A brief trajectory of the constituent elements of the Advocacy Coalition Framework (ACF)

This paper aims to retrace the trajectory of the Advocacy Coalition Framework (ACF), especially its policy-oriented learning process. Initially, we contrast the elements in Sabatier and Mazmanian (1980), where the first efforts to develop an analysis model are found, with the arguments found in Sabatier (1987, 1988, and 1993). Subsequently, the historical trajectory of updates and versions of the model is discussed based on the analyses in Sabatier and Jenkins-Smith (1999), Sabatier and Weible (2007), and Jenkins-Smith et al. (2018). It was possible to follow the modifications in the model, the main hypotheses built, the criticisms, and their unfoldings. As main findings, it was evident throughout the ACF construction trajectory: four versions of the model over almost 30 years and with the decisive participation of six prominent authors who contributed to its main developments since the first version, present in Sabatier and Jenkins-Smith (1993); and Sabatier and Jenkins-Smith (1999), the model gains greater analytical capacity with the intermediate variables between the external factors and the policy subsystem and analytically refines the categories internal to the subsystem. One gap identified in this trajectory, and consequent proposal for a future research agenda, is the influence of international actors and their implications on policy modifications, a condition not explicitly addressed by the ACF in its varied versions, as highlighted by Jenkins-Smith et al. (2018).


INTRODUCTION
The Advocacy Coalition Framework (ACF) is used widely in international literature as a model to analyze disputes that permeate the public policy decision-making process. According to Weible, Sabatier, Jenkins-Smith, Nohrstedt, and Henry (2011), the ACF, since its conception, has had the following characteristics: a) the main unit of analysis is the public policy subsystem; b) to understand the arrangements in the subsystem, it is essential to have a historical trajectory of public policy of no less than 15 years; c) actors involved in the public policy system may be aggregated into coalitions; and d) public policy designs are interpreted as translations of coalitions' beliefs. These assumptions are considered along the following analytical axes: coalition formation and maintenance; the process of public policy-oriented coalition learning; and changes in public policy. 1 Among these three axes, the policy-oriented learning process presents the most challenges and limits for operationalizing empirical studie 2 (Jenkins-Smith et al., 2018). However, it is the category that gives ACF its originality with the understanding that knowledge is not watertight but contingent and that the transformations guided by this process are the result of the clash between coalitions and their beliefs in the public policy subsystem. The model itself, from Professor Sabatier's early efforts in the 1980s, has incorporated contributions arising from its application globally and has shown that it is policy oriented and susceptible to change and adaptation like the learning process.
This work's main objective is to trace the trajectory of the construction of the ACF, highlighting the importance of its theoretical framework and its analytical categories for the understanding of the processes and dynamics involved in the formulation and implementation of public policies. The paper is arranged as follows: first, we contrast the elements contained in Sabatier and Mazmanian's work (1980), which marks the initial efforts to develop a model of analysis, with the arguments found in Sabatier (1987 and1988). The results of these seminal works were reaffirmed in Sabatier and Jenkins-Smith (1993). Next, to discuss policy-oriented learning, we refer to the work of Jenkins-Smith and Sabatier (1993). Subsequently, we discuss the historical trajectory of updates and versions of the model based on the analyses in Jenkins-Smith et al. (2018), Sabatier and Jenkins-Smith (1999), and Sabatier and Weible (2007). Thus, it was possible to identify the model's modifications, main hypotheses, and criticisms and their consequences. 3 To achieve its current configuration, the ACF has undergone several changes since its first versions. Notably, Sabatier, before coauthoring with Jenkins-Smith, together with Mazmanian throughout the 1970s, already proposed a prototype model for analysis to fill a gap in public policy studies -namely, the influence of the political system on the implementation of public policies. The proposal, at the time, addressed the following types of public policies: i) transformations from street-level bureaucracy to top-level bureaucracy; ii) changes in the behaviors of municipal or state bureaucrats in the distribution of resources; iii) behaviors of private actors during budget disputes (Sabatier & Mazmanian, 1980). Three elements can be noted in terms of the scope of what the ACF depicts: the relationship between bureaucrats at different levels, changes in the behaviors of those who decide public policy, and formal participation of actors outside the government who attempt to permeate the decision-making process. Sabatier and Mazmanian (1980, p. 540) express concern regarding the behavior of political actors when referring to the theoretical legacy built by other authors (Bardach, 1977;Rein & Rabinovitz, 1977;Van Meter & Van Horn, 1975) to understand the implementation decision-making process beyond the administrative and legal perspective: Collectively, these efforts provide a reasonable overview of policymaking in terms of its complexity and the variety of factors that can either assure or impede successful implementation. But we feel that more is needed. In the first place, more of an effort is needed in conceptualizing and empirically exploring the linkage between individual behavior and the political, economic, and legal context in which it occurs [...] 4 .

DIFFERENT VERSIONS OF THE ADVOCACY COALITION FRAMEWORK DIAGRAMS
The most current version of the ACF diagram resembles a feedback system, as can be seen in Figure 1. The representative analytical scheme of the model relies on two external categories, one being more stable, with no abrupt changes in the short term, called relatively stable parameters. In this set are the basic attributes of the problem of the public policy area and the distribution of natural resources, the fundamental sociocultural values and social structure, and the basic constitutional structure of the state. The set of events external to the subsystem aggregates the most dynamic variables. First, the variables indicating changes in socioeconomic conditions, such as accelerated urbanization, recurring economic crises, trade opening, and changes in public opinion reflect the change in the governing coalition system; that is, which political party or group has the greatest strength in front of the government. Finally, there are the changes in the other policies subsystem. Before affecting the internal subsystem of a specific public policy, external variables pass through the filter of two sets: opportunity structures for long-term coalitions and short-term constraints and resources available.

FIGURE 1 FLOW DIAGRAM OF THE ADVOCACY COALITION FRAMEWORK
group has the greatest strength in front of the government. Finally, there are the changes in the other policies subsystem. Before affecting the internal subsystem of a specific public policy, external variables pass through the filter of two sets: opportunity structures for long-term coalitions and short-term constraints and resources available.  Within the first set, the degree of consensus needed for deeper changes in public policy stands out. For example, if a public policy does not significantly affect the beliefs of the actors, it tends to have a high degree of consensus. The openness of the political system, a long-term category for coalitions, seeks to understand the context in countries in which corporatism is a significant element of permeability. Countries with strong corporatist traditions in state-society relations tend to restrict the participation of actors in the process. According to Milward and Francisco (1983), the intermediation of corporate interests occurs around public policy areas based on government programs. In such areas, state agencies both support pressure groups in the process of formulating public policy and rely on their support. Finally, the cross-cutting cleavage closes the relationship of intermediate intervening categories of longer duration with repercussions on the degree of consensus when there is a lot of polarization.
In the second set, short-term constraints and resources available, we observe the contingency of actors' actions in the public policy subsystem. Within this subsystem, the main unit of analysis is the advocacy coalition, tensioned by an internal competition consistent with its principles, values, and interests-a belief system. Coalitions usually give up their specific strategies and resources to permeate a government authority's decision in their favor. The result is the creation of rules that will bring consequences to events in the dynamic external subsystem, forging a feedback loop. Notwithstanding the behavior of actors, the content of public policies matters as does the political system of the country and its historical experience of democracy in addition to the international context-conditions that enable greater participation or contestation.
To deepen the study of public policy without disregarding the political system and the institutional dimension, Sabatier and Mazmanian (1980) created the analytical framework presented in Figure 2. Flowchart of the variables in the process of public policy implementation Source: Sabatier and Mazmanian (1980).
In the late 1970s study, the concepts important to the ACF model, such as advocacy coalition, belief systems, public policy-oriented learning, and the policy subsystem, were not yet delineated, but we can highlight some elements of similarity.
The problem tractability variable set, highlighted in Figure 1, is equivalent to the basic attributes category in the ACF's relatively stable parameter set-that is, what problem or public good the public policy proposes to solve/address. We also do not need much effort to identify the non-statutory variables that affect implementation, as well as events external to the subsystem in the ACF, such as categories external to the state that affect implementation (e.g., socioeconomic and technological changes). 3-Grupo alvo como percentual da população; 4-Alcance necessário para mudança comportamental.
The problem tractability variable set, highlighted in Figure 1, is equivalent to the basic attributes category in the ACF's relatively stable parameter set-that is, what problem or public good the public policy proposes to solve/address. We also do not need much effort to identify the non-statutory variables that affect implementation, as well as events external to the subsystem in the ACF, such as categories external to the state that affect implementation (e.g., socioeconomic and technological changes).
The category statutory ability to structure implementation, however, can be compared with shortterm constraints and resources of the subsystem actors. The institutional arrangement (form of selection of the bureaucracy) represents the membrane of separation between those inside and outside the community that decides public policy. Finally, the variable stages of the implementation process has similarities with the ACF public policy subsystem. Both are influenced by external variables; that is, they are dependent variables. However, in the case of the ACF, the subsystem is foreshadowed by the short-term constraints and resources of the actors in the subsystem, as well as exerting influence through the outputs and impacts of public policy.
In Sabatier and Mazmanian's (1980) work, although there is concern about how behavior, theories, and technologies influence actors, concepts like advocacy coalition and policy-oriented learning are not yet present. These concepts, which will provide more analytical consistency to the model, will only be absorbed by the public policy subsystem category in the ACF because although the implementation process stages category keeps a procedural form as well as the public policy subsystem, it still does not address the clash of coalitions arising from their belief systems. Notably, Sabatier and Mazmanian's (1980) project was strongly influenced by public administration and the concern with public policies' efficiency, and not necessarily with the analytical perspective formulated about 10 years later. This is clear because of the attempt to align institutional design with the action of political actors without resorting to a generalist and reductionist perspective while presenting an analytical tool to provide minimum predictability based on empirical evidence. Thus, the model laid out in Sabatier and Mazmanian's (1980) work reflects a paradigmatic framework for understanding the initial steps toward ACF.
The version of the ACF presented in Jenkins-Smith and Sabatier's (1993a) work is the result of contact between the two as early as 1985, when Professor Jenkins-Smith visited the University of California, Davis, where Professor Sabatier was teaching. At the time, they identified similar conceptions about the role of scientific information and public policy analysis in the policy process (Jenkins-Smith & Sabatier, 1993, p. 12). Then, they formulated the first version of the representative diagram of the model, as can be seen in Figure 3.   Sabatier and Jenkins-Smith (1993) arrived at the design presented in Figure 3 after two decades of studying environmental public policy in California. After years of research, they verified the following relevant aspects regarding changes in public policy: importance of scientific and technical debate in the understanding of the role of public policy; decentralization in the decision-making process of public policy that has moved from the US Congress to local authorities; formation of iron triangles around beliefs and values 5 ; and influence of external physical-environmental events, such as climate change and air pollution. This multiplicity of factors, according to Sabatier (1993, p. 20), validates Heclo's perspective (1974), which focuses on the learning of a given public policy developed by political elites within the professional community to understand the ongoing changes in public policy over the long term.
Supported by this work and in contrast to the model initially developed by Sabatier and Mazmanian (1980), the authors present four basic premises: an analysis perspective of 10 years or more is required to understand the process of public policy change and the knowledge acquired by the public policy community over that time; focusing on the policy subsystem is the most feasible way to understand policy change; intergovernmental dimensions and all levels of government must be considered for understanding the policy subsystem; and public policies can be considered belief systems consisting of a set of values and assumptions about realizing them. In their later work, Sabatier and Jenkins-Smith (1999) added one more assumption: the importance of the role of arguments and technical information about the magnitude and cause of problems, as well as the likely impacts of the various solutions presented.
In the debate about the motivation of the actors, Sabatier (1993) notably presents differences regarding the unit of analysis when compared to Heclo (1974) although both emphasize coalitions based on the dynamics of public policy-oriented learning. Heclo (1974, pp. 308-309) focused on formally organized groups and individual pursuit of self-interest.
Recently, Jenkins-Smith et al. (2018) complemented this premise by pointing out the interference of technical and scientific information in the subsystem because it is linked to the actors' belief system. In this work, the authors make a further assumption: individuals are boundedly rational, motivated by beliefs and values, as well as being relatively goal-driven but unclear about the way to achieve them. Sabatier (1993) starts from the idea that the actors in the subsystem are not restricted to formal groups, such as organizations and parties, and have bounded rationality. Further, based on the diagram shown in Figure 3, Sabatier (1993, p. 20) seeks to differentiate the more stable external factors affecting public policy from the more dynamic ones, as worked out in the comparison with Sabatier and Mazmanian (1980). Thus, in terms of their stability and ability to influence over time, the two sets of external variables are categorized as relatively stable parameters and dynamic external events. 5 Kingdon (1995) used the concept of "Iron Triangle" to describe the coalition formed by an interest group in association with members of the executive bureaucracy and representatives of the legislature. The concept was probably first used by journalist Ralph Pulitzer in 1919, a time of rising elitist thought, the dominant theoretical current in political science until the mid-twentieth century, whose content affirmed the concentration of political power in the hands of a minority, a political elite-an inevitable aspect in all societies. The critique of elite theory was formulated and deepened by pluralism, especially based on the seminal work of Dahl (1961), who argued that power in Western industrialized societies is widely distributed among different groups. No group is without power to influence the decision-making process and nor is any group dominant (cf. Grynspan, 1994). A genealogy of the ACF would have this discussion as its starting point.
Relatively stable parameters are more difficult to change in the short term and are less susceptible to the action of coalitions and actors. Environmental policy, for example, is significantly affected by air quality, air currents, and the political territorial boundary, which is relatively stable over time (Sabatier, 1993, p. 17). Within this dimension of external influence are the attributes of the problem, or how the disputed good regulated by public policy is characterized. Sabatier (1993) exemplifies the concern with air as an exhaustible commodity that needs regulation to maintain access. Fundamental cultural values and social structure can be evidenced with the different views on the state's participation in the economy if the social structure is sectarian in terms of race or religion, among other possibilities. Therefore, the relevance of social structure and context in coalition formation departs from any exclusively individual-based analytical level.
In the 1993 model, the basic constitutional structure refers to the normative-legal dimension of the state. External system events, however, are a constant challenge for the actors in the public policy subsystem. Actors must learn to anticipate and respond to these sudden changes in ways consistent with their beliefs and interests (Sabatier, 1993, p. 22). Systemic governing coalitions correspond to changes in the composition of Congress and the orientation of the executive. As an empirical example of this category, the authors discuss the change in the composition of the US Congress -from a majority of liberal Democrats to that of conservative Republicans -and how this condition affected the redirection of government priorities in environmental policy. The political decisions and impacts arising from other subsystems would be linked to the fact that the subsystems are partially autonomous and influenced by decisions in other areas. Sabatier (1993) explains this intersection between subsystems by highlighting the implications of British foreign policy in observing certain levels of pollution as a situation arising from its insertion in the European Economic Community.
The categories constraints and resources of the actors in the subsystem act as a bulkhead, a filter, to external pressures (relative stability parameters and dynamic system of events). Thus, external conditions, such as the country's socioeconomic structure and constitutional design, generate the set of resources, opportunities, and difficulties within the context of a specific public policy. However, it is up to the actors within the public policy subsystem to seize the moment to direct public policy in a way that is consistent with their beliefs. What is interesting about this category is that it considers an intermediate context between the external environment and the public policy subsystem itself (Sabatier, 1993;Sabatier & Jenkins-Smith, 1999). Notably, in the subsequent versions of the model, this is the point that undergoes the most changes, as can be seen in Box 1. Moving from their 1993 work, Sabatier and Jenkins-Smith (1999) add the degree of consensus needed for substantial changes in public policy as an intermediate variable between relative stability parameters and the public policy subsystem. They also added variable public opinions in the events of the dynamic external subsystem. The steps that make up the public policy subsystem also show Continue changes. Decisions according to sovereign institutional rules about regulation, resource allocation, and commitments present in the first version undergo two modifications: they are simplified to decisions by government authority and make institutional rules, resource allocation, and commitments a new stage of the subsystem. As seen in Sabatier and Jenkins-Smith (1999), the same intermediate variable between the external factors and the public policy subsystem undergoes further changes, as can be seen in the synthesis of the third version of the model (Box 2), presented in Sabatier and Weible (2007, p. 202). We added the category opportunity structures for long-term coalition, which absorbed the previously separate variable levels of consensus for public policy change and added the category openness of the political system.

BOX 1 COMPARISON OF THE FIRST AND SECOND VERSIONS OF THE ADVOCACY COALITION FRAMEWORK (ACF)
As Sabatier and Weible (2007, p. 200) note, these two categories were incorporated based on the work of Lijphart (1999) and inserted after observing the European corporatist context, present in the research of Larsen, Vrangbaek, and Traulsen (2006 2. Opening of the political subsystem.

Dependent a. Policy Subsystem
As shown in Box 2, when we compare the diagrams presented by Sabatier and Weible (2007)

POLITICAL SUBSYSTEM IN THE ADVOCACY COALITION FRAMEWORK: MODEL ASSUMPTIONS AND ADJACENT CONCEPTS
The public policy subsystem is the basic dimension for employing the ACF to look at the formations and interactions among actors within the decision-making process, as well as their responses to external and intermediate variables. It contributes decisively to the understanding about the changes in public policy and the characterization of advocacy coalitions. Regarding the category of advocacy coalitions, Sabatier (1993) proposes an approach distinct from the implementation steps present in Sabatier and Mazmanian (1980). In contexts marked by a tendency toward a specialization of technical and scientific knowledge, it becomes necessary to incorporate other actors into the public policy community, such as scientists, and mass media professional, such as journalists. Therefore, Sabatier (1993, p. 24) defines the subsystem as a set of actors dealing with certain problems concerning a public policy, without disregarding other potential participants in coalitions beyond the level of individual analysis. In these terms, it considers collective actors-such as private or public organizations, parties, and associations-also as a unit of analysis in the subsystem. 7 The subsystem is circumscribed to a thematically specific public policy with a defined territorial scope. The origin of the subsystem is the dissatisfaction or the issue that a group or a coalition concomitant to the subspecializations developed throughout the professionalization presents concerning the progress of the public actions aimed at a topic that affects a community located in space and time. Regardless of the origin, subsystems contain a huge and diverse set of actors, which should be aggregated into groups to reduce their quantity and make the analysis feasible. The advocacy coalition category can be defined as a collection of people from various positions who share a belief system and appear to have a non-trivial degree of coordination of activities over time. This conception can be an alternative to the difficulty in ascertaining the variations in the behavior of actors in the same institution, such as Congress, courts of justice, and even agencies (Sabatier, 1993, p. 25). In these studies, analyses focus on a range of topics, from the structure and stability of actors' belief systems to its formation and maintenance over time (Jenkins-Smith et al., 2018, p. 154).
The grouping of actors around a coalition is accomplished through the convergence of beliefs. However, there may be actors who do not conform to one coalition or the other, such as consultants, researchers, and bureaucrats, who participate only to offer their technical skills. 8 However, there is a category of policy broker, a function usually occupied by politicians elected to the Executive Branch whose concern is to ensure minimum limits to the competition among coalitions to find solutions to the problems faced by public policy. The distinction between participating in the advocacy coalition or being a policy broker lies in the outcome of the public policy itself (Sabatier, 1993). The model indicates that institutional affiliations tend toward the policy broker, while actors formally outside public power tend to participate in coalitions (Sabatier, 1993). Finally, the concept of advocacy coalition admits that shared beliefs are the link between actors and that core beliefs are significantly resistant to change.
The alignment of these elements confirms the first ACF hypothesis: when the core belief of public policy is in dispute within a subsystem, the formation of allies and opponents tends to be fairly stable over the period of a decade or more (HP1). With this, the ACF rejects the argument that actors are guided primarily by short-term interests and coalitions of convenience. Sabatier (1993, p. 27) mentions several empirical works that reaffirm such a hypothesis. Interests and actions are determined by a causal relationship. A certain actor intended to achieve a deliberate end or interest and acted in a certain way -that is, used a certain means to achieve their purpose. To arrive at this causal relationship, between means and ends, a parameter of beliefs about the causality of actions is needed. Soon, beliefs absorb self-interest and purpose. The reverse is more limited. Furthermore, it is very difficult to certify, with significant reliability, interest that is not restricted to economic activity, not quantifiable by a utilitarian logic, and not resulting from a set of values and beliefs (cf. Sabatier, 1993, p. 27). Therefore, beliefs and interests should not be understood as orthogonal concepts because they are complementary although beliefs are more feasible to identify than interests.
To understand a set of beliefs without necessarily resorting to relativism or reductionism, Sabatier (1993, p. 29) notes that the belief system has three basic starting points: Ajzen and Fischbein's (1980) theory of rational action, in which, for the authors, actions are derived from a range of possibilities to achieve a combination of goals in which actors' preferences are beyond a simplistic utilitarian view; rationality is bounded rather than perfect; and because the subsystem is composed of a select group within public policy, there are substantial grounds that confirm that many actors will have an internally consistent and relatively complex belief system in their areas of interest.
With rationality removed as the main indicator for the actors' actions, the question remains as to how a belief system can be explained. In this sense, Sabatier (1993) dialogues with the following authors: Putnam's (1976) review of the normative and cognitive orientation of political elites; Axelrod's (1976, as cited in Lakatos, 1971, distinguishing of the hard core from the elements of the scientific belief system; and Converse's (1964) understanding of abstract beliefs as more resistant to change than specific ones. Thus, the belief system adopted in the ACF pervades three structural categories: the hard core of normative foundations and ontological axioms that define the actor; the periphery of the hard-core beliefs, then called public policy beliefs (Sabatier & Weible, 2007), in which strategies and positions in public policy converge on the hard-core beliefs; and the secondary aspects, the instrumental decisions and the search for necessary information specific to public policy to implement the beliefs present in the periphery of the hard core.
Thus, the belief system meets a hierarchy and the propensity to change as it approaches secondary aspects. The beliefs coming from the hard core tend to be harder to change and more generic. Many coalition members will show agreement with various beliefs present in the hard core. However, the secondary beliefs depend on the area of public policy. Positions at a specific point in a public policy that interfere with the hard core will be more difficult to change than a position linked to secondary beliefs. Therefore, Sabatier (1993) presents two further hypotheses: actors in an advocacy coalition demonstrate more consensus on hard-core belief issues and less on the secondary aspects (HP2). Actors (or coalitions) will give up on the secondary aspects of their belief systems before recognizing weakness in the hard-core periphery beliefs (HP3). These two assumptions are important because they position the terms of negotiation on decision-making aspects. More recently, Jenkins-Smith et al. (2018, p. 154) added the following sub-hypotheses to hypothesis 3: in a coalition, administrative agencies generally advocate more moderate positions than their interest group allies (HPA) and actors in propositional groups are more limited in their expression of beliefs and policy positions than actors in material groups (HPB).
A central issue for the model that still requires further study concerns the free-rider problem of collective action. Sabatier and Weible (2007) discuss this issue through three perspectives: the costs of participation in the coalition are relatively low compared to other forms of collective behavior because the shared belief system generates high trust and goodwill to fairly distribute costs; the referred benefits of participation in a coalition are exaggerated, especially when public policy participants experience devil shift 9 in high-conflict situations; and the level of coordination in the coalition ranges from strong to weak at the extremes. Weak coordination has lower costs, reducing the risks for the free rider.
In the dispute in the public policy subsystem, one coalition plays a more prominent role than another. Each tries to follow its beliefs; to do so, they use strategies and resources provided by the relative stability and event parameters of the external dynamical system (Jenkins-Smith et al., 2018, p. 154;Sabatier, 1993). Sabatier (1993) then presents two more hypotheses concerning the learning movement of coalitions and external influences: the core (basic attributes) of the government program is unlikely to be revised as long as the main coalition remains in power (HP4) and the core (basic attributes) of the government program is unlikely to be modified without significant disruption external to the subsystem (HP5). With these hypotheses, it is possible to affirm that less expressive coalitions can exercise more permeability in public policy, but they will have to rely on significant external events to enhance their strategies. Similarly, public policy-oriented knowledge is more likely to change secondary aspects rather than hardcore public policy attributes.
Recent studies have attempted to refine the concept of advocacy coalition by operationalizing empirical studies around some gaps, such as a better understanding of the concept of policyoriented knowledge, inclusion of levels of conflict, attributes of actors, role of the policy broker, and nature of the stimuli and characteristics of decision forums. Such concerns are found in Weible et al. (2020), whose focus is on a better definition of advocacy coalition and a concern with understanding the dynamics of actors not only around the belief system but also the importance of coordinated action in public policy subsystems. These studies have defined five attributes of an advocacy coalition: the actors and their typologies; belief system; coordination of political activities; resources; and finally, stability. The ideal type of advocacy coalition meets all five attributes. The minimum condition for the existence of a coalition, on the other hand, is the sharing of beliefs. 9 This is a perception that places the antagonistic coalition in a Manichean position between "God and the devil" (Jenkins-Smith et al., 2018). Devil shift was first coined by Sabatier, Hunter, and McLaughlin (1987). The study starts with the premise that the rationality presented by the behavioral perspective becomes contestable when considering that actors disputing a certain outcome of a decisionmaking process compromise the efficient use of resources. In the context of information asymmetry, the difficulty in being able to accurately identify competitor values, objectives, and resources compromises the strategies, allocation of resources, and perception of the other. Thus, there is a risk of underestimating or overestimating the "enemy" as laid out in international relations literature. Given this issue, the central finding of the research supports the thesis that actors perceive opponents as stronger and meaner than they really are.

PUBLIC POLICY-ORIENTED KNOWLEDGE AND CORRESPONDING HYPOTHESES
After discussing the hypotheses and concepts dealing with the influence of external factors on actors in the formation of advocacy coalitions, we will address another fundamental concept of the ACF model: policy-oriented learning. Jenkins-Smith and Sabatier (1993b, p. 42) define it as enduring alternations of thought or behavioral intentions that result from experience and that are related to the attainment or revision of precepts in the belief system of individuals or collectives. It is the key concept for understanding the core of policy beliefs that guide actors subjected to constraints and opportunities arising from external events and opponents' actions. The knowledge resulting from exposure to external events and actions of opponents can cause a revision of the core of the belief system but have a more substantial impact on the modification of secondary aspects.
The process of subsystem formation, thus, begins with an attempt to solve a problem. If the formed coalition identifies the causes and gets political support, it proposes a public policy as a solution. If it is not opposed by another coalition, it is implemented. Otherwise, the coalition that challenges the public policy must seek resources and support for its alternative proposal. Both coalitions leave for what the authors call analytical debate, which is performed in various arenas/ forums and also affects the process. After the debate, if the result does not reach consensus, an agreement is sought. If there is agreement, both parties seek to expand their coalitions. If not, the government develops a research-intensive program of low coercion. Thus, the tension between the coalitions is eased and studies are conducted to resume the evaluation of the proposals (Jenkins-Smith & Sabatier, 1993b, p. 46).
Another relevant point concerns how actors identify and respond to changes in the antagonistic group's belief system, in the sense of even incorporating the beliefs against which they dispute. The example presented by Jenkins-Smith and Sabatier (1993a) refers to the changing position of economists critical of government spending in their attempt to control the emission of pollutants. Incentive programs and credit lines have been instituted for firms to reduce the emission of pollutants into the atmosphere. Such a decision did not affect economists' beliefs in economic efficiency and met the beliefs of the coalition concerned about the emission of pollutants. A transformation resulted from the incorporation of new aspects secondary to the public policy belief system within the analytical debate.
Committed to the intelligibility and study of ACF policy-oriented learning, Jenkins-Smith et al. (2018, p. 151) rearranged what Jenkins-Smith and Sabatier (1993, p. 42) called processes into the following basic questions: what components of belief systems changed or remained the same throughout the learning process? What contexts promote learning by coalition members? How is knowledge spread among allies and possible opponents?
To account for these perceived procedural issues, Sabatier and Jenkins-Smith (1993) listed the following explanatory dimensions: the level of conflict among coalitions reflects the degree to which basic beliefs among coalitions are incompatible. If the level is too low or too high, it tends to generate insufficient learning. If it is too low, coalitions empty the arenas and devote energy to other issues. If it is too high, the differences between each coalition's core public policy beliefs become more pronounced, making it more difficult to change positions. As for the intermediate levels, they tend to intensify the learning process directed toward public policy because there is sufficient motivation to consider or challenge the other party. Based on this concept, Sabatier and Jenkins-Smith (1993, p. 56) present the sixth ACF hypothesis: A policy-oriented learning scenario through belief systems is most feasible when there is an intermediate level of information conflict between two coalitions (HP6).
The other explanatory category includes analytical tractability. This category is somewhat confused in Jenkins-Smith and Sabatier (1993b, p. 56), but it refers to the ability of the analytical treatment of a particular public policy issue. Jenkins-Smith et al. (2018, p. 151) updated the concept to stimulus attributes. These are policies in which the theme is ambiguous without much technical-scientific basis. Therefore, a low level in public policy-oriented learning is expected.
From this point, Jenkins-Smith and Sabatier (1993b, p. 52) raised two more hypotheses: • Problems that accept quantitative data and theory as the basis for their solution are more prone to public policy-oriented learning through belief systems than those in which data and theory are generally qualitative, rather subjective, or completely absent (HP7) • Problems concerning natural systems are more likely to be learned than those involving purely social or political systems because, in some cases, they are more susceptible to arguments based on the classical scientific method of the natural and biological sciences (HP8).
Further, in relation to the analysis categories present in the model, it is important to highlight the number of people; who participates; and the forums, arenas, or environments in which the interactions among coalitions occur. Regarding this point, Jenkins-Smith and Sabatier (1993b, p. 54) present another hypothesis: • Policy-oriented learning through belief systems is more likely when there is a forum that is dominated by procedural norms and has enough prestige to force professionals from different coalitions to participate (HP9).
To continue updating the model, Jenkins-Smith et al. (2018, p. 151) supplemented with the following question: what role, if any, does a policy broker play in facilitating learning in the face of opponents? This question resulted in one more explanatory category: the actors' attributes, which are belief systems, resources, strategies, and network-that is, how does an actor, preferably the policy broker, engage in the process of policy-oriented learning? According to Ingold and Varone (2012), brokers can facilitate learning among opponents although there are no predefined criteria about their affiliation or position. The authors thus added one last hypothesis to the model: • Even when the accumulation of technical information does not change the perspective of the opposing coalition, it can have important impacts on public policy-at least in the short term-by changing the perspectives of the policy brokers (HP10).

FINAL CONSIDERATIONS
In the review and analysis of the ACF construction trajectory, it was possible to identify four versions of the model throughout almost 30 years, espoused by six prominent authors. On comparing the first version of the model presented by Sabatier and Jenkins-Smith (1993) and the one developed by Sabatier and Jenkins-Smith (1999), the model gained greater analytical capacity with the intermediate variables between the external factors and the policy subsystem; the categories internal to the subsystem were analytically improved by modifying the concept of sovereign rules to rules defined by government authorities and separating it from other categories arising from the characterization of the subsystem, such as institutional rules and resource allocation. This modification was important for the incorporation of the policy broker concept. Second, on comparing the third version of the model by Sabatier and Weible (2007) with the second version by Sabatier and Jenkins-Smith (1999), we found that the main modification occurred with the contribution of Larsen, Vrangbaek, and Traulsen (2006) in the incorporation of the category of openness of the political system in countries where corporatism is predominant. In doing so, the model has increased its applicability in contexts whose political structure is dissimilar to the pattern found in Western democracies. The fourth version, presented by Jenkins-Smith et al. (2018), is distinct from Sabatier and Weible's (2007) third version with the addition of the category of cross-cutting cleavage in the intermediate variable opportunity structures for long-term coalitions. It also removes the policy broker from the subsystem and leaves only the institutional rules as a direct result of the government authority's decision.
Throughout these transformations in the ACF diagram, it is worth highlighting the main attempts to respond to its limitations according to its three theoretical approaches: discussion about public policy changes, conceptualization and identification of advocacy coalitions, and the development of public policy-oriented learning. As Jenkins-Smith (2018, p. 144) notes, the study of public policy change, based on the ACF, has been the subject of many empirical studies. What is most striking about these works is their focus on explaining how and why some public policies change more than others. According to the author, while the original version of the model used two hypotheses to answer these questions -disturbances external to the public policy subsystems and changes in the power structure -the empirical studies have highlighted the combination of external and internal disturbances to the subsystem, public policy-oriented learning, and negotiated agreements as attempts to explain the changes or lack of changes of public policies (Jenkins-Smith et al., 2018, p. 147). Moreover, studies on such changes use different methods of data collection and analysis and adopt slightly different definitions of public policy (Jenkins-Smith et al., 2018, p. 147).
Policy-oriented learning is defined as a change in perception and behavior toward the solution of a public problem resulting from the revision of the belief system of individuals or groups. It is related to changes in public policy and affects the process of actors' membership in advocacy coalitions. However, even while contributing to the originality of the ACF, policy-oriented learning is the least consistent theoretical approach because of its subjective and difficult-to-measure character. Public policies with an intermediate level of conflict and with available technical and scientific information tend to result in studies with greater potential for ACF applicability.
Studies focused on identifying and typifying advocacy coalitions, however, tend to be more widespread. Publications by the forerunners of the model are moving in this direction, as can be seen in Weible and Ingold (2018) and Weible et al. (2020). This does not mean that such an approach is the simplest. Jenkins-Smith et al. (2018) have shown how imprecise and limited the definition of the concept of advocacy coalition continues to be. In this regard, Weible et al. (2020, p. 8) propose adding attributes to the original conceptualization of advocacy coalition. After all, this approach, in its limitations and imprecisions, has the greatest potential for operationalization for empirical research because it is substantially descriptive when compared to the more explanatory character attributed to the categories changes in public policy and policy-oriented knowledge.
Finally, as a gap in the literature that can be addressed in future research, the influence of international actors and its implications on public policy modifications are not explicitly addressed by the ACF, as highlighted by Jenkins-Smith et al. (2018). However, in an international environment marked by complex interdependence, considering the external context will certainly make contributions to public policy analysis. After all, it is necessary to question how international organizations and global networks of actors interfere in the subsystem of a public policy and which contextualization variables are more sensitive to the external environment. Perhaps these issues are not so important to discuss in US studies and experiences because the US influences the public policies of other countries more than it is influenced by them. However, for countries in secondary positions in the international system, this discussion is likely to be pertinent and promising.