypes, a participant’s game data is probabilistically equal to. When repeated for all participants, clustered results provide insight into strategies used by players, which we refer to as player archetypes. These probabilities cannot be calculated directly from game data. We use a Bayesian framework to reveal conditional probabilities that can be calculated using available game data. A discussion of our Bayesian framework is provided in our supplementary information.

Model Methodology

Steps of our methodology are illustrated in Figure 2, each outlined below.

Figure 2.

Methodology.

Step 1: Define Agent Archetypes

Agent archetype strategies are identified from unique combinations of TD-SIM participant actions. Actions are elicited, with parameters (discussed below), by analysing participant data. Actions and parameters identified are shown in Table 1 labelled with an action name, abbreviation, action description, and parameter range. Table 1 actions are; Target (TG) the total amount of technical debt a participant aims to achieve within a game, Scheme (SC) how quickly progress toward a technical debt target is made using TD-SIM’s mechanics, Management (MG) is to pay down accumulated technical debt during play, Optimisation (OP) is using TD-SIM mechanics to optimise game rewards (in-game cash per turn) using the warehouse to store cogs or spanners, Action 1 (A1) is the action of bypassing management activities, and Action 2 (A2) is the action of using all menu features (such as return-to-prior menu). TD-SIM’s mechanics are fully described in Gould et al. (2024).

Table 1.

Actions and Parameters Used to Identify Each Agent Archetype’s Strategy. Technical Debt (TD) Growth Parameters are Minimum (MIN), Maximum (MAX), or Randomised (RND).

Action name	Action abbreviation	Action description	Action parameters
Target	TG	Percent TD	85, 53, 43, 23
Scheme	SC	TD growth	MIN, MAX, RND
Management	MG	Pay down TD	T (True), F(False)
Optimisation	OP	Optimising in-game cash reward	T (True), F(False)
Action 1	A1	Bypass TD management activity	T (True), F(False)
Action 2	A2	Navigate to previous menu	T (True), F(False)

Among the actions in Table 1, Target is the only one parameterized as continuous data, affording sensitivity testing of our method to closely related agent archetypes. To prevent trivial attainment of a target, prime number targets are used (53%, 43%, 23%), given technical debt upgrades occur in +5% (MIN) or +15% (MAX) increments. RND is a random strategy with equal probability of MIN and MAX occurring. The maximum achievable target under all schemes is 85%. Remaining parameters are binary (T/F), either enabled or disabled respectively. Each agent archetype is uniquely defined by concatenating actions (e.g., paying down TD, bypassing technical debt management) and associated parameters.

Step 2: Apply Strategy Model

Using identified agent archetypes, we calculate the probability of each agent archetype’s strategy occurring given participant game data. We augment probabilities with participant outcome data so that player performance can be incorporated into subsequent classification and clustering. Specifically, technical debt generated and cash reward earned per game are extracted from participant games. The resulting data matrix contains columns of conditional probabilities for all agent archetypes alongside technical debt generated and cash rewards achieved; each row representing a single participant’s game. This matrix is imported into Python 2025.2 as a Pandas Data Frame for analysis.

Step 3: Classify and Cluster Strategies

Principal Component Analysis (PCA) is used to reduce the columns of our data matrix (called dimensions) down to a smaller set of Principal Components (PCs). Each principal component is a unique combination of dimensions from the data matrix that captures a specific portion of data matrix variance (Abdi & Williams, 2010). PCA is an established means of classification in serious games, used for instance by Rodrigues and Brancher (2018). We perform PCA using the Python library sklearn and associated packages.

We first apply a StandardScaler to normalize column data to a mean of zero with a standard deviation of one. This is important because our analysis incorporates agent archetype probabilities, technical debt generated, and cash rewards achieved, with different scale dimensions, something PCA is sensitive to. Second, PCA is fitted to the normalized data. The standard practice of considering principal components with eigenvalues greater than one is adopted to identify principal components for further analysis (Orji et al., 2014).

K-Means clustering is applied to identified principal components. It is an unsupervised clustering method used in serious game data analysis, often in conjunction with PCA (Kanungo et al., 2002; Rodrigues & Brancher, 2018). Clustering is achieved by minimizing the within-cluster sum of square (WCSS) distances between each point and its corresponding centroid, for a number, k, of pre-selected centroids (Kanungo et al., 2002).

Over an increasing number of k centroid trials, WCSS decreases. The elbow method for cluster analysis is used to assess the optimum number of clusters, the point at which WCSS decreases become insignificant to the analysis (Humaira & Rasyidah, 2020). The player archetype of each cluster is assessed using techniques from principal component analysis.

Step 4: Characterise Player Archetypes

Significant principal components for a cluster are the ones with the highest average principal component importance scores across members. Cluster member importance scores are calculated as principal component squared cosine values (Abdi & Williams, 2010). Strategic characteristics are derived for each cluster’s significant principal components.

Strategic characteristics of each significant principal component are identified by analysing the dimensions from the data matrix that contribute most to it. Overall strategic characteristics of a cluster’s most significant principal component are its player archetype, subsequently named to reflect the behaviours observed. The amount a dimension contributes to a principal component is measured as the square of each dimension’s correlation coefficient, it’s so called loading score (Abdi & Williams, 2010). To reduce analytical effort the six highest ranking dimensions by loading score of each principal component are considered for strategy analysis.

Results

Agent Archetypes

We identify 57 unique agent archetypes. Equation (1), provides an agent archetype example. The strategy of this agent archetype is to target mid-range technical debt growth TG (53%) by balancing management MG(T) with optimised game outcomes OP(T), whilst making use of TD-SIMs functionality, A1(T) and A2(T). Technical debt growth is achieved using minimal increments of +5%, SC(MIN). From a design perspective, this strategy reflects a balanced combination of management, optimisation, and mechanics-focused play, making it suited to eliciting behaviours that TD-SIM is designed to provoke.

TG (53)_SC (MIN)_MG (T)_OP (T)_A 1 (T)_A 2 (T)

(1)

Model Application

In prior work 17 participants play four games each with five turns, providing 340 games in total, to which our model is applied. We augment calculated probabilities with percentage technical debt generated, and quantity of cash achieved per game. Our complete data matrix consists of 57 columns of conditional probabilities alongside one column of percentage technical debt generated and one column of cash reward obtained, 59 dimensions in total, by 340 rows of participant data.

Principal Component Classification and K-Means Clustering

Principal component analysis reduces 59 dimensions to 17 principal components, accounting for over 70% accumulated data matrix variance. Principal components are ordered and labelled from PC_1 (12.6% variance), down to PC_17 (1.8% variance).

Successive trials of K-Means clustering are applied to the 17 principal components. Four clusters are optimal using the elbow method. Final clusters are presented in Figure 3 below, with clusters labelled A, B, C and D for reference.

Figure 3.

PCA Fitted Data as 4 Clusters, each Labelled Separately as A, B, C, D in a Black Diamond, each Designated with Corresponding Player Archetype.

Significant Principal Components (Importance Score Analysis)

Summarised importance results by cluster are presented in Table 2, with significant principal component/s for each cluster shown in the Significant PC/s column. Clusters A, B and C each have a single significant principal component, respectively PC_2, PC_1 and PC_3. The most significant principal component for cluster D is PC_3, accounting for 15% normalised importance. To strengthen and align to other clusters, PC_1 and PC_4 are also included, assigning three significant principal components to cluster D with total normalised importance of 42.4%.

Table 2.

Normalised Aggregated Importance Scores for Each Cluster for PC_1 to PC_17.

Cluster^b	Size^a	Median	Standard deviation	Max	Min	Significant PC/s
A	20%	1.3%	10.9%	45.7%	0.3%	PC_2 (45.7%)
B	11%	1.4%	12.8%	54.8%	0.5%	PC_1 (54.8%)
C	15%	1.4%	9.6%	41.3%	0.4%	PC_3 (41.3%)
D	54%	4.4%	4.3%	15.4%	1.7%	PC_1 (14.6%) PC_3 (15.4%) PC_4 (12.4%)

^aCluster size as a percentage of participant game data in the cluster (340 games in total).

^bSee Figure 3 for cluster labels.

Strategic Characteristics (Loading Score Analysis)

Our classification and clustering method yielded unique dimensions for each principal component, indicating little sensitivity to agent archetypes with similar strategies. Aggregated results for the top six dimensions for PC_1 to PC_4 are shown in Table 3. Dimensions for all principal components are mostly agent archetypes (AA) with PC_1, PC_2 and PC_4 equal to AA (100%). For PC_3, five out of six dimensions are agent archetypes, with one contribution from the technical debt dimension. To glean the strategy of each principal component, contributing agent archetypes are analysed across constituent actions; TG, SC, MG, OP, A1 and A2. The TD dimension for PC_3 is discussed separately below.

Table 3.

Dimension Summary for the Six Highest Contributing Dimensions of Principal Components.

	Dimension Type^a	TG	SC	MG^b	OP^b	A1^b	A2^b
PC_1	AA (100%)	53%	MIN(100%)	T (67%)	T (100%)	F (83%)	F (83%)
PC_2	AA (100%)	53%	MIN(50%)	T (67%)	F (100%)	F (100%)	F (100%)
PC_2	AA (100%)	53%	RND (50%)	T (67%)	F (100%)	F (100%)	F (100%)
PC_3	AA (83%) TD (17%)	72%	MAX (80%)	F (60%)	T (100%)	F (100%)	F (100%)
PC_3	AA (83%) TD (17%)	72%	RND (20%)	F (60%)	T (100%)	F (100%)	F (100%)
PC_4	AA (100%)	41%	RND (83%)	T (100%)	T (100%)	T (100%)	T (100%)
PC_4	AA (100%)	41%	MIN(17%)	T (100%)	T (100%)	T (100%)	T (100%)

^aAA = Agent Archetype / TD = Technical Debt.

^bT = Percentage True / F = Percentage False.

In Table 3, TG is the arithmetic average of percentage target technical debt, SC is percentage contribution of a scheme used. For management (MG), optimisation (OP), and functionality (A1, A2) percentages of enabled (T) or disabled (F) are shown depending on which is greater.

Cluster A Strategic Characteristics (PC_2)

The single significant principal component of cluster A is PC_2. Referring to Table 3, PC_2 has an average mid-range technical debt target of 53%, and balanced technical debt growth with scheme MIN(50%)/RND (50%). Some rapid technical debt growth is present as the RND scheme is an equal mix of MIN and MAX. Moderate technical debt management occurs T (67%), with no attempt to optimise cash rewards F (100%), or use A1 or A2 functionality, both F (100%). The bias toward technical debt management, with some exploratory technical debt growth, with no attempt to optimise cash implies cluster A represents a Managers player archetype.

Cluster B Strategic Characteristics (PC_1)

The single significant principal component of cluster B is PC_1 with a mid-range target of technical debt creation (53%), using a minimalistic scheme, MIN(100%). There is a slight bias towards managing technical debt T (67%) across contributing agent archetypes, with a single focus on optimizing cash creation T (100%). There is largely no attempt to utilise TD-SIM functionality with A1 and A2 both being F (83%). The strategic behaviour of this cluster is to cautiously target mid-range TD, with a focus on optimising cash, suggesting an attempt to game TD-SIM rather than experiment with management. For these reasons we assess cluster B as a Gaming player archetype.

Cluster C Strategic Characteristics (PC_3)

Cluster C also has a single significant principal component. PC_3 strategy is about targeting high technical debt TG (72%), creating it as quickly as possible with a MAX (80%) scheme, whilst optimising in-game cash reward T (100%). Minimal technical debt management occurs T (40%)/F (60%) with no attempt to use A1 or A2 mechanics, both F (100%). For these reasons cluster C is assessed as a Learning player archetype, with cluster members experimenting how far to push technical debt growth but not necessarily experimenting with how to manage technical debt. This assessment is reinforced by the occurrence of the TD dimension, the only principal component to feature a non-agent archetype dimension.

Cluster D Strategic Characteristics (PC_1, PC_3, PC_4)

Cluster D has three significant principal components, of which all but PC_4 have been addressed above. PC_4 strategy has a low technical debt target of 43%, achieved predominantly using a RND technical debt growth strategy RND (83%). MG, OP, A1 and A2 are all employed suggesting an all-in strategy with no obvious single intent. We attribute a Testing strategic theme to PC_4. Cluster D is a combination of strategies with cluster members spreading their game efforts approximately equally between Gaming (34%), Learning (36%) and Testing (29%) when apportioned by importance scores in Table 2 using data from the Significant/PCs column.

Discussion

Player Archetypes, Epistemic Change, and Insights

Evaluation of TD-SIM participant game data yields four clusters, representing four unique player archetypes, shown in Figure 3. TD-SIMs efficacy for epistemic change in technical debt awareness and insight into improvement opportunities stem from understanding player archetypes as a reflection of player behaviour.

Managers are exploring technical debt management through creating and paying it down, a positive re-enforcement that TD-SIM provided these cluster members an opportunity for epistemic change. This cluster however only represents 20% of total participant games, Table 2. TD-SIM is only 20% effective in achieving intended outcomes for the total player cohort. For TD-SIM to become an embedded technical debt management tool, its efficacy for epistemic change must increase. By quantifying current performance, future improvements have a baseline to compare against.

Cluster C players correspond to a Learning player archetype, exhibiting strategic patterns that closely align with Managers, differentiated primarily by use of optimisation mechanics (see Table 3). This proximity indicates a design lever. Targeted adjustments to optimisation-related mechanics may promote future Learning archetypes to adopt behaviours more aligned with Mangers. Accordingly, our method identifies and quantifies specific areas of the game for refinement.

The least desired player archetype, and one with the most opportunity for improvement, is cluster D, a mix of Gaming, Learning and Testing. Over 50% of participant game data resides in cluster D, suggesting further player archetype breakdown is warranted. As a data driven approach our method provides information for deep diving into potential sub-clusters. Further work is needed in this space to understand opportunities and capabilities of our methodology.

Cluster B player archetype is Gaming. The ability of players to game TD-SIM indicates an imbalance of reward and challenge. It is considered a positive outcome that only 11% of total games played used this strategy, small compared to the 50% of participant games employing a mixed player archetype. This contrast indicates where design improvements could be prioritised.

Our findings provide quantitative confirmation of insights reported in earlier TD-SIM research (Gould et al., 2024). For instance, TD-SIM was previously assessed to be unbalanced regarding technical debt management versus gaming-the-game. Our evaluation of player archetypes agrees, quantifying 11% of TD-SIM players employed a Gaming strategy.

Method Execution

Our model and methodology are conceptually complex. In addition to the mathematical rigour of our Bayesian analysis, the application of importance scoring and loading score evaluation represent non-trivial processes essential to achieving our outcomes. Despite being challenging, and the literature being silent on player strategy assessment, it is possible for player strategies to be measured. Notwithstanding this complexity, the execution of our method is relatively straightforward once established, as it is supported by automated Python scripts.

Some subjective interpretation of clustering, classification, and strategic analysis is necessary, such as the number of contributing principal component dimensions to use in strategy characterisation, or the elbow method to determine the optimal number of clusters, albeit guided by accepted data science principles (Humaira & Rasyidah, 2020). Sensitivity analysis of our method to such subjective decisions could provide guidance on the extent to which such choices influence the outcomes observed and thus the overall objectivity of our method. Correspondingly, no significant sensitivity to closely related agent archetypes is observed.

Model Findings

Our goal is to address the research questions, (RQ1) How can we measure strategies used by simulation participants? And (RQ2) What insights do such strategies reveal?

We answer (RQ1) by evaluating TD-SIM using our method, we find and measure four player archetypes. Our player archetypes are; cluster A with a Managers player archetype, cluster B a Gaming player archetype, cluster C a Learning player archetype, and with cluster D a mixed player archetype consisting of Learning, Gaming, and Testing archetypes.

In answer to (RQ2) we assess TD-SIM’s efficacy for promoting epistemic state change, finding 20% of participants have a Mangers strategy which best aligns to the intent of TD-SIM. Additionally, we find quantified opportunities to improve this outcome. Our approach advances serious game evaluation by offering a systematic method for characterising player strategies, measuring serious game effectiveness, and quantifying opportunities for improvement. Our method, while analytically complex, remains feasible to implement.

Limitations and Suggestions for Future Research

Notwithstanding these findings, the work presented here remains experimental and requires further validation. The available sample is limited in size and derived from a single serious game. Because our method is generalisable, it can be applied across a broader range of games, creating opportunities to extend this research with larger and more diverse data sets.

Our method requires subjective choices albeit guided by accepted data science principles such as the elbow method (Humaira & Rasyidah, 2020). Such choices naturally influence our outcomes and potentially our findings. Future work on sensitivity analysis of our method could address such weaknesses.

Our Bayesian model assumes all game states are visible, which holds for agent archetypes, but not necessarily for participants, as players for instance may not visualise all game states during a game. Future work to align player and agent archetype state visibility through potentially the use of Hidden Markov Models could address this.

Conclusions

This paper presents a novel method for measuring serious-game participant strategies through the construction of player archetypes. Using Bayesian analysis, principal component analysis classification, and K-means clustering, it is possible to derive and measure player archetypes of real-world game data. Although analytically demanding, our methodology demonstrates that strategy measurement is feasible. We improve serious game evaluation by providing a means of measuring a game’s effectiveness in supporting players’ epistemic development within a target knowledge domain, quantifying opportunities for serious game improvement.

Supplemental Material

Supplemental material - Measuring Simulation Participant Playing Strategies

Supplemental material for Measuring Simulation Participant Playing Strategies by David Gould, Tim French, Melinda Hodkiewicz in Simulation & Gaming.

Footnotes

Declaration of Conflicting Interests

All opinions and conclusions drawn are those of the author alone. To the extent that this publication draws on information provided by study participants in this or prior work, it should not be assumed that any views expressed herein are also necessarily those of such participants. The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

No funding was required for this research; it is a contribution toward the lead author’s PhD.

Informed Consent

The research submitted for consideration only uses data from prior research. The prior research did use human participants, to this extent the outcome of the ethics review as conformation of informed consent has been provided. Please note, no human participants were used in this current research.

ORCID iD

Dave Gould

Supplemental Material

Supplemental material is available online.

Author Biographies

Mr Dave Gould has more than 25 years of experience in the mining industry, beginning his career in Perth as a process technician in mineral refining. Over the course of his professional practice, he has developed a sustained interest in interdependent Manufacturing Execution Systems (MES) that support contemporary mining and processing operations. A decade after completing his MBA, he returned to postgraduate study to pursue a PhD in Software Engineering. His research examines technical debt within interdependent information systems, with a particular focus on large and complex mining supply chains. He is also passionate about the use of serious games as a means to enhance technical debt awareness and education in complex supply-chain environments.

Associate Professor Tim French is a computer scientist at the University of Western Australia whose research focuses on logic, artificial intelligence, and knowledge representation. His work examines reasoning about uncertainty, probabilistic reasoning, and awareness in multi-agent systems, with contributions spanning axiomatization, expressivity, and decision procedures. He also applies formal reasoning and AI methods to industrial contexts, including automated planning, expert-knowledge capture, and machine learning for complex processes. He serves as the Regional Contest Director for the South Pacific Programming Contest and has held leadership roles in major AI conferences.

Professor Melinda Hodkiewicz is an engineering academic at the University of Western Australia whose work focuses on improving maintenance, asset management, and safety practices through data-centric methods. Her research spans natural language processing, knowledge graphs, and ontology-based interoperability for industrial maintenance and safety data, and she has played a leading role in national and international standards development, including contributions to ISO 55000 for asset management. A Fellow of the Australian Academy of Technology and Engineering, she has held visiting fellowships at the Alan Turing Institute and the U.S. National Institute of Standards and Technology, and has been recognised with the MESA Medal for lifetime contributions to the asset management community.

References

Abdi

Williams

L. J.

(2010). Principal component analysis. Wiley interdisciplinary reviews: computational statistics, 2(4), 433-459.

Arthars

Markauskaite

Goodyear

(2024). Constructing shared understanding of complex interdisciplinary problems: Epistemic games in interdisciplinary teamwork. Journal of the Learning Sciences, 33(2), 405-442.

Bekebrede

Lukosch

(2015). Understanding complex systems through mental models and shared experiences: a case study. Simulation & Gaming, 46(5), 536-562.

Bellotti

Ott

Arnab

Berta

de Freitas

Kiili

De Gloria

(2011). Designing serious games for education: from pedagogical principles to game mechanisms. Paper presented at the Proceedings of the 5th European Conference on Games Based Learning.

Besker

Martini

Bosch

(2019). Software developer productivity loss due to technical debt—A replication and extension study examining developers’ development work. Journal of Systems and Software, 156, 41-61.

Catalano

C. E.

Luccini

A. M.

Mortara

(2014). Best practices for an effective design and evaluation of serious games. International Journal of Serious Games, 1(1), e1-e13.

Cheng

M.T.

Chen

J.H.

Chu

S.J.

Chen

S.Y.

(2015). The use of serious games in science education: a review of selected empirical research from 2002 to 2013. Journal of computers in education, 2(3), 353-375.

Cunningham

(1992). The WyCash portfolio management system. ACM SIGPLAN OOPS Messenger, 4(2), 29-30.

Frank

(2012). Gaming the game: A study of the gamer mode in educational wargaming. Simulation & Gaming, 43(1), 118-132.

10.

Gould

French

Hodkiewicz

(2024). Simulating Complex Adaptive Software System Technical Debt. Paper presented at the International Simulation and Gaming Association Conference, Aukland, New Zealand.

11.

Government of Western Australia . (2025). Western Australia Iron Ore Profile – August 2025. Government of Western Australia Retrieved from https://www.wa.gov.au/system/files/2025-09/waironoreprofileaug2025.docx

12.

Hauge

J. M. B.

Pourabdollahian

Riedel

J. C.

(2012). The use of serious games in the education of engineers. Paper presented at the IFIP International Conference on Advances in Production Management Systems.

13.

Humaira

Rasyidah

(2020). Determining the appropiate cluster number using elbow method for k-means algorithm. Paper presented at the Proceedings of the 2nd Workshop on Multidisciplinary and Applications (WMA).

14.

Kanungo

Mount

D. M.

Netanyahu

N. S.

Piatko

C. D.

Silverman

A. Y.

(2002). An efficient k-means clustering algorithm: Analysis and implementation. IEEE transactions on pattern analysis and machine intelligence, 24(7), 881-892.

15.

Klabbers

J. H.

(2009). The Magic Circle: Principles of Gaming & Simulation: BRILL.

16.

Kriz

W. C.

(2003). Creating effective learning environments and learning organizations through gaming simulation design. Simulation & Gaming, 34(4), 495-511.

17.

Kruchten

Nord

Ozkaya

(2019). Managing Technical Debt: Reducing Friction in Software Development: Addison-Wesley Professional.

18.

Lakshmanan

Roy

Gorain

(2019). Case Study: Digital Disruption in the Mining Industry. In Innovations and Breakthroughs in the Gold and Silver Industries (pp. 199-241): Springer.

19.

Leigh

Spindler

(2005). Congruent facilitation of simulations and games. In Gaming, simulations, and society: Research scope and perspective (pp. 189-198): Springer.

20.

Leigh

E. E.

Levesque

L. L.

(2024). Facilitating simulations: Edward Elgar Publishing.

21.

Loh

C. S.

Anantachai

Byun

Lenox

(2007). Assessing what players learned in serious games: in situ data collection, information trails, and quantitative analysis. Paper presented at the 10th International Conference on Computer Games: AI, Animation, Mobile, Educational & Serious Games (CGAMES 2007).

22.

Loh

C. S.

Sheng

I.H.

(2015). Predicting expert–novice performance as serious games analytics with objective-oriented and navigational action sequences. Computers in Human Behavior, 49, 147-155.

23.

Mayer

Bekebrede

Harteveld

Warmelink

Zhou

Van Ruijven

Wenzler

(2014). The research and evaluation of serious games: Toward a comprehensive methodology. British journal of educational technology, 45(3), 502-527.

24.

Meijer

(2009). The organisation of transactions: Studying supply networks using gaming simulation (Vol. 6): Wageningen Academic Publishers.

25.

Orji

Vassileva

Mandryk

R. L.

(2014). Modeling the efficacy of persuasive strategies for different gamer types in serious games for health. User Modeling and User-Adapted Interaction, 24(5), 453-498.

26.

Petri

von Wangenheim

C. G.

(2016). How to evaluate educational games: a systematic. Journal of Universal Computer Science, 22(7), 992-1021.

27.

Ramasubbu

Kemerer

C. F.

(2021). Controlling Technical Debt Remediation in Outsourced Enterprise Systems Maintenance: An Empirical Analysis. Journal of Management Information Systems, 38(1), 4-28. doi:https://doi.org/10.1080/07421222.2021.1870377

28.

Rodrigues

L. A.

Brancher

J. D.

(2018). Improving players' profiles clustering from game data through feature extraction. Paper presented at the 2018 17th Brazilian Symposium on Computer Games and Digital Entertainment (SBGames).

29.

Shaffer

D. W.

(2006). Epistemic frames for epistemic games. Computers & education, 46(3), 223-234.

30.

van den Hoogen

Meijer

(2014). Debriefing in gaming simulation for research: Opening the black box of the non-trivial machine to assess validity and reliability. Paper presented at the Proceedings of the Winter Simulation Conference 2014.

31.

Wehrle

Wiens

Schultmann

(2022). Application of collaborative serious gaming for the elicitation of expert knowledge and towards creating Situation Awareness in the field of infrastructure resilience. International Journal of Disaster Risk Reduction, 67, 102665.

32.

Westera

(2017). How people learn while playing serious games: A computational modelling approach. Journal of Computational Science, 18, 32-45.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.18 MB