As mentioned earlier, rest breaks are distinguished from other forms of intervals separating work periods in time (e.g., preparation, monitoring, etc.), and other type of activities that require the further maintenance of attention (Gillie and Broadbent, 1989; Goschke and Kuhl, 1993; Allport and Wylie, 2000). A characteristic of these aspects is that they be distinguished theoretically while the empirical test of prediction derived from theory depends on the quality of design and measurement precision. However, there are some related phenomena obtained in a paradigm similar to a rest-break situation (at least could be framed as such) but with an entirely different underlying mechanism. One example is the so-called incubation paradigm, where individuals are typically administered with a problem-solving task, with the solution being dependent on sudden insight that is often prevented due to fixation or when misdirected toward another pathway (e.g., Vul and Pashler, 2007; Bilalić et al., 2008; Sio and Ormerod, 2009). The incubation effect is demonstrated by comparing a condition where individuals are administered to take a rest, relative to an alternative (distracting) activity and a control (no-rest) condition. The outcome, relatively often shown, is that those participants (a) who took a rest during an incubation period performed better than (b) those who performed a task during this period, (c) relative to the baseline (no-rest) condition, though opposing findings have also been reported (Sio and Ormerod, 2009).

Another related line of research concerns studies on memory consolidation during nocturnal sleep, daytime napping, or other kinds of resting periods. The term memory consolidation refers to a category of processes considered to support the stabilizing of a memory trace after its initial acquisition. Importantly, psychological research is not so much concerned with low-level consolidation processes such as synaptic consolidation, which has clearly been shown to occur 1–2 h after initial learning. Instead, the focus is on cognitive consolidation, that is, on how mental representation are formed or restructured, and how episodic traces are finally transformed into abstract codes with flexible retrieval structures (Tulving and Thomson, 1973; Hintzman, 1974, 1986; Tulving and Watkins, 1975). Classically, memory consolidation is studied during sleep, though psychological research is more focused on shorter time scales, studying mnemonic consolidation after resting relative to an active-work condition (Wamsley, 2019; Martini and Sachse, 2020). The participants in such studies are typically required to memorize material, with recall performance being tested afterward at a later time point, under a relaxed condition (e.g., closed eyes, napping, etc.), relative to a distractor condition (e.g., mental work, watching videos, etc.). The theoretical prediction is that waking rest supports the consolidation of previously learned memory content, relative to an interference condition, though the empirical findings are rather mixed with this regard (Martini et al., 2019). While this line of research addresses the aspect of “resting” on memory consolidation, it is important to understand that the de facto research question in this field is not on how rest breaks restore attention, but more on the benefit of non-interference (vs. interference) on memory recall performance (Wamsley, 2019; Martini and Sachse, 2020).

By definition, the term restart costs refer to a cost point that is incurred when the time taken for rest is too long, and in this way, it could be viewed as a non-beneficial effect of rest on performance. It is directly connected to the question of how long a rest break should be to achieve the best results, which has been debated already by earlier studies (Graf, 1922; Manzer, 1927; Poffenberger, 1928; Schubert, 1932; Barmack, 1939; Ross and Bricker, 1951). In the literature on rest-break research, this is studied by experimentally varying the rest-break length and to determine the functional relationship between rest length and recovery. In the specific literature on task-switching performance, the term restart costs is not in the same way conceived of as a rest break, but taken in the more restrictive sense as the relative costs of intercalated time intervals in task repetition relative to alternation trials (Allport and Wylie, 2000; Wylie and Allport, 2000). In this way, re-start costs in task-switching research are linked to aspects of forgetting, or other aspects of losing proper memories for the upcoming task (Altmann, 2002; Altmann and Gray, 2008; Kiesel et al., 2010; Vandierendonck et al., 2010; Vallesi et al., 2013). We therefore argue that the conception of re-start costs are in its specific form (as used in the task-switching literature) different from those studied in the rest-break literature. While the rest-break literature deals with the aspect of attentional replenishment, the task-switching literature is concerned with aspects of forgetting, and in this way, re-start costs are not the primary concern but a side-show phenomenon in the task-switching literature.

The concept of rest breaks exhibits some remarkable similarity with the concept of an interruption, which refers to a temporal interruption of an activity that is not necessarily needed, and neither wished, nor intended, and not expected as such at a certain time point during ongoing task performance. In the early literature that has its starting point in Lewin’s (1928) field theory, it was shown that when individuals were interrupted during an ongoing task (stringing beads, solving puzzles, etc.), but were allowed to continue with other tasks, then the interrupted task was recalled more often than the uninterrupted ones (Zeigarnik, 1927). Even more so, if they were allowed to freely decide what do to at some point, the individuals tended to re-start and completing the interrupted tasks (Ovsiankina, 1928), which indicates that the memory for non-completed tasks tend to further persist in memory and thus guiding ongoing decisions (Gillie and Broadbent, 1989; Goschke and Kuhl, 1993; Einstein and McDaniel, 2005). Thus, the study of (completable) task interruptions like solving a puzzle is an interesting counterpoint to the study of rest breaks because it demonstrates the energizing effect of intrinsic motivation on cognitive persistence such that individuals are prevented to perceive a need for rest before completing a task such as a puzzle, or when they are given an objective (or purpose) to aim for Kruglanski et al. (2012), Suri et al. (2015), Krishna and Strack (2017), Steinborn et al. (2017).

While the theorizing in the rest–recovery model literature is focused on mechanisms of attentional replenishment, considering rest as to result in an improvement or at least stabilization of performance, the literature on delayed processing or unpredicted interruptions during ongoing action sometimes shows the opposite effect, as often a detrimental effect on performance is reported which is considered to originate from kinds of distraction. In fact, delays can severely interrupt workflow and may also result in affective responses, or emotional disturbances, such as increased distress, despite a measurable decrease in performance (Kohlisch and Kuhmann, 1997; Szameitat et al., 2009; Thomaschke and Haering, 2014). A crucial difference between the research on rest breaks and those on interruptions lies in the contextual semantics of situational prototypes where breaks or interruptions typically take place. By definition, rest breaks are studied in situation where rest is naturally indicated such as to counteract the time-on-task effect in sustained-attention and vigilance tasks. In contrast, interruptions are meant to disrupt the workflow and are typically infused in situations where they are unwanted. While people can clearly distinguish between both in everyday situations, it is difficult to determine the contextual semantics of everyday concepts in the artificial situation of a laboratory experiment. For example, the loading bar in computer games would unlikely be conceived of as a rest break, but when one attempts to study this situation in the laboratory experiment, it becomes difficult to distinguish (Suchotzki et al., 2017; Suchotzki and Gamer, 2018).

The energetic capacity concept is one of the central elements in the research field of sustained attention and vigilance (Jennings and van der Molen, 2005; Langner and Eickhoff, 2013). The critical variable is the volume, which is decreased by cognitive work and recharged by breaks. In contrast to the computational concept of resources in dual-task research, the central issue here is not the allocation of a flexible resource to multiple forms of concurrent tasks, nor is it about the allocation of separate pools of resources to tasks with a need for a specific type of resources. Rather, the central issue revolves around the distribution of the capacity that can be provided by the resource over the time during which mental work is performed. In this respect, the energetic resource model rather corresponds to the metaphor of everyday language use (Greenwald et al., 1986; Kahneman, 2013, ch. 2): during intense mental work, the reservoir of mental resources is reduced or emptied, and it is recharged during breaks. The empirical indicator for the state of resource disposal is the decrease in performance in the task over time, measured with the respective characteristic values. The metaphor of an energy reservoir that is discharged and recharged by work or breaks is the core element of almost all theoretical approaches in this area. Differences often relate to the dynamics of these processes, the nature of the effect variables, and the weighting of additional variables such as self-reported motivation or psychophysiological parameters (Langner et al., 2010; Lim and Kwok, 2016; Steinborn and Huestegge, 2016).

Kaplan’s (1995) attention-restoration theory is a highly regarded theoretical approach that (in a close definition) relates to long-term recovery. In a strict sense, it is a psycho-sociological theory that addresses the dynamics of recovery and stress in the context of the real world of work and leisure. An outstanding feature of this theory, relative to other accounts theorizing on rest and recovery, is that it holds clear implications for urban planning (Berman et al., 2008; Atchley et al., 2012; Anguluri and Narayanan, 2017; Pasanen et al., 2018). At its core, the model states that energetic resources are claimed by goal-directed work and restored in free time (during non-goal-directed activity). Yet, recovery is not only a function of spare time, as the model makes a distinction regarding the context where rest takes place. In other words, spending time in nature is assumed to be more beneficial than spending time in (typical) urban environments (but see Ouellette et al., 2005). According to this idea, natural environments are rich in what they term “gentle fascinations,” that is, sensations that can be processed with effortless attention that is automatically directed by stimuli and not tied to goals (Kaplan, 1995; Berman et al., 2008; Kaplan and Berman, 2010). These include, for example, clouds moving across the sky, leaves rustling in a breeze, or water flowing over rocks in a stream. Critical to this approach is the theoretical distinction between reflective and automatic attentional control (cf. Strack and Deutsch, 2004; Krishna and Strack, 2017), albeit related to a more sociologically relevant context. There is also empirical support for his proposal showing that population satisfaction is higher in urbanized areas with green walkable parks than in areas without them, even when relevant variables are statistically controlled (cf. von Lindern et al., 2017, for an overview).

In the perspective of their analytical framework, Kaplan and Kaplan (1989) have identified four key principles of recuperation that can be generalized to the field of break research in general, (1) change of scenery, (2) conceptual distance, (3) fascination, (4) and extent of attracting involuntary engagement. Additionally, one more principle can – but rather indirectly – be extracted, (5) the slowing of pace (Levine, 1998; Hoffmann et al., 2021). In terms of break effects, it can be said that people recover well when they make a change from the work task to the break activity, when it offers sufficient distance from the work activity, when the break activity is beautiful or fascinating, and when it engages attention in a non-goal directed way. Finally, it is especially restful when time pressure is removed. Because these principles have high face validity, they have very often served as the basis for predicting rest-break effects in experimental studies. Despite this inspirational power, some methodological problems arise here. For example, it is unclear to what extent predictions can be transferred from large to small time domains. Related to this, many of the operationalizations chosen in “experimental” settings are questionable because of their reductionistic approach. The fundamental question here is whether reaction-time based performance is appropriate to test these principles (Miller and Ulrich, 2013). Therefore, the inspirational power of the theory consists mainly of its high generality, intuitive plausibility, and the logical consistency of its basic arguments (cf. Greenwald et al., 1986; Greenwald, 2012; Gray, 2017).

The recreational function of temporary slowing of behavior was already recognized by early authors, and is of high relevance in the current theorizing. For example, Bills (1943) already made the argument that during continuous speeded activity, the cognitive system can only operate at a maximum level of performance for a very limited time, until it experiences a transient depletion of the required resource pool, which then manifests itself in mental blocking. Bills (1943) argued that this phenomenon could be conceived of a type of pause enforced by the cognitive system, similar to the pause taken on a scheduled basis (cf. Jersild, 1926; Bertelson and Joffe, 1963). The finding that the frequency of exceptionally long responses increases after a prolonged period of uninterrupted processing (while there are typically no changes in the fastest responses) has often been used as evidence for this cognitive-energetic view (Sanders, 1998, ch. 9). Bertelson and Joffe (1963) generalized this principle of alternation of strain and recovery in continuous activities, arguing that performance in self-paced continuous tasks is heavily affected by the ability of an individual to regulate speed and accuracy for the purposeful completion of a task in such a way that it can be performed with optimal efficiency (Neubauer and Knorr, 1998; Stahl and Rammsayer, 2007; Steinborn et al., 2018). Following this view, numerous authors conceptualized “energetic regulation” a basic ability to attain and maintain a state of general optimal activation for upcoming demands, to set an optimal rhythm, and to maintain this rhythm over the duration of the demand.

One important aspect of research on rest concerns the rate of recovery from mental activity that takes place during breaks. Connected with this point, it is crucial to determine whether recovery potentially occurs (in a non-registered manner) during the active-task period itself. The early observation by Bills (1931) that periods of “enforced” rest take place during the task itself (and against the instruction to perform best) gives a clear indication that the answer is likely “yes.” However, such an observation challenges the logic underlying a straightforward version of an energetic–capacity model, because differences between tasks or individual differences could be interpreted as differences in the employment of strategies (Pashler, 1998; Sanders, 1998). The question that follows is to what degree individuals have strategic control over the available resources to perform a task (Inzlicht et al., 2014, 2018). For example, individuals seem to be able to either delay (Kunde et al., 2004; Jentzsch and Leuthold, 2006) or to speed-up responses (Strayer and Kramer, 1994; Kleinsorge, 2001; Steinborn et al., 2017) when prompted to do so by proper cues, indicating that there is some flexibility in changing strategy. The strategic-resource idea emphasizes the allocation policy of resources, which refers to the basic principles underlying the distribution of capacity over the activity period. In other words, if one knows that there is a long way to go, one more likely allocates resources in a different way than if the way is expectedly short. Yet, it is difficult to exactly determine the strategies used in a task because there are not only differences between experimental conditions but also differences between individuals and, even more so, individuals might change their strategy over the task in a rather qualitative way, which complicates a straightforward interpretation of performance effects in the light of theory (Vandierendonck, 2017).

In fact, there is psychophysiological evidence for such a demand-adaptive mobilization of (physiological) capacity. A typical finding is that the courses of cardiovascular parameters are completely different in sprinters and long-distance runners, with the former showing an increase in heart rate from about 90 to 120 beats per minute immediately before the start of the race, while the latter showed no or only minor changes (e.g., Faulkner, 1964; Hilton, 1975; Inui, 1987; Baden et al., 2005). This is a remarkable finding in that this energetic mobilization occurs before the actual demand, implying anticipatory behavioral adaptation (cf. Requin et al., 1991; Jennings and van der Molen, 2005). It should be noted that the term “strategic” in this context does not strictly imply that the allocation policy results from reflective planning. Rather, it can also be “triggered” by experiences of effort (cf. Goschke and Kuhl, 1993; Koriat and Goldsmith, 1996; Strack and Deutsch, 2004; Mojzisch and Schulz-Hardt, 2007). More critical is the aspect of prevention, which is in contrast to a straightforward formulation of a resource model presupposing a functional relationship between performance and test length (i.e., the time-on-task effect). Many early authors stated that individuals may not always be completely focused, but that moments of rest occur in between, and that these moments of rest serve a regulatory function. Bills (1931, p. 244), for example, has contended that “the rest afforded by these mental blocks maintains the objective performance of the individual at an average level.” Therefore, one might wonder whether explicitly administered (micro-)breaks would reduce block frequency, and whether a specific distribution of such microscopic breaks might lead to an optimization of performance and performance stability (Leth-Steensen et al., 2000; Ballard, 2001; Flehmig et al., 2007a; Langner et al., 2010; Steinborn et al., 2010; Unsworth and McMillan, 2014).

Bertelson and Joffe (1963) gave their participants a continuous four-choice task that lasted for about 30 min and required them to press one of four keys assigned to one of four digits (1–4). Their result also suggests that mental blocks enforce a rest period to ensure efficient performance afterward. While mental blocks were always preceded by a slowdown in response and a deterioration in accuracy, both were followed by a sudden improvement. Unsworth and Robison (2019) examined exactly these unintentional pause structures in the psychomotor vigilance test (PVT). The task requires simple responses to targets which are separated from each other by a random interval separating trials from each other (Langner et al., 2010, 2011; Steinborn et al., 2016; Massar et al., 2018). Even here, where targets are separated in time, a similar pattern to that observed by Bertelson and Joffe (1963) emerged, likely because the individuals are engaged in monitoring the time flow until target occurrence (Miller and Schröter, 2002; Steinborn and Langner, 2011, 2012). Lapsing occurred more or less in a rather periodical way and was not related to the time position as given by the length of intervals. In some way, this confirms Broadbent’s assumption that preparatory time (and any watchkeeping time in vigilance tasks) is not resting time but a rather arduous mental state of effortful engagement. Connected with this point, it has often been debated whether it is not the “rest” per se but the “change” in the nature of a current activity that leads to a recovery of mental functions. Helton and Russell (2015) have examined this particular aspect. The individuals were administered either with a baseline condition (no rest), a pure rest condition, and several “change” conditions, including typical mental operations (e.g., letter-matching task, etc.). As a result, performance was best for rest, worst for the continuous condition, while the other (“change”) conditions were somewhere between the extremes, indicating that any change in a task cannot compete with having rest, which again means that rest is always the best option to recover (even though experimenters usually lack precise control over what exactly participants do when instructed to “simply rest”).

The strategic model bears an important implication that is pertinent to pause research. From any straightforward formulation of a non-strategic model, the observation of a time-on-task decrement would indicate a decline or total depletion of resources (Baumeister et al., 1998; Hagger et al., 2010; Inzlicht and Schmeichel, 2012). In contrast, the strategic-resource model incorporates that individuals take preventive measures to either economize resource expenditure, or doing this rather impulsively, for example, when they suddenly experience effort (Thayer et al., 1994; Inzlicht et al., 2018; Hoffmann et al., 2021). Contemporary theorists such as Thayer (1989), Matthews (2021) proposed that it is the internal experience of states of the two aspects “energy” and “tension” that serves as the “tachometer” indicating whether to save or to spend effort to the task at hand. Such an approach automatically implies the question of how these indicators can be measured, either as emotions (experienced effort) or behavior (experienced lapsing), or a combination based on somatosensory experience and self-observation, arising from a stumbling of performance fluency at some point during the task (Kerr, 1973; Dreisbach and Fischer, 2011; Langner et al., 2011; Steinborn et al., 2016). In his pioneering work, Smallwood et al. (2004) introduced a technique where individuals are occasionally asked during the session whether they were still focused on the task or whether their thoughts were wandering elsewhere (on-task vs. off-task). Across numerous tasks and contexts, it was reliably found that individuals were far from being perfect as indicated by the proportion of “off-task” moments, typically increasing from about 20% to around 50% during a testing period, irrespective of whether task performance showed a decrement or not (e.g., due to practice effects), which is one of the reasons why only a few studies were able to show a correspondence between self-reports and the time-on-task effect in performance.

Although the theory of mental satiation is a highly relevant framework with historical roots in the field theory of Kurt Lewin (1890–1947), it is surprisingly underrepresented today. In this perspective, the critical variable that is relevant to performance is not a hypothetical volume of an energetic reservoir (reduced by work and replenished by rest), but the level of accumulated mental satiation that is experienced and considered the main driving force of behavior (e.g., Mojzisch and Schulz-Hardt, 2007; Kruglanski et al., 2012, 2014; Kurzban et al., 2013; Häusser and Mojzisch, 2017; Krishna and Strack, 2017). The satiation model therefore utilizes a diametrically opposing metaphor when compared to the resource model. This becomes clear when one compares the two following everyday phrases. (1) The engine runs until the fuel supply is exhausted. (2) The engine runs until it has run hot (i.e., exceeded a critical temperature). In the first example we have a preconditioning parameter while in the second one we have a delimiting parameter, but despite that, both metaphors may allow for similar predictions. Continuous activity leads to an increase in mental satiation, while breaks lead to a decrease in the same. This implies that one could derive exactly the same predictions based on a resource model or the satiation model, which is why these approaches should not, in a strict sense, be contrasted as opposing theories. Rather, they represent separate cultures that, despite having a common essence, differ partially in experimental approach as well as in somewhat different weighting of key arguments (Greenwald et al., 1986; Greenwald, 2012). Studies on psychological satiation, for example, typically put strong emphasis on the aspect of subjective experience as well as on those variables that contribute to or influence experience.

Psychological satiation is a phenomenon that arises when an action is carried out frequently and in a repetitive manner, so that the activity, which might initially be perceived as neutral, is increasingly perceived as being aversive in the course of continuous repetition. This occurs mainly in task types that consist of homogeneous and repetitive forms which are not intrinsically motivating by themselves (Pieters, 1985; Donk and Hagemeister, 1994; Van Breukelen et al., 1995; Neubauer and Knorr, 1998; Steinborn et al., 2018). Typical tasks that are very well suited to induce satiation experimentally can be found in the classic psychometric instruments for measuring active forms of sustained attention, the so-called speed tests, sometimes also termed concentration tests. Lewin (1928) assumed that the persistent and excessive repetition of the very same action over and over again undermines the built-in tendency of any cognitive system to strive for a gain of information and the experience of agentic competence, whereby a resistance gradient develops against the further continuation of the same action, which is subjectively experienced as aversive. However, when individuals are instructed to continue the task over a longer period of time, an increasing conflict develops between two opposing tendencies that can only be resolved by exerting willpower. Although individuals experience aversion as a cause, Lewin considered it a perceivable indicator in terms of a phenomenological by-product of the internally created tendency (Robinson and Clore, 2002; Strack and Deutsch, 2004). More recently, Mojzisch and Schulz-Hardt (2007) have studied the determinants of mental satiation in various contexts, based on field theory. In one series of experiments, groups of individuals were instructed to complete speeded test of different workload levels over a period of 20–40 min, using a loaded mental addition test (Düker, 1949), and the subjective experience was assessed before and after the test. It was found that a high level of satiation developed particularly when the task did not allow for any resting by mind wandering, that is, when it was both occupying and repetitive.

Based on classic theorizing on accumulating mental satiation and the resulting distraction as well as on the compensatory function of rests, Van Breukelen et al. (1995) presented a psychometric model to characterize and predict performance and performance fluctuations during sustained mental work. In essence, it is posited that the averaged reaction times (typically RT mean) are composed of two components: responses emitted in the state of mental focus and those emitted in the state of distraction. The model is a generalization of earlier psychometric models of cumulative inhibition, which are based on broadly similar assumptions but different in their specific statistical parameters. The theoretical framework of the inhibition model is Hull’s (1943) theory of reactive inhibition, and its implementation in the Rasch model for speed tests (Rasch, 1980; Pieters, 1985; Baghaei et al., 2019), which postulates that during ongoing monotonous processing, a kind of negative drive develops that forms an opposing gradient to the current goal orientation. In the context of performance testing, this can be seen as a “distraction tendency”. The gradient increases as a function of the throughput processed and immediately leads to distraction when a certain threshold is reached. The inhibition tendency thus increases monotonically during task processing and decreases during periods of distraction or during rest breaks.

Some studies are particularly suited to highlight theoretically fruitful research approaches in this context. For example, Sanders and Hoogenboom (1970) presented their participants with a six-choice RT task characterized by a rapid pace (response-stimulus interval = 60 ms) with either a continuous work or a rest-pause condition. The digits 1–6 served as targets and were mapped to six separate buttons. Responses became faster on average in the rest-break condition, while they remained the same in the continuous work condition. Furthermore, a cumulative distribution function (CDF) analysis revealed that the two conditions did not differ in terms of the fastest, but only in terms of the slowest CDF percentiles. Sanders and Hoogenboom (1970) argued that this reflects that rest breaks proactively prevent the occurrence of mental blocks and in this way reduce performance variability. This interpretation is consistent with earlier suggestions (Jersild, 1926, p. 34). More recently, Steinborn and Huestegge (2016) examined the effect of rest on performance and experience as a function of the factors “rest,” “demand,” and “time on task.” As a result, rest (vs. no rest) had beneficial effects on performance, which increased with time on task and was more pronounced for hard than for easy arithmetic. The CDF percentile analysis revealed that rest particularly reduced the frequency of dropouts and lapses. The pre–post assessment of experience revealed a differential pattern: while energy and engagement tended to decline, there was no effect of tension and distress. Taken together, these studies emphasize the importance of distributional analyses to draw relevant theoretical conclusions here.

The majority of studies theorizes (in some or the other way) on grounds of resource models borrowed from the cognitive-experimental literature on multitasking research. An essential detail of particular models concerns the distinction of a variable vs. a fixed volume of capacity, with the former assuming a mobilization of capacity by immediate demand and the latter assuming an alternation or dynamic partial allocation of capacity between channels processing task-related vs. task-unrelated information. Another potentially important detail concerns the role of how feelings (before or during) task processing affect performance levels (e.g., energy, tension, motivation). The corresponding research traditions can be classified into three categories, (1) considering feelings as phenomenological by-products of cognition, (2) viewing them as a sort of “tachometer” of internal state indicating room for vigorous action or need for recuperation, or (3) as the underlying “cause” of the observed performance differences. Despite these pure categories, a transactional perspective would argue that while feelings may primarily have an indicator (tachometer) function, the mere act of reading out the internal state can by itself lead to distraction or conflict, similar to that in multitasking situations. Accordingly, an observed performance decline can actually be the result of distraction and self-referential processing (Wells and Matthews, 1996), or more precisely, from the act of monitoring and comparing actual values with an internal standard, and from subsequent evaluation or self-regulation (Carver and Scheier, 1990; Strack and Deutsch, 2004; Hewig et al., 2011; Krishna and Strack, 2017; Steinborn and Huestegge, 2020).

According to a mobilization model of sustained–attention performance, capacity is primarily demand-driven, which means that it is the immediate demands of the task that triggers capacity supply and not a deliberate decision. To a certain extent, it can be controlled at will, for example, when advance information is provided (Brown and Braver, 2005; Botvinick and Braver, 2015), or by a prompt (or reminder) to increase focus (Strayer and Kramer, 1994; Kleinsorge, 2001; Falkenstein et al., 2003; Steinborn et al., 2017), though it is impossible to control it precisely or to keep it steady. In a situation that emphasizes the aspect of “maintaining” performance levels over extended time periods, precise control would require a time-scheduled capacity threading between states of focusing and those of monitoring over the task period (cf. Craik, 1948; Humphreys and Revelle, 1984; Van Breukelen et al., 1995; Fernandez-Duque et al., 2000; Steinborn et al., 2017). This means that capacity varies over the duration of the task and thus can be characterized by a hypothetical ratio of utilized and spare capacity. According to Kahneman (1973), such variation is due to the fact that the allocation policy is not always set on focusing but sometimes sways capacity to other activities (monitoring or mind wandering), resulting in slower or even sluggish responses during these periods. Such a view of intermittent resource allocation to active operating vs. passive monitoring provides a natural way to explain the trial-to-trial response-speed variability that is commonly observed in reaction–time series (Flehmig et al., 2007a; Steinborn et al., 2016; Klein and Robinson, 2019).

A formal way to represent the aspect of performance fluctuation is to model data within the framework of the mixture-models type of speeded performance (Pieters, 1983, 1985; Miller, 2006; Schwarz and Miller, 2012). Basically, it is assumed that observed performance (i.e., reaction time) is composed of trials where the individual was under a state of focusing or a state of reduced focus (e.g., through non-registered rest, either enforced or taken). There are also many accounts that rather implicitly (and in a less formal way) refer to a similar idea without explicating the precise mechanisms of how exactly this is reflected in performance parameters (e.g., Humphreys and Revelle, 1984; Jensen, 1992; Leth-Steensen et al., 2000; Stuss et al., 2003; Robinson and Tamir, 2005; Flehmig et al., 2007a; Cheyne et al., 2009; Thomson et al., 2015). At a more sophisticated level, response variability is studied by analyzing the individual response-time distributions, often computed as a vincentized percentile function. Typically, reaction-time distributions tend to be leaning toward the right side (i.e., corresponding to the slower percentiles), in other words, to having a long tail toward the right. However, there is an important distinction that is often misunderstood in the literature. Precisely, to say that an experimental factor produced an effect on RT variability (i.e., to say it affects the ratio of focused vs. non-focused trials) requires that the factor has a selective effect on the slower percentiles of the distributive function that goes beyond mere scaling variability (De Jong et al., 1994; Miller, 2006; Steinborn et al., 2017), as would be indicated by the RT coefficient of variation (Flehmig et al., 2007a).

A capacity-mobilization model of sustained-attention performance can be characterized by some important key aspects: First, the hypothetical value of “capacity” is a latent (or theoretical) variable that is represented by the empirical indicator (or performance) variables, that is, in the speed of processing or the throughput of information processed per unit of time (Thorne, 2006; Szalma and Teo, 2012; Vandierendonck, 2017; Steinborn et al., 2018). Further, the mental resources devoted to continuous performance are essentially limited in two ways, (a) that the individuals punctually engage is only one mental activity, and (b) that they can maintain engagement for only a brief period. Task-relevant processing is referred to as utilized capacity and indicated by performance, while task-unrelated processing is referred to as spare capacity, which is the residual that can only be inferred indirectly (from the lack of optimal performance). The relation between both (i.e., the spare–utilized capacity ratio) varies across a sequence of trials, so that as individuals engage in task operations, spare capacity is reduced at the costs of utilized capacity, and vice versa, when they disengage from the task-relevant operation (i.e., microbreaks, either enforced or taken), utilized capacity is reduced at the cost of spare capacity (Steinborn and Huestegge, 2016, 2017, 2020). To know the dynamics of these alternation between spare and utilized capacity, it would be necessary to inspect not only the distributive function but also the time series of reaction times, separately for each individual. With this respect, only a very few studies examined this in greater detail (e.g., Kraepelin, 1902; Bills, 1931; Laming, 1979; Cheyne et al., 2011).

While a spare–utilized capacity model is both prudent and parsimonious in its base assumptions, more recent theoretical proposals seem to overstretch their structural basis, which generates a problem when it comes to using theory for deriving predictions. For example, Thomson et al. (2015) made an attempt to assume that an exactly equal amount of capacity alternates between two processes, one devoted to the task-related and the other reflecting task-unrelated processing (i.e., mindwandering), and over a series of trials, individuals are assumed to alternate between these two channels, while the capacity remains equal across all trials, thereby refuting any process of mobilization completely. According to this conception, performance is measured by standard parameters, while mindwandering occurs either by (passively) drifting off or by (actively) being on pause as assessed by self-report probes. The proportion of these conditions is assumed to change over time (i.e., the proportion of “off-task” conditions increases) which gives a measure of the amount of mindwandering, that is, not being on task. However, a problem here lies in a confusion of using natural language and theoretical formalism when speaking about performance effects. In natural language, one would agree that the brain does not consume more energy overall during phases of mental work as compared to alternating phases of mindwandering (but see Gailliot et al., 2007, for an opposing view). In the same vein, everyday language would link the term mobilization to changes in physiological activity, which is not addressed in the formulation of a spare–utilized capacity model. The problem is that any formal model that specifies capacity for alternative hypothetical processes (e.g., A = on-task; B = off-task) creates a problem in the formal structure in a way as the capacity budgeted for the on-task process is determined by the performance measures while those budgeted for the off-task process is indicated by the self-report probe trials. These self-reports give an indication that the individuals were not focused but absentminded, but they do not deliver an equivalent that allows to determine the capacity required during both alternating phases (Miller and Ulrich, 2013).

In summary, we argue that a formal model must be able to specify the degree of engagement of both the task-related and the task-unrelated process, and it must provide rules for determining the behavioral indications signaling this engagement. This is not possible in a model case where performance can be measured only for one task, while the other process (i.e., rest pauses, mindwandering, etc.) is indicated only indirectly, by the absence of a performance optimum (Van Breukelen et al., 1995; Miller and Ulrich, 2013). In order to be able to formulate a model that allows us to predict the direction or amount of capacity change during time on task, or afterward during rest breaks, it is crucial to link the theoretical variables to a connected measurement concept that specifies how the values of the model parameters are indicated by behavior and performance (most often performance speed and accuracy). Further, a measurement theory must also include a specification of performance characteristic trade-offs that presumably covary with the hypothetical state of mental fatigue (or overload, etc.). In most theoretical accounts, it is rather implicitly assumed (or simply presupposed) that both performance speed and accuracy deteriorate over time, indicating reduced mental efficiency, while pauses are assumed to restore mental efficiency. Empirically, it is often found that information processing slows down with no substantial effects on accuracy, and sometimes a speed–accuracy trade-off is observed (e.g., individuals become faster but more impulsive). Another aspect concerns the use of averaged performance parameters. Since mindwandering in continuous performance tasks is particularly evident in the slow percentiles of the reaction time distribution, it would be essential to additionally inspect time series data (Laming, 1979; Cheyne et al., 2009). In the field of rest-break research, these aspects are lacking, and in fact there are currently only a few studies that actually address these aspects in some or the other way (as outlined above).

As noted, all existing theoretical approaches can broadly be classified into two groups, which we term “resource model” and the “satiation model.” According to the resource model, mental work leads to a decrease in the energetic reservoir, while rest breaks lead to an increase. According to the satiation model, mental work leads to an accumulation of satiation, but rest breaks lead to a reduction. In their simplest form, both model categories utilize a linear metaphor that is merely inversely poled (either as a permitting or delimiting condition). One particular difference lies in the emphasis on aspects of subjective experience. While the satiation model ties the loss of perceived intrinsic motivation to objectively measurable performance, the resource-model theorizes on the reduction of an energetic reservoir by mental work, and any reference to subjective experience is neither necessary nor mandatory. In fact, the proposal that experience and performance is connected or even causes each other is by no means trivial, at least it cannot automatically be taken as a presumption (Langner et al., 2010; Matthews, 2021). The literature is completely mixed in this regard. While a substantial number of studies reported a correspondence of pre-tested subjective state (e.g., energy, motivation, etc.) and subsequent performance, only a few were able to demonstrate a correspondence of the time course of these variables. For example, Thomson et al. (2014) showed a congruent temporal trajectory of self-reported mind-wandering tendencies and performance in one (relatively easy) task, while there was an opposing tendency of these measures in another (more complex) task. Clearly, the arrangement in one task was simpler and more repetitive than in the other one, providing fewer possibilities for procedural learning (e.g., Compton and Logan, 1991, see Figure 2; Healy et al., 2006).

There are many aspects to be taken into consideration when conducting an experimental study on the effect of rest breaks on feelings and performance. Most of them relate to design, tasks, and performance measurement and are part of the basic knowledge repertoire of a skilled experimenter (Greenwald, 2012; Miller and Ulrich, 2013). But there is one aspect that pervades the entire research, the potentially concealing effects of test practice and item-specific learning (Flehmig et al., 2007b,2010), which is difficult to control properly (Donk and Hagemeister, 1994; Hagemeister, 1994, 2007). Precisely, performance-based tests such as typing, cancelation, or mental-arithmetic are typically used as propagation tasks to induce ego depletion (Hagger et al., 2010; Vohs et al., 2021). These test forms are typically administered as self-paced versions thus requiring sustained-information transfer (Humphreys and Revelle, 1984; Steinborn et al., 2018), they are highly demanding, and most crucially, they achieve exceptional test–retest reliability, since many items can be processed per unit of time. Yet, it should be considered that the time-on-task effects on performance do not always produce a global response slowing, and further they are (to an unknown degree) subject to procedural learning. Finally, fatigue might affect these processes differently at different points in practice (Healy et al., 2004, 2006). Some authors opted for the use of simple tasks like the sustained-attention-to-response test (SART, Manly et al., 1999; Seli et al., 2013; Head and Helton, 2014), the psychomotor vigilance test (PVT) (PVT, Dinges et al., 1997; Steinborn et al., 2016; Unsworth and Robison, 2019), or even detection-based vigilance tests as measure of choice (Warm et al., 1974; Thomson et al., 2016). While these tasks are less susceptible to learning, due to their simplicity, the items in these tests are typically spaced in time so that a relatively low number of trials is obtained during testing as compared to self-paced tests, where each item follows after the previous one with no time in-between trials. Consequently, test–retest reliability has often been found to be a weakness of the former (vigilance tests) as compared to the latter (self-paced tests) if one considers the economy of testing (Miller and Ulrich, 2013; Steinborn et al., 2018).

Many published studies actually suffer not merely from methodological weaknesses of aspects connected to measurement methodology, but also from inadequate (or incomplete) design issues. As noted, fundamental to the study of rest breaks is that several distinct effects of rest can be distinguished with a slightly different meaning of each of them, not to speak of the contextual effects that often change the dynamics of these effects, for instance, when testing effects of rest under sleep deprivation (Wilkinson, 1959; Sagaspe et al., 2006; Bratzke et al., 2009, 2012). This is not considered in most of the literature. Essentially, rest-break effects as obtained via an experimental design that contains a time-on-task condition can be divided into global and local difference effects (cf. Steinborn and Huestegge, 2016). In order to examine the global effect of rest on performance, the performance trajectory between the control (no-rest) condition and the critical (rest) condition is typically examined. This can be done by using a group-based design where the critical conditions are compared in terms of a between-subject comparison. The global benefit of rest is obtained as the relative difference in the time course of the performance curve (i.e., the slope of the time-on-task effect), as indicated by the relevant performance parameters. The local-rest effect can (for obvious reasons) only be studied in the critical group (where rest-breaks are included). Here, the performance immediately before and after rest (i.e., the pre–post effect) of only the critical group is contrasted. This approach is formally equivalent to an analysis of trial sequences surrounding a critical event, such as errors (Brewer and Smith, 1984; Steinborn et al., 2012; Jonker et al., 2013), or attentional lapses (Bills, 1931; Bertelson and Joffe, 1963; Steinborn et al., 2016).

While the effects of rest can basically be studied in any task form, speeded tests (e.g., mental arithmetic, letter cancelation, coding, etc.) are particularly suitable for performance measurement (cf. Van Breukelen et al., 1995). Since they require continuous work and because items are not spaced but rather compressed in time (items follow immediately after each other), they usually provide superior test reliability (Steinborn et al., 2018; Wühr and Ansorge, 2019). Individuals in these tasks are typically required to continuously respond to a series of successively presented targets, with each target following immediately after responding to the previous one, with no feedback given after errors. It should only be noted that individuals can efficiently detect any errors that they make, and they would even do so even when instructed to ignore them (Maylor and Rabbitt, 1995; Steinborn et al., 2012). One advantage is that they can be administered both as computerized versions of paper-and-pencil forms without changing the nature of the task (Van Breukelen et al., 1995; Steinborn et al., 2018). Since these tests offer the opportunity to simultaneously consider several performance aspects, they deliver some additional information about the working style. Typical measures are the speed, accuracy, and variability (Pieters, 1983, 1985; Flehmig et al., 2007a), or combined-efficiency indices (Thorne, 2006; Szalma and Teo, 2012). To summarize, a task must be applicable to experimentally study rest breaks, which means in the first place that it must put some strain on the individual, and it must allow for a reliable performance measurement. Although this may seem trivial, this condition is not met in the majority of empirical studies. However, without precise measurement, a sufficient number of trials, and adequate performance output variables, it will not be possible to derive correct theoretical conclusions from empirical results (Miller and Ulrich, 2013; Steinborn et al., 2018), and a proliferation of false conclusions would finally produce a chaotic plentitude of contradictory knowledge clusters rather than a systematic understanding of the underlying mechanisms of rest.