Systemic Design employs Causal Loop Diagrams (CLDs) to document and visualize the dynamics of complex systems, describing their relevant factors, called variables, and the causal relationships between them. A systemic approach is essential to understanding causal relationships and improving the decision-making process, especially in complex, multidisciplinary contexts where the implementation of reactive and proactive measures is pivotal (e.g., public health, transport, urban development, public involvement). While CLDs provide a consolidated and widely adopted visual representation, they have yet to be formalized in the context of data-driven modeling and analysis.
Systemic Design is an emerging field driven by the ambitious objective of understanding, making
sense of, and addressing complex problems in terms of “relationship and global dynamics”, rather than
isolated components. Being addressed mainly by design schools, the systemic design community has
created a stronSygstemic Design Association (https://systemic-design.or),gw/hich produced scholarly
publications (journals and conferences) and educational programs and is gaining interest and
adoption for approaching complex problems, typically fostering domain expertise exc1h].aFnrgoem[ a
foundational point of view, it capitalizes on well-established design approaches to complex challenges,
includingdesign thinking [
An important instrument of systemic design is the description of complex systems’ dynamics by means of Causal Loop Diagrams (CLDs6, ][). These diagrams, of which Figur1eshows an example instance, illustrate systems’ behaviors at an abstract level, by means of nodes and edges. Nodes typically representvariables describing factors causing or afecting the problem. Only a few variables are
CEUR Workshop
ISSN1613-0073 measurable (e.g., can be quantified precisely using a well-defined metric), and most variables express generic concepts (e.ge.x,ploitation of natural resources, orproduct quality vs product lifespan). Directed edges between nodes express causal relationships (from a source to a destination node), and – in particular – emphasize the influence of one w.r.t. another, in qualitative terms. Such influence, denoted as polarity, is positive when the growth of the value of the source variable causes the growth of the value of the destination variablen,aegnadtive when the growth of the value of the source variable determines the reduction of the value of the destination variable.
Based on these simple descriptions, diagrams are inspected for fincdaiunsgal loops, i.e., cyclic paths looping from one variable back to the same variable. Each loop is associated with a given characterization, namelybalancing (B) orreinforcing (R), based on a simple inspection of the edges involved in the loop (edges with negative polarity along a loop are counted, and a loop is balancing when the count is odd, reinforcing when the count is even).
In addition to causal loops, we define the new concepctauosfal routes, consisting of paths of edges connecting two nodes, denoted as source and destination. Similar to causal loops, causal routes can also be associated with positive and negative polarity (edges with negative polarity along a route are counted, and a route is denoteidncarseasing when the count is evedne,creasing when the count is odd).
Diagrams provide a clear picture of mutual influences; their analysis facilitates the generation of novel insights that can be leveraged to assess areas or intervention points, even those not immediately apparent, and their potential impact. The simple Causal Loop Diagram in1Fiingculruedes causal loops and causal routes.
We then consider cases when causal loops share a common variable and classifyatghreeeminags (i.e., all balancing or reinforcingd)isoargreeing (i.e., when both options are present). Similarly, we consider causal routes between the same pair of variables and classifyatghreeeminags(i.e., all routes share the same polarity)doisragreeing (i.e., some routes have diferent polarity). These analyses can help with reasoning about possible balancing or counterbalancing efects within the same diagram.
While systemic design is gaining increasing interest in the design community, so far it has not influenced technical, engineering-oriented communities much; in particular, it is not well-known to conceptual modeling scholars. In our article, we aim to build a first bridge by describing a CLD metamodel (i.e., a model of diagrams’ components) and then reasoning upon the insights that such formalization can bring. We first present a chronology and brief description of some major references on CLDs – without any claim of being exhaustive (Sec2t)i.oTnhen, we present our metamodel of CLDs (Section3), allowing us to formally define a series of new concepts, which we deem interesting for deepening a causality analysis. For describing the power of the CLD model, we present two large use cases (Section4s–5). The first one is used to explain our concepts in action, applied to the COVID19 pandemic; the second one targets the fashion industry, allowing us to apply systemic design at large, generating insights that can be leveraged to assess areas or intervention points, even those not immediately apparent, and their potential impact. Finally, we discuss the importance of assisting systemic design with our data-driven approach (Se6c)t.ion
Causal Loop Diagrams were introduced in a work written in 1986 by Richa7r],dasnodn t[hen formalized by Haraldsson in 20046[]; the latter introduces variables and their connections, explains positive and negative polarities, and the essence of balancing and reinforcing loops based upon edge polarities. It also introduces observed behavior patterns and uses them to illustrate several loop dynamics.
The concept of Stock/Flow Diagrams (SFD) was introduced by Binder8e]tinald.i[agrams explaining causal relationships; all variables are quantifiable and represensttocekist(haecrcumulations) or lfows (activity rates); SFDs are compared/contrasted with CLDs because, while the latter are informal descriptions of reality, the former are quantifiable descriptions, e.g., of physical processes. The article includes a method for progressively transforming CLDs into SFDs. SFDs are also extensively described, in a plain style, by Meadow3s][.
A mathematical perspective on CLDs is ofered by Haywar9]d. [In this work, several mathematical rules describe the variation of stock levels when connected within one or more loops, further describing each flow as a mathematical function, with first and second derivatives. The author explains several biological models, including epidemic growth and the predator-prey model. Among more speculative works, CLDs are explained in terms of Algebraic Quantum Field Theory in modern mathe1m0]a.tics [
Methods for advancing the development and use of CLDs are constantly being researched (e.g., in the
Ph.D. thesis of Kenzie1[
A formalization of the design process for complex CLD, with multiple stakeholders, is desc1r8i]bed in [ in the context of renewable energy technology (RET) adoption for hotels in Queensland (Australia). It shows several versions of CLDs, each one undergoing phases of proposition and approvals. In the ifnal CLD version, most variables are provided with some reference that explains/justifies them, and some variables introduced in the early phases of the process are removed based on deeper analysis. The process is very laborious and time-consuming, but in the end, the system designers succeed in their mission, i.e., to show the pros and cons of RET adoption at a high conceptual level; they acknowledge that achieving convergence (e.g., about which variables and relationships should be selected or retained) through several rounds of discussions involving diferent stakeholders requires a lot of time and energy.
While CLDs are efective as an instrument for describing complex systems at a high level of abstraction, bridging them to design activities concerning the development of information systems is not trivial; to our knowledge, the only method for bridging CLDs to information system design is provided by Tulinayo et al. in20[] and further discussed i2n1][.
Conceptually, CLDs can be compared to it*hferamework 2[
Causal Loop Diagrams (CLDs) represent networks of variables connected by directed edges; next, we describe the metamodel of CLDs, in the form of an ER diagram, to make their semantics accessible to the Conceptual Modeling community. The graphical language of CLDs is represented i2n, aFigure metamodel (in the sense o2f4[]) showing the available graphic modeling primitives and their abstract syntax (to provide a unique interpretation of CLDs introduced next); here we avoid specifying contextconditions, which are made clear in the text. The metamodel describes all possible CLD models, which in turn provide a description of reality. In addition to modeling the standard graphic elements, we also model new elements introduced by us, enclosed by dotted lines and highlighted by shaded (yellow) iflling.
Nodes, edges, and their properties. In CLDs, Variables are visualized as nodes; they are
characterized by anidentifier and aname; the identifier is sometimes omitted from the representation. Variables
may be quantified by means of aMeasure, in turn characterized bmyeaasureUnit and, sometimes,
anintendedLevel. In particular, some nodes may be regardedstoacsks, to represenatccumulations of
material or information that have built up over time [
Variables are connected throCuaguhsalityLinks (i.e., connections), which are directed arrows indicating how the change in one variatabileo(f the link) influences another variabhleead( of the link). Each connection haspaolarity that can be positive or negative; positive connections are represented in CLDs by continuous-line arrows, and negative connections by dashed-line arrows. Positive polarity of an edge from X to Y occurs wheangrowth of X causes a growth of Y, or, equivalentlay,reduction of X causes a reduction of Y1; conversely, negative polarity of an edge from X to Y occursawghroewnth of X causes a reduction of Y, or, symmetricallya, reduction of X causes a growth of Y. Positive and negative polarities can be defined even when they connect variables that are not measurable. Finally, the efects of a causality link can be delayed; when this aspect deserves attention, th//e issymusbeodl on the arrow, and the attribwuittehDelay is true.
FeedbackLoops, CausalRoutes and their properties. Variables and CausalityLinks enable representing arbitrary graphs over thematic regions; the relevant patterns associated with a graph can be derived from them, and represent FeedbackLoops, CausalRoutes (introduced in this paper, not included in the original definition6s][), and their properties.
A FeedbackLoop is created for each cyclic path of two or more edges going from one variable, called “target variable”, back to the same variable. Each loopbiaslaenitcihnegr(B) orreinforcing (R). This notation is typically placed on a round sticker along the loop; usually, only loops that have been identified as interesting by designers are marked on the visual diagram. The association of loops with stickers may be ambiguous when several loops interfere with each other; for this reason, we denote loops by a list of node identifiers along the cyclic path.
The loop dynamic is dictated by a simple ruloloep:s are balancing when they contain an odd number of negative connections, they are reinforcing when they contain an even number of negative connections. This rule is motivated as follows: if the number of negative connections is odd (1, 3, 5, ...), we can compose the edges’ interpretations and say that the growth of the target variable along the outgoin edge will eventually be compensated by a decrease of the same target variable produced by the last (incoming) edge of the loop, regardless of the number of intermediate steps. Thus, growths will be balanced by decreases, whereas decreases will be balanced by growths. Instead, if the number of negative connections is even (0, 2, 4,...), then a growth along the outgoing edge eventually causes a growth of the target variable along the incoming edge, generating a positive reinforcement along the loop – and likewise for a negative reinforcement. Feedback loops rheafevrenaceBehaviourPattern referred to their target variable; these are (purely conceptual) functions, drawn on a temporal scale providing an indication of how the variables will behave as a consequence of changes occurring in the loop. Patterns reflect typical stereotylpineesa:r, superlinear, and sublinear; once associated with either growth or decrease, this yields six possible behavior patterns for each loop. In particular, linear and superlinear stereotypes denote reinforcing loops, while sublinear stereotypes denote balancing loops, as they show a trend toward stabilization.
In addition to feedback loops, we introduce the new concCepatusoaflRoute, consisting of a sequence of connections from a source node to a destination node. Causal routes are characterized as increasing/decreasing, based on the even vs. odd count of negative polarities along their connections, as discussed for balancing and reinforcing loops. In additionc,utmhueliartiveTrend provides a (purely conceptual) indication of how the destination variable grows or decreases as a function of the growth or decrease of the source variable; trends reflect typical sterelointeyapre,su:perlinear, and sublinear; once associated with either growth or decrease, this yields to six possible behavior patterns for each causal route.
As an addition to the original formalization, we alsoAadltdetrhneativeLoops concept, representing those loops that have an arbitsrhaarreydVariable (not necessarily the target one); theaygarereeing when they are all balancing or reinfordciisnagr,eeing when at least one is balancing and at least one is reinforcing. Similarly, we add the conceAptltoefrnativeRoutes, representing two or more causal routes connecting the same source and destination variablesa;gtrheeeiyngarwehen they are all increasing or decreasing; theydiasragereeing when at least one is increasing and at least one is decreasing. 1We provide both readings because, when causality links form loops or routes, one of the readings more naturally explains the link role in relation to the entire route.
Feedback loops in some CLD examples are also described by means oOfbasnervedBehaviourPattern, listing several ordered observations of measured variables to show their phased evolution (growth or decrease along the cycle). As these patterns are not well formalized and we consider them as not very efective in the description of complex systems, we disregard them in CLDs that are presented next.
We deliberately excluded from the metamodel some CLD aspects, for instance, the discussion of loop
dominance [
The COVID-19 pandemic has moved the scientific community at large in order to study its efects;
a number of models have studied in-depth mitigation meas2u6r,e2s7[] economic aspects2[
Our first analysis is concerned with feedback loops that target this variable. Confirmed cases rise when a new dangerous variant becomes dominant. We recognize a main loop B1 (circuit 12, 9, 6, 12) that sees anincrease of intensive care admissions (an immediate measure of the disease’s spread and severity) causing ainncrease of interventions bringing, with a given delay, to rtehduection of confirmed cases. Another similar loop B2 (circuit 12, 8, 6, 12) justifies an increaisneteinrventions as a result of a (delayed)government action. These two alternative loops agree on their balancing efect.
Another interesting loop is B3 (circuit 12, 14, 15, 16, 9, 6, 12), which also agrees with B1 and B2. It is also balancing, along the reasoning thinactreasne of confirmed cases causes anincrease of load on health workers, hence areduced health service capacity that translates treodauced access to health services, and then to ainncrease of intensive care admissions because patients reach the hospital when their health status is already very severe; then B3 merges with B1’s last two edges. Note the reading of this balancing loop, where the odd number of negative loops produces, in the end, a balancing efect over the target variable.
However, the increase in health worker load is at the base of two reinforcing loops, R2 and R3. Loop R2 (circuit 14, 15, 14) indicates a well-knodweandly spiral in which healthcare has been trapped during the pandemic, where, as a consequence of the high load of health workers, health service capacity has been reduced, leading to even higher health worker load due to COVID-19 and other emergencies. Loop R3 (circuit 14, 15, 16, 18, 14) illustrates that a reduction of health services causes, with some delay, a reduction of prevention practices for many pathologies other than COVID-19, causing -in the long runheavier health worker loads. Thus, when R2 and R3 are observed by takhienaglththweorker load as their shared variable, these two alternative loops are both reinforcing and agreeing.
The (relatively) simple CLD shown in Figu3rsehows another reinforcing loop R1 (circuit 4, 1, 2, 4), at the intersection of the economy-government thematic regions, reflectingdaenadoltyhsepriral, this time relative to the job market, where interventions cause business restrictions, which in turn cause a reduction of economic growth, which in turn causes ainncrease of unemployment and therefore a further increase of business restrictions.
The just discussed R1 reinforcement loop introduces an interesting alternativeirnoteurtvenftrionms toeconomic growth (see Figure4A), where stimulus packages introduced by the government lead to increasing economic growth and compensate, although with delay, for the restrictions on travel and businesses, which instead cause a reduction of economic growth (as discussed above). Summary edges represented at the center of figures describing alternative routes clarify each route’s polarity: the route on the left side is increasing (it has no negative edges, hence an even number of negative connections), whereas the route on the right side is decreasing (one negative connection). This simple set of disagreeing causal routes hints at the huge, complex decision processes leading to the deliberation, by the world’s governments, of efective stimulus packages for combating the negative efects of COVID-19 on economic growth.
Another intriguing decision process is discussed in F4igBuarned concerns the connectio
ninotefrventions to theone health concept 3[
Our second use case is dedicated to analyzing the fashion industry; in particular, it is focused on a
critical assessment of the industry’s footprint and sustainability, looking for hidden/interesting aspects.
The CLD, withfashion industry footprint as its target variable, is illustrated in5F;iigtusrpeans over the
Market, Consumption, Production, and Policies thematic regions, and is weakly inspired by the CLD
presented in3[
As in our first use case, we start with a focus on feedback loops, shown in F6,igwuhriech highlights several alternative loops sharinpgutrchhease of new clothes variable. Reinforcement loop R1 describes a simple consumerist behavior (circuit 7, 9, 12, 7): with purchases of new clothes, owned clothes increase, then thrown-away clothes increase, and finally, new clothes are purchased. Along this reasoning, loop B0 (circuit 9, 12, 9) simply indicates that an increase in thrown-away clothes causes a decrease in owned clothes.
More interestingly, the balancing loop B1 (circuit 12, 10, 7, 9, 12) signals that when clothes are thrown away, there is a higher availability of second-hand clothes, andrtehduiscecathne purchase of new clothes; thus, R1 and B2 are alternative and disagreeing. An increasing causal route, common to loops R2 and B2, goes frompurchases of new clothes toclothes thrown away (sequence 7, 8, 11, 13, 12) occurs because, with increased purchases of new clothes, production intensity increases, but then the quality of clothes decreases; this reduces the lifespan of clothes, and eventually, more clothes are thrown away. This route can be completed both as a reinforcing loop R2 (by considering the direct connection to purchases of new clothes) and a balancing loop B2 (through second-hand clothes availability, node 10).
Other feedback loops share tphuerchase of new clothes; in particular, reinforcing loop R3 (circuit 7, 6, 3, 4, 7), along classic mechanisms of expanding markets, indicates that an increase in purchases of new clothes increases the fashion industry’s profit, which in turn increases investments in the market, which in turn increases customer’s desire to buy, yielding to increasing of purchases of new clothes; this loop only includes positive connections. A more subtle balancing loop B3, however, involves consumers’ awareness of the fashion industry’s footprint (circuit 7, 8, 5, 1, 2, 7). Along with an increase in purchases of new clothes, production intensity rises, which then causes greater resource exploitation and therefore an increase in the fashion industry’s footprint. This may, in the long run, afect consumer awareness and cause a reduction in new clothes purchases.
At this point, six loops insist on the shared vapruiarcbhlaese of new clothes, out of which three are balancing (B1, B2, B3) and three are reinforcing (R1, R2, R3); understanding their interdependencies and determining the strengths of each of them requires deeper analyses, but their underlying mechanisms are well identified.
The analysis of some alternative and disagreeing causal routes of the original CLD5pinroFvigiduerse more insights. We first consider, in Figu7rAe, the polarity oenfvironmental regulations upon thefashion industry profit . The standard causal route, with a negative polarity, indicates that environmental regulations require higher industry investment in sustainability, causing a reduction of production intensity and, therefore, an increase in production costs, and then a decrease in profits. However, a less obvious causal route, with a positive polarity (contributed by a sequence of connections with positive polarities) indicates that greater investments in sustainability can generate greater innovatio for supporting the sustainability transition, followed by higher investments in marketing that highlight these achievements, and these in turn may rise the customer’s attention and desire, leading to higher purchases of new clothes and eventually to an increase of profits.
We next consider, in Figur7Be, the influence ofpurchases of new clothes upon thefashion industry footprint. As before, the standard causal route, with a positive polarity, indicates that an increase in purchases causes a more massive production, a higher production intensity, hence higher exploitation in all six considered categories, and eventually a rise in the industry footprint. However, an alternative route considers the rise of profits, which descends from higher purchases of new clothes, which are used for investing in sustainability, thereby facilitating the industry’s transition and eventually obtainin higher sustainability, leading to a reduced industry footprint.
CLDs are a formalism providing high-level descriptions of complex systems; as such, they allow focusing on areas of intervention that can be deepened/assessed by means of further analysis. An example is provided, in our fashion design use case, by the insight that improving the fashion footprint not only causes an increase in the cost of production but also can be a factor for pushing innovation and a means for stimulating the consumer’s awareness and creating a new strategy for market penetration, thus balancing the “negative” impact of increased costs with possible “positive” outcomes – process innovation and marketing positioning. A complementary analysis could be perfoframste-dfaoshnion influence on countries from the Global South, such as Bangladesh, with a garment industry worth 55B US dollars a year, now facing an unsettled future after protests, due to low pay and poor working conditions, particularly for wom3e5n]. [
In parallel work36[], we describe a demo application that implements all our concepts; the prototype includes as predefined cases the COVID and Fashion Design use case; it also includes the use case about renewable energy technology (RET) adoption for hotels in Queensland (Australia) des1c8r]i,btehdein [ largest and most documented CLD that we found in the literature, with 42 variables, 74 relationships, 143 causal loops and 62171 causal routes. Our prototype incoporates L3O7]O, PaYo[pen-source visual tool, for entering a new CLD; it supports the systematic analysis of causal loops and causal routes, the selection of specific loops and routes for direct comparison, and the production of a PDF report where the result of an explorative interaction can be extracted, together with selected loops or route The prototype demonstrates the potential of computational tools to support systemic modeling and reasoning; practitioners can use it to identify leverage points and explore the systemic consequences of interventions, while researchers gain a framework for formalizing and querying alternative models within the same domain.
The current prototype is a basis for building a more robust infrastructure, capable of documenting the design process for a given use case (using diferent progressive CLD versions) and organizing several use cases within a CLD repository. The combination of metamodeling, structured representation, and a queryable CLD repository will provide the foundations for more deliberate and evidence-informed systemic design.
Looking ahead, future work could include the integration of dynamic simulation capabilities, the use of external data sources for real-time or evidence-based modeling, and the implementation of collaborative features. Additional potential lies in the application of machine learning to detect recurri causal archetypes and in the development of natural language interfaces to support accessibility. An additional interesting research concerns the exploitation of suitably trained Large Language Models t automatically recognize causal relationships and their polarities from texts.
As a further opportunity for investigation, we noted possible synergies wi*itfrhatmheework. Currently, the systemic design community andi*thcoemmunity both ofer valuable perspectives for modeling complex systems, but they difer significantly in focus and methodology. While CLDs provide a macro-level view of system dynamics, tih*eframework ofers a micro-level view of stakeholder motivations and interactions. Taken together, these approaches could complement each other, with CLDs shedding light on system-level feedbackia*nrdevealing the underlying intentions that drive individual and organizational actions.
The authors have not employed any Generative AI tools. [13] T. J. Gandee, P. J. Giabbanelli, Combining natural language generation and graph algorithms to explain causal maps through meaningful paragraphs, in: International Conference on Conceptual Modeling, Springer, 2024, pp. 359–376. [14] N.-Y. G. Liu, D. R. Keith, Leveraging large language models for automated causal loop diagram generation: Enhancing system dynamics modeling through curated prompting techniques, arXiv preprint arXiv:2503.21798 (2025). [15] J. McGlashan, et al., Quantifying a systems map: network analysis of a childhood obesity causal loop diagram, PloS one 11 (2016) e0165459. [16] S. Allender, et al., A community based systems diagram of obesity causes, PloS one 10 (2015) e0129683. [17] R. Iannone, et al., Modeling fashion retail supply chain through causal loop diagram, IFAC
PapersOnLine 48 (2015) 1290–1295. [18] N. Dhirasasna, O. Sahin, A multi-methodology approach to creating a causal loop diagram, Systems 7 (2019) 42. [19] L. S. Liebovitch, et al., Approaches to understanding sustainable peace: qualitative causal loop diagrams and quantitative mathematical models, American Behavioral Scientist 64 (2020) 123–144. [20] F. Tulinayo, et al., From a system dynamics causal loop diagram to an object-role model: a stepwise approach, Journal of Digital Information Management 10 (2012) 174–186. [21] F. Tulinayo, et al., Enhancing the system dynamics modeling process with a domain modeling method, Int. J. of Cooperative Information Systems 22 (2013) 1350011. [22] E. S. Yu, J. Mylopoulos, Enterprise modelling for business redesign: The i* framework, ACM
SIGGROUP Bulletin 18 (1997) 59–63. [23] X. Franch, et al., The i* framework for goal-oriented modeling, Domain-Specific Conceptual
Modeling: Concepts, Methods and Tools (2016) 485–506. [24] G. Guizzardi, On Ontology, ontologies, Conceptualizations, Modeling Languages, and (Meta)Models, in: O. Vasilecas, J. Eder, A. Caplinskas (Eds.), Databases and Information Systems IV, volume 155, IOS Press, 2006, pp. 18–39. [25] C. Batini, et al., Conceptual database design: an entity-relationship approach, Benjamin-Cummings
Publishing Co., Inc., 1991. [26] R. M. Anderson, et al., How will country-based mitigation measures influence the course of the
COVID-19 epidemic?, The Lancet 395 (2020) 931–934. [27] A. Bernasconi, et al., Data-driven analysis of amino acid change dynamics timely reveals SARS
CoV-2 variant emergence, Scientific Reports 11 (2021) 21068. [28] G. Bonaccorsi, et al., Economic and social consequences of human mobility restrictions under
COVID-19, Proc. of the National Academy of Sciences 117 (2020) 15530–15535. [29] F. Pierri, et al., Online misinformation is linked to early COVID-19 vaccination hesitancy and refusal, Scientific Reports 12 (2022) 5966. [30] K. Zięba, How can systems thinking help us in the COVID-19 crisis?, Knowledge and Process
Management 29 (2022) 221–230. [31] N. Strelkovskii, E. Rovenskaya, Causal loop diagramming of socioeconomic impacts of COVID-19: state-of-the-art, gaps and good practices, Systems 9 (2021) 65. [32] O. Sahin, et al., Developing a preliminary causal loop diagram for understanding the wicked complexity of the COVID-19 pandemic, Systems 8 (2020) 20. [33] World Health Organization, One hehalttthps,://www.who.int/news-room/questions-and-answers/ item/one-healt, h2023. Last access: August 8th, 2025. [34] System Mapping Academy, System Mapping Toolkihtt,tps://miro.com/app/board/uXjVP0ou690= /?fromEmbed=1, 2025. Last access: August 8th, 2025. [35] N. Inamdar, Fast fashion drove bangladesh - now its troubled economy needs more (2024). [36] A. Bernasconi, et al., CLD-Explorer: Toward a Tool for Causal Loop Diagrams Analytics, 2025.
Companion Proc. 44th Int. Conf. on Conceptual Modeling: Posters and Demos. [37] LOOPY, https://ncase.me/loop,y2/025. Last access: August 8th, 2025.