Practicing policy learning during creeping crises: key principles and considerations from the COVID-19 crisis

Abstract Policy learning plays a critical role in crisis policymaking. Adequate learning can lead to effective crisis responses, while misdirected learning can derail policymaking and lead to policy fiascos, potentially with devastating effects. However, creeping crises such as the recent COVID-19 pandemic pose significant challenges for doing “good” policy learning. Such crises pose persistent threats to societal values or life‐sustaining systems. They evolve across time and space while stirring significant political and societal tensions. Given their inherent features, they are often insufficiently addressed by policymakers. Taking the COVID-19 crisis as an illustrative example, this article aims to draw practitioners’ attention to key features of creeping crises and explains how such crises can undermine critical policy learning processes. It then discusses the need for “policy learning governance” as an approach to design, administer and manage crisis policy learning processes that are able to respond to continuous crisis evolutions. In doing so, it helps practitioners engage in adaptive and agile policy learning processes toward more effective learning by introducing four key principles of policy learning governance during creeping crises. Those are: identifying optimum learning modes and types, learning across disciplines, learning across space, and learning across time. Practical tools distilled from emerging research are then introduced to help apply the proposed principles of policy learning governance during future crises.


Introduction
Policy problems come in different shapes, sizes, and forms.To address them, policy actors often engage in increasingly systematic policy learning processes.Doing so entails the pursuit of knowledge and information about policy problems (Zaki, Wayenberg, and George 2022).This can be done by different means, for example, reflecting on past experiences, policy evaluations, scientific evidence, engaging with experts, or with stakeholders both from within the government or across society.This can help develop better understandings of policy issues and identify potential solutions (Dunlop and Radaelli 2013).Understanding these policy learning processes helps us decipher the fundamentals of policy behavior.Simply put, the way policy actors learn (i.e. who do they learn from?How? Where, when, and why do they learn?)explains their pursuit of certain policy decisions, the appropriateness of these decisions, and thus policy outcomes.
Policy learning is even more pressing within crisis contexts.This is especially during "creeping crises," where urgent action is continuously needed over long periods in response to ambiguous and continuously evolving problems.Such crises can be exceptionally hard to define and address.Thus, they evoke significant social and political tensions, where the public's tolerance for policy mistakes becomes even more limited.This is especially as various political interests create tensions around which objectives need to be realized.There, the quality of policy learning or doing "good policy learning" can enhance policy effectiveness, foster public trust, and compliance, while inadequate learning can cause significant policy failures and fiascos, sometimes costing lives and affecting livelihoods (Dunlop, James, et al. 2020;Nolte and Lindenmeier 2023;Cairney 2021).However, despite being sought after, practicing good policy learning within such "messy" crisis contexts is quite elusive.This is why emerging research recently called for generating insights into improving policy learning as a practice and maneuvering its common pitfalls, particularly within such crisis contexts (see Dunlop, James, et al. 2020).
Having said so, taking the COVID-19 pandemic as an illustrative example, this policy practice article has two main objectives.First is pointing practitioners' awareness to an emerging, disruptive, and relatively understudied context of policy learning, i.e. creeping crises (Boin, Ekengren, and Rhinard 2020).In doing so, it offers a deeper understanding of the key characteristics of such crises and illustrates how they can undermine effective crisis governance.This helps practitioners develop more nuanced assessments of future crises.The second objective is to help practitioners do "better policy learning" within creeping crisis contexts.This is through drawing on emerging crisis learning research to emphasize the need to undertake policy learning governance during such crises.i.e. proactively managing policy learning processes to cope with creeping crisis evolutions.To help do so, four interdependent key principles for learning governance during creeping crises are crafted: Identifying optimal learning modes and types, learning across disciplines, learning across time, and learning across space.Then, drawing on emerging research, a toolkit of practical learning governance interventions is synthesized across four key dimensions of policy learning: actors, information and Knowledge, systems and Structures, and policy issue formulation.

Setting the scene: what are creeping crises?
Policy learning processes are highly context sensitive.Thus, understanding the context where learning is to be practiced and overseen is critical to learning effectiveness.This section briefly elaborates on creeping crises as one of the most understudied crisis types and contexts for policy learning practice.
Crises come in different shapes and forms, hence naturally, there are several approaches to classify them.One approach can be by looking at the duration of a crisis event, and how fast a key crisis indicator develops (i.e.how it increases in intensity).Key crisis indicators vary from one sector to the other.For example, in cases of infectious diseases, the main indicator can be the spread of an illness.As explained by DeLeo et al. (2021) we can have crises with short duration and rapid accumulation (e.g.SARS), short Duration and gradual Accumulation (e.g.Ebola outbreaks), long Duration and gradual Accumulation (e.g.Avian influenza, MERS), or long duration and rapid Accumulation (COVID-19, Swine influenza).Another way of classifying crises comes from 't Hart and Boin (2001) where we look at the speed of crisis onset and closure.There, we have fastburning crises (fast onsetfast closure), long-shadow crises (fast onsetslow closure), cathartic crises (slow onset-fast closure), or creeping crises (slow onset-slow closure).
Our particular crisis strand of interest here is creeping crises.A creeping crisis can be defined as a "threat to widely shared societal values or life-sustaining systems that evolves over time and space, is foreshadowed by precursor events, subject to varying degrees of political and/or societal attention, and impartially or insufficiently addressed by authorities" (Boin, Ekengren, and Rhinard 2020).These crises are less frequently observed and researched compared to fast burning crises (e.g.natural disasters, major accidents, etc.).Those on the other hand are often defined as "serious threats to the basic structures or the basic values and norms of a system, which under time pressure and very uncertain circumstances necessitates making critical decisions" (see Rosenthal, Boin, and Comfort 2001;Saetren et al. 2023).
Creeping crises are relatively harder to detect, and address compared to fast burning ones.This difficulty can be attributed to several inherent features of creeping crises.First, they have slow onsets, with relatively low intensity, thus not invoking immediate or proactive action (Saetren et al. 2023).In other words, they can easily slip through cracks of policymaking and institutional agendas.However, being left to fester in the background, soon after, creeping crises can unpredictably crescendo into high intensity "full blown" problems, suddenly presenting evolved and large-scale challenges that require immediate attention (Beamish 2002;Boin, Ekengren, and Rhinard 2020).Creeping crises are highly multidimensional and transboundary, i.e. large in scale, cutting across different societal sectors and geographic regions (Zaki and George 2022).At such scale, they interact with different cultures, values, and demographics characteristics, producing risk perceptions and impacts.Temporally, they linger for extended periods, sometimes reaching years where they "simmer" or stay present within the policymaking context, with varying intensity cycles (alternating between high and low severity).Accordingly, creeping crises are "fuzzier" and more challenging to define, with their meanings changing over time (Boin, Ekengren, and Rhinard 2020).Thus, they spur social contestation and tensions, leading to politicization and conflict underlying crisis response, especially given the public's conflicting ideas of what the crisis is, its severity, and thus the severity of crisis response required (Zaki, Pattyn, and Wayenberg 2023).
Across different classification approaches, the COVID-19 pandemic is considered one of the century's most significant creeping crises (see Boin, Ekengren, and Rhinard 2020;Di Mascio, Natalini, and Cacciatore 2020;Nolte and Lindenmeier 2023).It created a global long-term situation for almost three years where policy actors struggled to cope with continuous crisis evolutions cutting across almost all policy domains with far reaching implications.Of course, notwithstanding the social and political tensions it has created.To apply the concept more clearly, Table 1 highlights how features of creeping crises manifested clearly within the COVID-19 case.
Given their complexity, and continuous evolution, creeping crises such as the COVID-19 pandemic require deliberate, and continuous policy learning processes (Lee, Hwang, and Moon 2020;Kim, Shin, and Kim 2023).There, focusing on agility is critical (Nolte and Lindenmeier 2023).However, these features put together make the practice of policy learning feel like playing a game of chess, against an invisible opponent who is always two steps ahead, while the room is on fire, and with everyone is yelling conflicting game-winning strategies.So, how can policy learning be better approached within such contexts?The next section proposes four key principles of policy learning during such crises .However, before doing so, it is important to highlight a few caveats that can help both practitioners and researchers assess crises more comprehensively.
First, it is necessary to acknowledge that perceptions, and thus classifications of crises are socially constructed.Perceptions of issues such as threat, urgency, and complexity of crises can vary from one social group to another.Thus, the same crisis can be experienced differently across different social and professional groups with varying risk perceptions and values (Boin, Ekengren, and Rhinard 2020;Dunlop and Radaelli 2013).For example, for certain groups such as frontline healthcare workers, COVID-19 can also be experienced as a series of discreet fast-burning sharp crises during high intensity waves of the pandemic (e.g.Zaki, Pattyn, and Wayenberg 2023).In the same vein, a single situation can be a crisis for a certain group, but an opportunity for another (see Rosenthal, Boin, and Comfort 2001).Hence, this article does not claim one crisis definition across the board.Rather, the COVID-19 pandemic is hereby discussed as a creeping crisis given that it is experienced as such at the analytical level of interest, which is of policy actors who handle the crisis at a more systemic, long term and strategic level (e.g.Nolte and Lindenmeier 2023;Zaki 2023). 1   Table 1.The COVID-19 pandemic as a creeping crisis.

Multidimensionality and meaning shifts
Crisis has significant economic, social, and political implications (Zaki and George 2022).The crisis manifests under different meanings as it evolves, from starting as a public health issue to one of freedoms, access to healthcare, education, economic subsidy models and more (Boin, Ekengren, and Rhinard 2020;Zaki and Wayenberg 2021).

Reoccurrence and outbursts
Crisis intensity indicators have significant variations (i.e.outbursts) where waves of infections rise and fall resulting in varying perceptions of crisis intensity and threat (DeLeo et al. 2021).Crisis suddenly evolves from the background to the foreground without attracting significant attention (Saetren et al. 2023).

Subjective and conflicting threat perceptions
The crisis' nature creates tensions between different social group preferences.It also facilitates political schisms and politicization arguing for varying approaches to tackling the pandemic ranging from laissez-faire to highly stringent policies (Bohle et al. 2022;Zaki, Pattyn, and Wayenberg 2023).
Second, and relatedly, there are no clear categorical thresholds in theory for when a crisis can be classified under this or that type (see DeLeo et al. 2021).Rather understanding a crisis is a reflexive process that needs to consider positions of different actors and the context where the crisis unfolds.Third, the above discussed features are not necessarily exclusive to this particular crisis type, rather they can be seen as part of a sliding scale.Several of the creeping crisis features can be found in other crisis types or general policy problems.For example, having issues that are difficult to define is also a feature of commonly studied "wicked problems" (Head and Alford 2015).Thus, this article highlights that these features manifest more intensely and comprehensively in creeping crises as opposed to other types of crises, rather than them being mutually exclusive.Now, with the creeping crisis setting explained, how can practitioners improve policy learning processes therein?In the next section, I draw on state of the art research on policy learning in general, and crisis learning in specific to emphasize the need for proactively managing and governing policy learning processes, i.e. "policy learning governance".This is as opposed to engaging in adhoc or stopgap (knee jerk) policy learning adjustments in response to crisis fluctuations.There, I introduce four key principles of robust policy learning governance: Identifying optimal policy learning modes and types, learning across disciplines, learning across time, and learning across space.To help realize these principles in practice, I then present a toolkit of practical learning governance interventions across key elements of the policy learning process.

Coping with creeping crises: policy learning governance
Research has frequently pointed out that policy learning processes are embedded within political, administrative, and social systems that influence how learning takes place (Zaki, Wayenberg, and George 2022).Policy actors invoke different learning heuristics (i.e.ways of thinking about, and using learning) depending on learning objectives.These learning objectives can vary depending on how the context looks like (e.g.how pressing and contested the issue or crisis is).Learning can be leveraged for political capital accumulation when problems are underpinned by high actor polarization yet are not very salient or pressing.On the other hand, learning can also be leveraged to deliver effective technical solutions when there is limited polarization (i.e.some consensus), but issues are highly pressing (e.g.Trein and Vagionaki 2022).Even when pursuing technical solutions, definitions and understandings of problems can still vary, leading to different learning approaches, and hence outcomes (e.g.Zaki, Pattyn, and Wayenberg 2023;Crow et al. 2023).Furthermore, learning can be used to solve technical policy problems, substantiate, or legitimize policy positions (Boswell 2008).These approaches to learning are not only shaped by external contexts.They also have implicit roots.Afterall, policy actors (both individual and institutional) have bounded rationality.They do not necessarily have unbiased assessments of policy problems and solutions.The way by which they approach learning and understand policy issues is affected by their individual biases, belief systems, fears, political interests, and reasoning paradigms (e.g.Nowlin 2021).They are also shaped by existing legacies which create biases toward certain approaches to understanding problems, issues often labeled under the concept of "path dependence" where simply put, previous action and legacies shape future options.
The different ways of approaching learning (and different learning objectives) translate into on-the-ground interventions that shape policy learning outcomes.Such interventions can be understood as processes of "policy learning governance".Ongoing processes by which policy actors strategize, design, and administer policy learning processes to achieve political or technical learning objectives within continuously evolving conditions.Examples of such interventions include deliberate selection of experts involved (e.g.Cairney 2021), forming new advisory structures or consolidating existing ones (e.g.Zaki and Wayenberg 2021), or establishing values and paradigms that drive learning, such as tolerance for experimentation and failure, transparency, or creating certain information flow pathways (e.g.Lee, Hwang, and Moon 2020;Kim, Shin, and Kim 2023), among several others.
These deliberate learning governance choices have significant implications for the outcomes of learning and can make the difference between optimal learning that effectively delivers technical or political objectives, or misdirected learning, which leads to policy failure.This has been repeatedly observed even surrounding relatively stable policy problems (ones that are relatively well-understood and do not significantly fluctuate over time).Within creeping crisis contexts, proactive and continuous policy learning governance is even more critical.Why?Because as discussed, such crises linger for extended periods, their meanings shift over time, and they present themselves with varying intensity cycles.Hence, they spur variations in levels of societal and political contestation, and fluctuating levels of support and compliance.Resultantly the knowledge needed to address the crisis also changes.As such, policy learning processes (as resources for policy analysis and policy formulation) need to cope with the crisis dynamism through continuous adjustments and more reflexivity (Auld et al. 2021).There, policy learning needs to be an even more dynamic process rather than a "single variable" or a one-time "check the box exercise" that exists separately from its surroundings.This approach comes in contrast to relatively more traditional forms of learning that occur through engaging conventional more static and unifocal subject matter expert advisory groups on well-defined issues which can indeed function relatively well in non-crisis or relatively more stable conditions (see the accounts by Dunlop 2017; Mavrot and Sager 2018). 2 While acknowledging the need for rational interventions, it is important to note here that an approach of learning governance acknowledges the aforementioned (and often implicit) bounded rationality and biases that can hamper learning.Real-life learning governance is neither apolitical, nor value free.While striving toward effective technical solutions during crises, actors bring in their own interests, political preferences, learning objectives and lessons to be learned (e.g.Nowlin 2021; Cairney 2021).While this approach cannot alleviate or nullify the influences of such biases, it aims, as much as possible, to proactively mitigate them (even if indirectly) through proposing deliberate governance interventions.
Having outlined the notion of learning governance and its importance, now we turn to how can policy learning governance be approached within a context of creeping crises.Here, four key principles are crafted.

Identifying optimal learning modes and types
At the most fundamental level, successful learning begins with correctly identifying and administering optimal learning modes.Research highlights four main modes by which learning interactions can take place.Each of these modes entails different stakeholders to be engaged and different institutional setups.According to Dunlop and Radaelli (2013), these four main learning modes are: Epistemic (from experts); hierarchal (through established structures and rules), bargaining-oriented (through interaction and negotiation), and reflexive (through continuous adjustment, socialization and multistakeholder dialogue).
The functionality (or appropriateness) of these learning modes is determined by high-low combinations two factors: certification of actors (i.e. is there a socially endorsed group to solve the problem?), and Problem tractability (i.e.high tractability means problem is easily understandable, and relatively non-technical, low tractability means a problem is technically complex and ambiguous).For example, if we have a problem of low tractability (high complexity and ambiguity), and high certification of actors (there is a specific group that are endorsed by society as problem solvers), then epistemic policy learning (i.e.learning from experts) becomes a highly functional learning mode (e.g.Zaki and George 2022).Conversely, if we have a problem with low tractability and low certification of actors, then reflexive learning becomes more functional.There, understandably, in the absence of clear solutions, a common problem view , and certified problem solvers, more open debates with various stakeholders can deepen problem understandings and contribute to solutions that enjoy nuance and consensus (e.g.Millar, Davidson, and White 2020;Auld et al. 2021), particularly within changing conditions (e.g.Zeitlin and Vanhercke 2018).Identifying optimal learning modes is critical for policy success.Research is rife with examples where the misidentification of learning modes derailed policymaking leading to policy failure.This can be due to the absence of deliberate approaches to learning (in other words letting learning processes just drift like a rudderless ship where "unsuitable" learning modes prevail) or misconstruing the two above mentioned scope conditions that drive learning mode selection.For example, within the context of the Brexit crisis, Dunlop, James, et al. (2020) find that while the crisis positioned epistemic and reflexive learning as optimum learning modes, this did not work out due to the absence of key scope conditions.This included lack of open attitudes and negotiation settings, fluid group identities, absence of a trust culture, and disputes over the legitimacy of expertise.It can also be due to politicization of the crisis where existing legacies or political rivalries compromise the implementation of a correctly selected learning mode.For example, this can be due to tampering with or influencing the functioning of expertise (e.g.Zaki and Wayenberg 2021;Cairney 2021).
While there are different modes of learning types, different types of learning also exist.Types of learning are concerned with the policy units or parts of a policy are being learned about.For example, instrumental and managerial learning are focused on refining our understanding of policy instruments, rules and calibrations or settings or how can we better design and implement them (as opposed to updating our understandings of core policy beliefs, objectives, and paradigms).For instance, within the COVID-19 crisis, this means decisions on the recommended number of social contacts, closure of schools, facilities, number of vaccinations, type and scale of economic subsidies and their schemes, etc. (e.g.Biegelbauer 2016;Zaki, Pattyn, and Wayenberg 2023).On the other hand, social learning is geared toward deepening our understanding of the social construction of policy issues and what they mean within the broader societal context, potentially paving the way for new policy objectives or even paradigms.This is a type of learning that is closely associated with reflexivity.For instance, developing a deeper understanding of how society perceived the COVID-19 pandemic?Is it only a public health problem?A threat to liberties?Healthcare investment?Should we try to suppress the pandemic or live with the pandemic?And so on.A third major type is political learning, there we aim to better understand, reach, and advocate politically acceptable, and as such increase their acceptance, both within the political system as well as across society (Trein and Vagionaki 2022).For example, better understandings of advocating our positions on providing economic subsidies to certain businesses, or support to parties to a military conflict beyond national borders, etc.There is also organizational learning, whereby public organizations develop better understandings of crises, and attempt to drive crisis governance (Lee, Hwang, and Moon 2020). 3 Engaging in different learning modes has implications for learning designs, especially concerning which groups to learn from, what is the key focus of learning, and how to administer and organize the learning process.For example, primarily learning from experts can be done through focused expert advisory groups (Cairney 2021), while reflexive learning can require setups more conducive to societal and public participation (Auld et al. 2021).Learning in the shadow of hierarchy for example, requires a relatively more focused engagement of the civil service, and tacit knowledge therein, while considering politico-administration traditions and norms (Zaki and George 2022).Organizational crisis learning requires focusing on cultivating organizational capacities, values, and tolerance for failure and experimentation (Lee, Hwang, and Moon 2020).The diversity of setups and pathways underlying these different modes and types of learning urge policymakers to deliberately identify and engage in optimal learning modes and types in light of the crisis at hand.
Given that learning modes and types require engaging with different groups, this takes us to the second principle, that is, learning across disciplines.

Learning across disciplines
The scale and multidimensional nature of creeping crises and their scale means they cut across different sectors and affect different facets of societal life (Boin, Ekengren, and Rhinard 2020).This makes them deceivingly hard to define.Under urgency pressures, policy actors can define the crisis at face value, in doing so overlooking its many other dimensions.Harkening back to the COVID-19 pandemic offers one of the clearest examples of doing so.At its onset, the crisis was novel, highly technical, and ambiguous.Thus, under the above-explained framework of four learning modes, epistemic policy learning, i.e. learning from experts, was optimal.Yet, "which experts should we learn from?" was a pressing question (Radaelli 2022).Initially, the crisis was perceived and defined as one almost exclusively of public health.Hence, policymakers practiced epistemic policy learning by focusing on a very narrow group of experts such as virologists, and other medical professionals.This meant other facets of the crisis were not represented through adequate expertise at the policy learning table.For example, expertise on the social, behavioral, or even bureaucratic implications of medically driven crisis responses (see Zaki and George 2022).Doing so overlooks the systemic nature of such creeping crises (El-Taliawi and Hartley 2021).This often led to sub-optimal learning outcomes and policy responses that fail to balance and reconcile competing priorities.Consequently, resulting in short-term solutions that contribute to depreciating compliance, and eventually undermining the effectiveness of crisis responses (see Zaki, Pattyn, and Wayenberg 2023).Misidentifying an optimal interdisciplinary mix can be due to difficulties in understanding the crisis, or even pre-established scientific inclinations and traditions (Jensen, Lynggaard, and Kluth 2022).It can also be due to the political interests and preferences underlying learning, where certain lessons become more favorable leading to "cherry-picking" experts who are likely to endorse certain preferences (Cairney 2021;Lee, Hwang, and Moon 2020).It can be also due to established institutional learning preferences (Jensen, Lynggaard, and Kluth 2022).
The principle of learning across disciplines is inherently concerned with ideas and knowledge, rather than just scientific expertise and epistemic policy learning.For example, when the adequate scope conditions exist, engaging in reflexive learning could entail sourcing knowledge, information, and ideas from different social groups of concern, or what is called "mini publics" (e.g.Sancino et al. 2021).Another example is when employing learning in the shadow of hierarchy, knowledge and ideas can be sourced from practitioners within the civil service, given their knowledge of citizen-state interactions, and the viability of policy solutions (e.g.Zaki and George 2022).Put together, this principle of learning governance urges thinking beyond oversimplified definitions, toward more nuanced interdisciplinary understandings of creeping crises.
The first two principles discussed have shed light on the "how" and the "who" of learning, the third principle focuses on the "Where" or the spatial dimension.

Learning across space
Large-scale crisis responses are often overseen by national governments, there the bulk of policy learning processes, and major crisis responses take place.However, in growingly decentralized governance systems, subnational governments also enjoy some leeway to interpret and apply national level policy directives.They also have their own competences, whereby they can engage in policy learning, and issue regional and local crisis response policies (Maggetti 2020;Zaki and Wayenberg 2023).So, during long-term creeping crises, policy learning processes can simultaneously take place across different levels of the governance systems: Supranational (e.g.Quaglia and Verdun 2023), national and subnational (Mattei and Del Pino 2021).This paints a vibrant image where learning happens everywhere.However, across different levels, learning is also not a monolith, i.e. naturally and by default, these learning processes can be heterogenous or even contradictory.This can be attributed to differences across jurisdictions within the governance architecture (differences between cities, provinces, and municipalities engaging in learning).These differences can include contextual and demographic properties such as regional economic conditions, population densities, political orientations, administrative capacities, access to expertise, among other factors (Di Giulio and Vecchi 2019).Resultantly, this creates varying interpretations of a single policy problem, varying priorities, and thus varying learning preferences that can lead to diverging responses to the same crisis (see for example Crow et al. 2023;Casula and Pazos-Vidal 2021;Busetti and Righettini 2023).For example, while at some jurisdictions governors or mayors can exclusively involve medical experts in making sense of national directives or formulating subnational policies, others can involve a broader interdisciplinary mix.A third group might not engage in any significant learning from experts altogether, preferring direct implementation of national directives or deliberation within the civil service.While this can create the space for tailored and contextualized policies, it can also lead to confusion, suboptimal learning, or even policy incoherence across regions (see the account by Zaki and Wayenberg 2023).
Here, while engaging in learning, research shows that policy actors need to consider that learning not only occurs at the national level, but also at different levels and units within a multilevel governance architecture.As such, there is a need to create overarching mechanisms that maintain coordination between learning processes at different levels and units while not stifling the space for the localization of learning.This can be for example through establishing institutional mechanisms to foster alignment between organizational and multilevel learning groups (Carter and May 2020), setting guidelines for the type of expertise that needs to be represented in learning processes across different units, and creating information flows and spaces for different learning groups to interact with each other throughout the crisis (Zaki and Wayenberg 2023).Now, we move to the last principle, specifically focusing on bringing the three discussed principles together by looking at the temporal dimension of the crisis.

Learning across time
As discussed above, creeping crises linger for extended periods.Over such durations, their inherent features (e.g.scale, multidimensionality, varying degrees of intensity), lead to the meanings and manifestations of a creeping crisis to vary over time.So, a learning type and mode that was considered optimal at the onset (let's say at t 0 ) might not still be optimal one year later (let's say at t 1 ).Let's look at how this happens in practice, also by reflecting on the case of the COVID-19 pandemic.Changes in crisis characteristics over time lead to changes in scope conditions.At the onset, the pandemic was largely an issue of low tractability, high actor certification (thus, epistemic policy learning was optimal).Under these conditions, the crisis pressure, limited amount of information, makes it relatively easy for government to certify undisputed expertise (Baekkeskov and € Oberg 2017).However, as more knowledge and information are generated and become publicly available, more actors find themselves certified to enter the debate, and problem tractability increases.Thus, expertise becomes disputed, and the "irony of epistemic policy learning" kicks-in (Dunlop 2017).A process by which expert authority is undermined and displaced by the knowledge they generated themselves over time (in other words, over time 'everyone becomes an expert').Additionally, contestation and push-back from political and societal actors can increase, particularly as the "rally-around-the flag effect" wanes.As the crisis takes on different meanings over time (e.g.becoming more economic, social, legal, and legislative in nature), other learning modes and types can become more optimal to cope with the evolved crisis contexts.For example, reflexive learning can replace epistemic learning as a primary learning mode (Auld et al. 2021).Also, other learning types can also be needed, for instance, social learning to facilitate deeper debates and understanding of the problem.Naturally, this requires different learning setups, for example reshuffling the range of experts involved, or involving a broader range of societal actors in the debate.Failing to anticipate adjust policy learning approaches, and maintaining defunct learning mode and type setups can lead to depreciated public compliance, and policy failure.Research shows that this can trigger changes in learning approaches.However, research also indicates that it is favorable to proactively adjust learning approaches and engage in learning mode/type transitions to cope with the dynamic nature of the crisis (Zaki, Pattyn, and Wayenberg 2023), even engaging in anticipatory or preemptive learning (e.g.Crow et al. 2023).
So, what can practitioners be on the lookout for as cues for learning mode transitions?Two things come to the fore.The first lays at the heart of the policy learning process, which is changes in policy issue formulation and its social construction, or definition.How the policy issue is defined determines its analytical features (i.e.issue tractability, and certification of actors).This in turn helps determine optimal learning modes (Dunlop and Radaelli 2013).Changes in such features should be assessed to discern whether different approaches to learning are needed.The second factor concerns one of the main factors that shape learning, which is the social and political context (Zaki, Wayenberg, and George 2022).Continuous assessments of the evolving social and political contexts within a creeping crisis can contribute important insights as to the functioning of existing policy learning processes and the suitability of learning modes.Polarization, political contestation, and issue salience assessments can help policy actors employ bargaining-oriented and reflexive learning to address the crisis at different points.
It is also necessary to highlight that learning over time is not only about coping with the dynamism of creeping crises, but also about identifying past cases from which lessons can be extracted.Research points out the importance of identifying past cases, especially those from similar contexts, whether successful as exemplary (Lundin, € Oberg, and Josefsson 2015), or unsuccessful as cautionary (Dunlop 2017).Best practices indicate that institutionalization of such lessons provides handy resources for swift and agile crisis responses (Lee, Hwang, and Moon 2020;Kim, Shin, and Kim 2023).This is in addition to continuous scanning of emerging lessons as a crisis unfolds.It is important to note however that the fungibility (i.e.applicability and fit) of lessons from past crises needs to be considered to ensure that only appropriate lessons are transferred and implemented (Greener, Powell, and King-Hill 2021).
Before concluding this principle, it is also important to point out that learning types and modes are not mutually exclusive.Rather, different modes and types can occur simultaneously.However, at a given point in time, some modes can be more prominent and pronounced given that they can provide more functional appropriateness at a particular phase of the crisis.For example, learning from experts can still take place at a secondary level while reflexive learning is primarily engaged.This is particularly as experts can serve different roles with varying contributions, ranging from being teachers, to facilitators, producers of standards, or contributors (See Dunlop and Radaelli 2013).
Having discussed the four principles of learning governance during creeping crises, the following section synthesizes an action-oriented learning governance toolkit drawing on state of the art COVID-19 policy learning literature.This helps equip practitioners and researchers alike with examples of learning governance resources.

Learning governance interventions: a toolkit of actions
To present this toolkit, it is first necessary to highlight the key elements of policy learning or 'what goes into a policy learning process?'This helps us establish what needs to be proactively managed or governed when practicing policy learning, particularly during creeping crises.
Policy learning can be understood as the process by which policy actors (individuals or organizations) deliberately seek and "consume" different types of information and knowledge (from different sources) regarding a certain policy problem.These policy actors exist within governance and management structures (across different levels, with rules, regulations, and administrative traditions) while continuously interacting with an evolving policymaking context with changing preferences, public opinions, tensions, etc. (Zaki, Wayenberg, and George 2022).
So, with that said, what is there to be governed in terms of policy learning?Policy actors have the room to proactively adjust combinations of key learning process elements to better achieve learning objectives.For example, by adjusting who participates in learning processes.They can also adjust policy learning structures or reformulate the rules and regulations that govern learning and engage in updated understandings and formulations of policy issues, which in turn reorients the view of what do we need to learn about.Furthermore, changes in actors, policy issue formulations, or learning rules can also lead to the reliance on different sources of knowledge and information to drive the learning process.Table 2 shows examples of such governance interventions across the different elements of the policy learning process drawing on the COVID-19 crisis experience.
Research and practice show that failing to engage in adequate policy learning governance can lead to learning failures, and hamper policy effectiveness (e.g.Dunlop 2017;Dunlop, Ongaro, et al. 2020).This can be even more problematic in cases of crises where policy learning contributes more directly to policymaking due to the limited time for reflection and the need to address complex and often ambiguous issues.On the other hand, there is also evidence that policy learning can be made more successful when policy actors proactively ensure that adequate learning and advisory groups are brought into the debate and institutional setups and rules ensure that they can work together effectively (e.g.Osei-Kojo et al. 2022).Before concluding this section, it is worth noting that the above table does not include an exhaustive list of all interventions possible.Rather, it offers a preliminary account of different actions that have been shown to be important for effective policy learning to take place throughout the COVID-19 crisis.Hence, practitioners and future researchers are encouraged to explore additional measures that could further enhance learning, especially while considering specific contextual factors (e.g.political administrative traditions, administrative capacities, population characteristics, etc.).

Concluding remarks: bringing it all together
Policy learning plays a critical role in crisis governance.However, the inherent features of creeping crises pose several challenges for the practice of "good" or "adequate" policy learning.This article drew on emerging research with a special

Policy issue formulation
Reframing policy issues in light of crisis evolution in a manner that reflects updated policy paradigms and objectives if needed (Zaki, Pattyn, and Wayenberg 2023).
Considering how the crisis is perceived across different regions and jurisdictions the implications this has for how policy learning takes place within different jurisdictions (Zaki and Wayenberg 2023;Crow et al. 2023).

Context
Engaging in continuous contextual scanning to ensure evolutions in policy issue formulation and political settings are accounted for (Raudla 2021).
Undertaking anticipatory or preemptive learning through forecasting changes in external contexts (Crow et al. 2023).
Engaging in quadruple loop learning, i.e. considering the continuous interactions between the external and internal learning environment of crisis governance organizations.This means continuously revising the relationship between backstage components (Time, targets, and crisis context), and front stage components (Assumptions, Actions, and results) (Lee, Hwang, and Moon 2020).
focus on lessons from the recent COVID-19 crisis to propose a novel way of approaching policy learning during creeping crises.It highlights the need for proactively and continuously managing policy learning processes to cope with crisis evolutions, i.e. policy learning governance.Four principles of effective policy learning governance are proposed: Identifying optimal learning modes and types, learning across disciplines, learning across space, and learning across time.Then, a set of practical interventions drawn from COVID-19 policy learning research are provided to help engage in said effective policy learning governance.While this approach argues for the importance of learning governance, it does not attach a normative outcome to it.Meaning, "more learning governance" or "more interventions" are not necessarily a recipe for success.Misguided interventions derail crisis governance, for example deliberately building learning structure susceptible to politicization and misidentification (Cairney 2021), identifying inadequate expertise (Zaki and Wayenberg 2021), or directing learning into the suboptimal modes (Dunlop, James, et al. 2020).There are also several tradeoffs that should be considered when practicing deliberate governance of learning processes.The crisis introduces several policymaking tensions, for example: the need for quick yet deliberate and well-considered policymaking, multilevel consideration of contexts (individual and collective), focused and expert driven policymaking while maintaining openness, representativeness, and transparency, among several others (Zaki and George 2022).Attempting to reconcile these tensions, while learning across time, space, and disciplines can also contribute to a form of paralysis by analysis.For example, while learning across disciplines, involving several strands of expertise requires diligent efforts to make sense of diverging (even contradictory evidence), also with the risk of information overloads.While learning across space, establishing an extensive number of advisory bodies can cause confusion, and blur lines of accountability, making it hard to develop coherent understandings of the crisis (Zaki and Wayenberg 2023).While learning across time, policy actors can be overwhelmed by a huge number of potential lessons from past crises or other jurisdictions, many of which are not necessarily "fungible" or transferable (Greener, Powell, and King-Hill 2021).
Hence, when practicing policy learning governance, practitioners are urged to consider approaches that balance a focus on coping with a dynamic crisis context and acknowledging the systemic limitations in the management of learning processes (e.g.Lee, Hwang, and Moon 2020;Osei-Kojo et al. 2022;Kim, Shin, and Kim 2023).This includes understanding existing value systems, organizational learning capacities, and politico-administrative contexts, among others.

Notes
1. Scholars can also have different views on crisis classification depending on their ontological positions and fields of interest.For example, while Boin, Ekengren, and Rhinard (2020), Saetren et al. (2023), andZaki (2023) view COVID-19 as a creeping crisis, Vince (2023) views plastic pollution as a creeping crisis when compared to COVID-19, with the pandemic in relative terms is considered more urgent.2. This is not to say that learning governance is only necessary during creeping crisis contexts and not others.Rather, that the dynamism and nature of creeping crises put more emphasis on the necessity of learning governance interventions.

Table 2 .
Examples of learning governance interventions.