Perceptual rhythms by phase-aligned perceptual performance peaks across trials Authors: Tobias Schoeberl, Stefan Treue Published: 2026 ABSTRACT AbstractPerceptual processing may rhythmically alternate. There is a line of evidence from dense sampling psychophysics supporting this claim: In such studies, performance in visual target detection or discrimination is repeatedly tested at various fine-grained time intervals relative to a reference event. This reference event is presented at the beginning of each trial, and it is thought to reset some internal oscillation. Performance time courses following the reset often show rhythmic oscillations in the sub-second range. The interpretation has been that several quickly alternating peaks and troughs of perceptual efficiency exist. Here, we propose an alternative. Instead of being rhythmic, the perceptual processing efficiency time course might exhibit only a single peak in each trial. If the temporal location of this peak is constrained to co-occur with one of the peaks of the internal continuing oscillation, rhythmicity in overall performance time courses can be explained when performance is computed across many trials. Applying this perspective to three published datasets (N = 44, N = 34, N = 11) and running simulations, we confirm the possible existence of such phase-aligned peaks in perceptual efficiency. We show that foreperiod priming (behavioural advantage of repeating the same time interval between reference and target from one trial to the next) alternates across time. This is predicted if peaks in perceptual efficiency are consistently allocated to those time intervals that coincide with a peak in the underlying oscillation. Priming therefore works best at these intervals, which coincide with a peak in the underlying oscillation.Subject terms: Perception, Attention, Human behaviour, Sensory processing Introduction Answering the question how cognitive functions such as perception, attention or working memory are implemented by the neuronal system is one of the core challenges for cognitive neuroscience1–3. Over the past two decades striking evidence for the rhythmic nature of such cognitive processes seemed to accumulate2,3. As the temporal dynamics of neuronal activity in nervous systems is to some extent rhythmic, these observations could help to link perception and cognition to their presumed neuronal substrate4–9. An influential theory postulates an intrinsic perceptual sampling rhythm operating in the high theta or low alpha range that could be identified with frontoparietal neuronal oscillations3,6,10. By default, the perceptual system might sample the environment at a constant rate (7–8 Hz) with alternating peak and trough phases of increased and reduced perceptual efficiency. When more than a single item becomes task-relevant, this rhythm could be divided across these items so that rhythmicity becomes apparent at a slower pace11,12. The scope of findings that seem to support the theory of rhythmic perception goes beyond a single domain, but the first compelling examples illustrating the idea come from studies investigating spatial attention in vision3,11,12. Landau and Fries used a dense sampling approach to show rhythmic patterns in behavioral performance time courses: a salient reference event (an attentional cue) was used in each trial to reset the presumed perceptual oscillation. Performance in target detection was then repeatedly tested at many fine-grained time intervals relative to the reference, so that accuracy time courses could be computed. The time courses were then Fourier-transformed, and permutation tests were used to assess the statistical significance of spectral amplitude peaks. Observing a significant spectral amplitude peak in such Fourier-transformed time courses is usually interpreted as evidence for a rhythmic oscillation in perception11,12. Indeed, Landau and Fries observed 4 Hz oscillations in performance time courses when the attentional cue, the reference event, was in the right visual hemifield. In addition, oscillations with targets at the cued and uncued location appeared to be in anti-phase, suggesting that an attentional factor was involved that caused a rapid switching between the two locations. This result has been interpreted as the division of the basic 8 Hz perceptual rhythm across the two locations11–13. Numerous dense sampling studies have used this behavioral approach in the past decade. Nevertheless, the exact mechanisms that give rise to oscillations in performance time courses in this type of study are not fully understood11–26. Brookshire recently pointed out that the commonly used permutation tests, that have been used to test the statistical significance of peaks in the Fourier transformed performance time courses operate on the null hypothesis that there is no temporal structure in the performance time course22. Thus, the test can show significant results with periodic as well with aperiodic temporal structure. The majority of the results in the recent literature are therefore not uniquely proving the presence of oscillations, because aperiodic temporal structure could explain the significant results. Although no consensus has been reached regarding this important caveat22–26, the results of dense sampling studies are often regarded as evidence that alternating phases of high and low perceptual processing capacities exist in the sub-second range. This is because the psychophysics is in accordance with results from brain imaging and electrophysiology in humans and non-human primates, which show statistical associations between the phase of neuronal oscillations and performance in some behavioral task (see also the discussion below) 4–9. Perception is thus often thought to have a rhythmic component. No matter if the presumed rhythm of perception is sinusoidal or not, we understand this idea to imply that the perceptual efficiency time course (PEtc) shows several, equally spaced local maxima and local minima in the sub-second range. We refer to this interpretation in what follows as the true/intrinsic rhythmicity model13. Here, we present an alternative to this intrinsic rhythmicity model. This alternative is, as we argue, fully compatible with the existing evidence, but it has important consequences for whether we think about perception as truly rhythmic or not. We grant that there are robust statistical associations between the phase of neuronal oscillations and perceptual performance4–9. We also grant that there are rhythmic and maybe even nearly sinusoidal patterns in behavioral performance time courses in dense sampling studies11–26. However, all of this can be explained, even when the temporal evolution of perceptual resources, henceforth referred to as perceptual efficiency time course (PEtc), does not follow the rhythm of the underlying oscillation. Why is that? Perceptual efficiency is a system property of the brain. We define the PEtc as the potential of the neuronal system to perceive a stimulus at each point in time, if a stimulus would be presented at that time22–24. As a result of temporal orienting, the PEtc may increase and decrease, resulting in local maxima and minima within the PEtc, but the questions how fast these alternations are and if these alternations are really rhythmic is debatable. We argue that the PEtc can, in fact, fluctuate in a non-rhythmic way or at a much slower pace compared to what the prior intrinsic rhythmicity interpretation of empirical results would suggest. The reason is that statistical associations between perceptual efficiency and certain phases within a cycle of some underlying oscillation can be explained when local maxima in the PEtc systematically phase-align to specific time points within a cycle of an underlying oscillation. This is true even when the local maxima in the PEtc are rare. They might, for example, occur only once per trial in a dense sampling study (that is, they might only occur in every 7th or 8th cycle of the underlying oscillation). Figure 1A illustrates this idea. Peaks in the PEtc are constrained to align to very specific, equally spaced time points that represent specific phases within the cycle of an ongoing oscillation. However, peaks in the PEtc only occur once per trial, not in all cycles of the underlying oscillation. We therefore regard the PEtc as potentially non-rhythmic. It does certainly not need to follow the peaks and troughs of the underlying oscillation it aligns to. We refer to this conception in the following as the phase-alignment model (Fig. 1B). Illustrates the idea of the phase-alignment model as opposed to the intrinsic rhythmicity model. A Depicted in red in the top row is an underlying neuronal oscillation (anchor oscillations) which is reset in each trial of a dense sampling study to start oscillating at time point zero at a specific phase. According to the intrinsic rhythmicity model, this oscillation determines perceptual efficiency such that the perceptual efficiency time course (PEtc) unfolds across time with the peaks and troughs of the underlying oscillation. When performance is repeatedly tested at many intervals relative to the reset, the peaks and troughs become apparent in the behavioural time course. By contrast, according to the phase-alignment model, the PEtc only has a single peak in each trial. If subjects focus their attention in time the peak in the PEtc aligns to one of the peaks in the underlying anchor oscillation (black line in the rows below the anchor oscillation). Rhythmicity in the overall performance time course in dense sampling studies can be explained when this peak shifts to different locations across trials (with the constraint that it is always aligned to one of the peaks of the anchor oscillation), even when the PEtcs in each single trial are not rhythmic. Notice that we assume in the phase alignment model that the PEtc alone determines perceptual efficiency at each point in time, not the anchor oscillation. The anchor oscillation plays a role only insofar as the peaks in the PEtc align to one of the peaks of the anchor oscillation. B illustrates how rhythmic oscillations emerge in the average time course of a dense sampling study, according to phase-alignment model. There is only a single peak in the perceptual efficiency time course in each trial. The temporal location of this peak can vary, with the constraint that the peak is always aligned to the phase of an ongoing underlying anchor oscillation (these peaks are represented by the vertical dotted lines). If this underlying oscillation has, for example, five cycles within the observed time window (range of ISIs), there will be five different PEtcs. With a very high number of trials, the observed performance time course will approximate these PEtcs, which assign a probability of a correct response to each ISI. The average time course across all trials, in turn, is then the average of these five PEtcs (red line, lowest panel on the left). The panel on the right shows the Fourier-transformed average time course in the frequency range of interest. Let us consider how rhythmic structure may in this way emerge in dense sampling studies which use experimental designs similar to that used by Landau and Fries11,12. It is known that temporal orienting can allocate perceptual resources to certain time intervals within a structured sequence of events27,28. As explained above, we assume that an underlying neuronal anchor oscillation exists, which is reset by the reference event (e.g., the cue in Landau and Fries’ study)11. This oscillation, however, does not determine the PEtc. Instead, the good phases of the anchor oscillation only define windows of opportunity for the local maximum of the PEtc to align to. The question is which window of opportunity relative to the reference will be selected in each trial, if there is indeed only a single peak in the PEtc in each trial. Reports of inter-trial effects that could determine this selection are abundant in the attention literature29–34. Much research on inter-trial effects with attention has focused on spatial or feature-based attention, but such effects exist in the temporal domain too27,28,35. For example, in a rapid serial visual processing task, reacting to a target stimulus which is presented at a given timepoint relative to a reference in a trial n, is facilitated when it is preceded by a trial with a target at the same relative time point in trial n−127. It is possible that such foreperiod priming will determine trial-by-trial shifts of selected intervals to a considerable degree given the strong impact of inter-trial contingencies in attentional control27–35. One can now see how rhythmicity in overall performance time course will emerge even without rhythmicity in the PEtc on the single trial level: In a subset of trials, the benefit associated with the phase-aligned peak of the PEtc will become behaviorally relevant, namely in those trials in which a similar time interval is repeated from trial n−1 to trial n. If the effect is large enough, this should induce rhythmic patterns in the overall time course, because performance will be better at the phase-aligned timepoints associated with peaks in the PEtc compared to timepoints “in-between”, to which peaks of the PEtc cannot effectively align to (see also Fig. 1B). The reason for this is, that resources cannot be precisely allocated to the timepoints in-between, because the PEtc is always phase aligned to specific phase of the anchor oscillation. In the following, we provide a proof of concept for the above-described idea by using simulations. We assume that the phase-alignment model holds true. We show that the above-described mechanism can explain rhythmicity in dense sampling studies, when a singular peak in the PEtc is in each trial phase-aligned to one of the peaks in an underlying oscillation, and when this peak shifts according to when the observer expects relevant events. We also present a way to test this phase-alignment model. If we assume that foreperiod priming – the priming of the attended time interval in trial n by the observed time interval between the reference and the target in trial n–1 plays a role, we predict rhythmic fluctuations in the magnitude of foreperiod priming across the observed time window, because only certain timepoints can be efficiently primed, those to which peaks of perceptual resources can be allocated. Finally, we test this prediction with three publicly available datasets from the dense sampling literature, confirming the existence of rhythmicity in foreperiod priming36–38. Methods Transparency The present study was not pre-registered. Conventional statistical analysis of rhythmic performance We simulated results of dense-sampling experiments with a detection task (trial outcomes: hit and miss) with trial counts of 30,000, 60,000 and 90,000. For each experiment, data were simulated for 81 ISIs (inter-stimulus intervals between reference event and target) ranging from 0 to 800 ms following a hypothetical reference event with a virtual sampling rate of 100 Hz. In each trial, the ISI was determined randomly. To determine the behavioral outcome in each trial, we used probabilities specified by a PEtc that assigned a probability of a hit to each ISI. The PEtc was uniform at 0.6 in the first trial. In each subsequent trial n, we simulated temporal orienting as a result of foreperiod priming by adding a Gaussian onto the flat PEtc. Formally, the probability of a hit at each ISI was determined by the following formula: with σ ∈ {20 ms, 40 ms, 60 ms}; effectSize ∈ {0.1, 0.2, 0.3}. μ was in each trial selected such that the Gaussian was centered at one of the local maxima of a sine function O (in each trial, the local maximum which was closest to the ISI in trial n−1 was selected) given by where i denotes the time interval from the reference in seconds, and F = 7.5 Hz. The Gaussian was thus in each trial n centered at the peak of a 7.5 Hz oscillation which was closest to the ISI in the preceding trial n−1. That is, peaks of the PEtc occurred in a phase-aligned manner. The simulated data were analyzed first in a conventional way by computing the behavioral time course, detrending with a second-order polynomial and employing a multitaper spectral density estimation technique (bandwidth: 1 Hz; no zero padding; using Chronux library version 12.2: http://chronux.org/; function mtspectrumc39). In order to assess statistical significance of the spectral power values at the frequency bin of interest at 7.5 Hz, we used a permutation test, shuffling the ISI across trials and computing 500 surrogate data from which p-values could be determined. 100 simulations were conducted in each condition and we counted the number of simulations in which we observed a significant result at the frequency bin closest to 7.5 Hz (Fig. 2, upper panels). Results of simulations illustrating the emergence of rhythmicity in performance time courses by phase-aligned peaks of perceptual efficiency. A Depicted on top are the results of proof-of-concept simulations. Data were simulated with different trial counts and with different model parameters. In each trial, the phase-aligned peak of the perceptual efficiency time course (PEtc) was simulated as a Gaussian on top of an otherwise flat hypothetical performance time course indicating the probability of a hit. Shifts of the attended timepoint across trials was modeled as a result of foreperiod priming27,28. In each trial, the ‘good’ phase of the anchor oscillation which was closest to the foreperiod in the preceding trial n−1 was selected for aligning the peak in the PEtc. Rhythms at the corresponding frequency of the anchor oscillation (7.5 Hz) are detected in a high proportion of simulated experiments, when data are analyzed with the conventional technique of computing power spectra of performance time courses and permutation testing11,12. This is true across a range of model parameters with empirically plausible trial counts. B The heatmaps below depict the results of the simulations in which rhythmicity detection was tested with the conventional technique (assessing rhythmicity in the average performance time course) compared to the testing of rhythmicity in foreperiod priming. Rhythmicity as a result of phase alignment of single peaks in the PEtc and foreperiod priming was simulated as before with a trial count of 30,000 across various model parameters. Detection rates (the proportion of simulated experiments in which a significant oscillation was observed at 7.5 Hz) are color-coded (between black and white) in the two panels on top. The panel below shows the difference. Results indicate that, dependent on the model parameters, analyzing rhythmicity in inter-trial priming can be more powerful than doing the analysis in the conventional way. Rhythmicity in foreperiod priming A second series of simulations was conducted, each with 30000 trials. Various parameters for the effect of temporal orienting as a result of foreperiod priming were tested (σ ∈ {20 ms, 40 ms, 60 ms, 80 ms}; effectSize ∈ {0.1, 0.15, 0.2, 0.25 and 0.30}). The number of significant results was counted with conventional analysis (see the procedure described in the last section). In addition, we tested for rhythmicity in the time course of foreperiod priming. That is, we computed the performance time course with a 30 ms running window, and we computed at each ISI the foreperiod priming effect: we computed the performance across all trials with an ISI in the corresponding window and in which the ISI in the trial n−1 was also within a 30 ms window centered at the corresponding ISI value. For each ISI, we subtracted the resulting performance value in primed trials from the corresponding performance value of the overall time course. We thus obtained an estimate of how efficient foreperiod priming was at each ISI. The resulting time course of foreperiod priming was detrended with a second-order polynomial and a multi-taper spectral density estimation technique was employed as described above. For assessing statistical significance of the observed spectral powers, we shuffled the ISIs in trial n−1 randomly across trials, thus removing inter-trial contingencies, and we repeated the analysis 500 times to obtain surrogate data from which p values could be computed (ps < 0.05 were considered statistically significant; see Fig. 2, panels below). Re-analysis of public datasets We tested for rhythmic foreperiod priming effects in public datasets from three studies18,36–38,40–42. Although these studies differed substantially in terms of the exact method (sampling rate, number of ISIs, experimental conditions), all three studies tested performance with varying ISIs relative to some reference event, and data were published in a way so that the trial sequence with the corresponding ISI could be inferred. We could thus test for rhythmic foreperiod priming in all cases. For detailed information about the methods and research participants involved, we refer to the corresponding publications18,41,42. The exact processing pipeline was adapted according to the particular requirements for each dataset (see sections below for more information). Notice that we used a non-parametric permutation test to test the hypothesis that there is (rhythmic) temporal structure in the time course of foreperiod priming. Thus, we do not make specific assumptions about how the data are distributed. However, as permutation tests have been criticized, because they might inflate the rate of false positives in the presence of non-periodic temporal structure, we confirmed the results in an additional analysis with Brookshire’s AR surrogate method (Supplementary Note 1)22. Michel & Busch (2023) Data of this study were retrieved from https://osf.io/b5q2u/18,36. The corresponding study had ethical approval18. Participants in this study (N = 44; five of those were excluded in the original study; here, we analyze data from all participants; reported values: 30 female, 9 male, the way sex and gender was determined was not further specified; there is no information about whether data on SES, communities of descent, or race was collected18; Mage = 22 years, range age: 18–29 years) had conducted 660 trials of a behavioral task. In this task, performance had been tested relative to cue onset in an attention task with ISIs ranging from 41.7 to 1041.7 ms. Cues had either appeared at the location of an upcoming target (cued condition) or away from this location (uncued condition). For the present purpose, we collapsed data across cued and uncued conditions and across participants. We determined for each trial the ISI in trial n−1. No-go trials [trials without target, (target position “0”)] and trials in which trial n−1 was a no-go trial were excluded. Miss trials (trials in which participants failed to respond, even though a target was present) were also excluded from the analysis. The dependent variable was reaction time (RT): we computed the overall performance time course T by computing reaction time at each ISI bin (RT, trimmed mean by excluding 5% of the most extreme reaction times to account for the skewness of RT distributions). Next, we computed at each ISI the RT only for those trials in which the corresponding ISI was preceded by the same ISI in trial n−1. The corresponding value was subtracted from the value of the overall time course T at this ISI bin. After doing this for all ISI bins, we had estimated the time course of foreperiod priming. This time course was detrended with a second-order polynomial and transformed into the frequency domain (by using the above- described multi-taper technique). Statistical significance of the spectral powers was assessed as described above by using a permutation test (shuffling the trial label of trials n−1 across trials, 1000 iterations). Previous reports of attentional oscillations estimated the effect of rhythmic perception/attention to be at around 8 Hz3,10,11. We therefore primarily tested for significant spectral power peaks at the two frequency bins closest to 8 Hz. The control for multiple comparisons was performed across these two bins (setting the significance threshold to a p-value of p = 0.025). Huang et al. (2015) Data were retrieved from https://osf.io/69by5 (data published as part of a subsequent study by Wang et al.)37,40,42. The corresponding study had ethical approval. Participants provided written informed consent and were compensated for their time40. Data from two experiments from the study by Huang et al. were reanalyzed40, from one with a masked priming task (N = 18 participants; 8 male, 10 female) and one with an unmasked priming task (N = 16 participants, 6 male, 10 female). The way sex and gender were determined was not further specified; there is no information about whether data on SES, communities of descent, or race was collected 40,42; Average age of participants across all subject in that study was 24 years. In the corresponding study, participants had performed a two-alternative forced-choice task responding to the pointing direction of an arrow target which was preceded by a prime [a second arrow either pointing into the same direction (congruent condition) or in the opposite direction (incongruent condition) as the target arrow]. There had been 50 ISIs sampling performance at 50 Hz: ISIs were ranging from 0 to 980 ms in 20 ms steps (16 trials per ISI per condition were recorded for each participant; 1600 trials in total per participant). In addition, there had been a masked (where the prime’s pointing direction was visible) and an unmasked condition (with the prime’s pointing direction being invisible). For the purpose of the present analyzes, we collapsed data across congruent and incongruent trials and across participants in order to test for rhythms in foreperiod priming. Data were analyzed in a similar way as described for the first dataset with the following exceptions: The performance time course T was computed with a 3-ISI running window. Also, the time course of inter-trial priming was assessed at each ISI with a window of 3 ISIs. That is, we computed an estimate of the performance at each ISI for all trials that were preceded either by the same ISI or by ISI-20ms or by ISI + 20 ms. These measures were taken to increase the number of trials for the analyzes and to avoid missing datapoints at some ISI bins. In addition, we computed the overall effects across the masked and unmasked conditions by computing the time courses separately and then we averaged the time courses. See Fig. 3, panels in the middle. Empirical evidence for rhythmic foreperiod priming. The figure Depicts the results of the re-analysis of three public datasets regarding rhythmicity in foreperiod priming [with N = 44 (Michel & Busch, 2023); N = 34 (Huang et al., 2015; N = 11 (Galas et al., 2025)]18,40,41. The upper panels show the estimated time courses of foreperiod priming across ISIs, the panels below show the corresponding spectra in the frequency domain in the theta and low alpha range. The dashed red line corresponds to the significance threshold (95% of surrogate data are smaller than the value of the dashed red line at each frequency). There are significant spectral power peaks at 7–8 Hz and 10 Hz. Galas et al. (2025) Data were retrieved from https://osf.io/cxkbu38,41. The corresponding study had ethical approval. Participants provided written informed consent and were compensated for their time41. In the corresponding study, participants (N = 11; an additional eight participants had been excluded in the original study, data of these excluded participants were not available for re-analysis; Mage = 23 years; range: 18-31 years, 11 out of the initial 19 participants were female; the way sex and gender were determined was not further specified; there is no information about whether data on SES, communities of descent, or race was collected41) had performed 22 sessions of the behavioral task. Each session consisted of 504 trials. In each trial, participants reported the identity of a Landolt ring (opening at the left or right side) which was either cued or uncued by an endogenous attentional cue (a line indicating the side where the target would likely occur). After the Landolt rings disappeared, there were two probes at both cued and uncued locations and participants had to additionally indicate whether the probe only contained noise or noise with an additional Gabor pattern. The probes appeared randomly with 14 equally spaced time intervals (ISIs) relative to the Landolt rings ranging from 40 to 495 ms. All responses were given at the end of the trial. For a more detailed description of the method, we refer to the original publication41. For the present analyzes, we collapsed across valid and invalid trials and across participants. Separately for probes on the left and for probes on the right, we computed at each ISI the sensitivity measure d’ as the z-transformed probability of hits (correct reports of a Gabor within the noise) minus the z-transformed probability of a false alarm (incorrect reports of a Gabor within the noise). We did this first for all trials, and secondly, only for those trials in which the ISI was repeated across trials. We then subtracted the two values to obtain a time course of foreperiod priming for d’. We then transformed the time course of foreperiod priming into the frequency domain (due to the short signal length (455 ms), we had to adjust the parameters for estimating spectral powers with the multi-taper technique: Bandwidth (W) was set to 2 Hz). Power values were then averaged for trials with probes on the left and for trials with probes on the right. Again, we conducted a permutation test by shuffling the trial identity in trial n−1—that is, the ISI in trial n−1—across all trials (see above). We repeated the analyzes steps with the shuffled labels and compared the original spectral amplitudes to the surrogates to obtain p-values. To alleviate concerns regarding non-periodic temporal structure, we applied a second reshuffling procedure22,41:We first computed the d’ priming time courses and reshuffled the resulting values across time bins (not the original trial-by-trial outcome). Surrogate power spectra were then generated from these time courses. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Results Rhythmic performance in dense sampling studies via phase-aligned peaks of perceptual efficiency A series of simulations confirms that rhythmic performance time course in dense sampling studies can be explained by the phase-alignment model. Data were drawn from a PEtc which was initially flat (0.6 hit probability at each ISI)16,21–26. In each subsequent trial, the effect of temporal orienting was modeled. That is, the phase-aligned temporal peak of the PEtc was simulated by adding a Gaussian to the flat PEtc. This Gaussian was centered at the relative temporal position of one of the peaks of a 7.5 Hz sinusoid (the sinusoid was modeled to be reset at the beginning of each trial, so that it would start in each trial with the same phase at timepoint zero). A peak of the PEtc was thus in each trial aligned to one of the peaks of the sinusoid, that one which was closest to the foreperiod (ISI) in the preceding trial n−1. When we analyzed these simulated datasets in the common way by testing for rhythmicity in performance time courses relative to the presumed reset (ISI = 0), we detected rhythmicity at 7–8 Hz in a high proportion of simulations with a range of parameters [standard deviation, σ, and height of the Gaussian (effect size) which models the effect of temporal orienting]. As is clear from the results depicted in Fig. 2 (plots on top), the number of significant results increased with increasing sample size. In addition, we observed significant results in many cases when the precision of temporal orienting was high (σ < 40 ms) and when the magnitude of the effect was large (>10%). These results show that the present-phase alignment model can in principle explain observations of rhythmicity in behavioral performance. Rhythmicity in foreperiod priming of temporal intervals uncovers phase-aligned peaks of perceptual efficiency How could the assumptions of the phase-alignment model be tested against the standard model that postulates a true alternation of peaks and troughs of perceptual resources in each single trial? Under the assumption that foreperiod priming plays a key role in orchestrating the presumed switching of phase-aligned perceptual peaks, there may be a relatively simple empirical test, because we expect rhythmicity in the time course of foreperiod priming: As we assume that peaks in the PEtc can only be exactly allocated to certain timepoints that are phase-aligned to peaks in an underlying anchor oscillation, it is clear that priming of the foreperiod will appear to be enhanced at some timepoints compared to other timepoints. That is, if the n−1 foreperiod matches a timepoint to which the peak of the PEtc can be allocated, the priming effect should be large at the corresponding interval. However, if the n−1 foreperiod does not match a timepoint to which a peak of the PEtc can be allocated—that is, if the foreperiod primes a bad phase of the corresponding oscillation—priming should not work well at this exact timepoint. In fact, adjacent, phase-aligned timepoints might profit more from priming in this case. We illustrate this rationale with a second simulation. We simulated data (n = 30,000 trials) for different σ and effect sizes of the temporal perceptual peaks. As described above, the corresponding Gaussian, which modeled this effect, was in each trial centered at the peak of the sinusoid which was closest to the foreperiod (ISI) in the preceding trial n−1. This should induce the predicted rhythmicity in foreperiod priming. We then analyzed the simulated datasets, once by testing rhythmicity in the overall performance time courses, and secondly by assessing the time course of foreperiod priming (see “Methods”). Figure 2 (lower panels) shows on overview of the percentage of significant results obtained with the foreperiod priming analysis as opposed to the conventional analysis. Depending on model parameters (effect sizes > 15% and σ < 6 ISIs), both techniques unveil rhythmic temporal structure. The foreperiod priming analysis might even be more powerful as is evident from the higher detection rates we obtained by analyzing foreperiod priming. As expected, when we simulated data with a truly rhythmic PEtc (amplitude of sinusoid 0.1; 30,000 trials), but without priming from one trial to the next, only the conventional analyzes led to above chance detection rates: Rhythmicity in inter-trial priming at the frequency bin of interest were only detected in four cases out of 100 simulated experiments, which is the below chance level of 5%. Empirical evidence for rhythmicity in foreperiod priming – re-analyses of public datasets In order to test the prediction of rhythmicity in foreperiod priming, we re-analyzed public datasets from three independent studies. One of these studies had not reported rhythmic performance with the conventional analysis technique, the other two had18,40,41. We observed significant spectral power peaks in the time course of foreperiod priming in reaction time in two cases: In the first study by Michel and Busch, our analysis revealed rhythmicity in foreperiod priming at 7.7 Hz (uncorrected p = 0.012; norm. spectral power = 3.22; 95%-CI of the surrogate distribution obtained by shuffling the n−1-SOA label across trials = [−1.20, 2,64]) and at 8.6 Hz (uncorrected p = .022; norm. spectral power = 2.73; 95%-CI of the surrogate distribution obtained by shuffling the n−1-SOA labels across trials = [−1.18, 2.59]; spectral powers exceeding the upper limit of this CI were deemed significant; notice that spectral powers were z-scored relative to the average of the surrogate distribution at each frequency bin). These results were also statistically significant after correction for multiple comparisons across the two frequency bins of interest. Figure 3, leftmost panels, provides an overview of the results with the foreperiod priming analysis. In the dataset from Huang et al., we again observed a significant spectral power peak in the Fourier transformed time course of foreperiod priming at 7.5 Hz (p = 0.002; norm. spectral power = 4.57; 95%-CI of the surrogate distribution obtained by shuffling the n−1-SOA labels across trials = [−0.98, 2,16]; spectral powers exceeding the upper limit of this CI were deemed significant; effect maintained after correction; see Fig. 3, panel in the middle). In addition, rhythmicity in foreperiod priming was seen with the perceptual sensitivity measure d’ in a recently published dataset from a highly powered study (>100,000 trials) by Galas et al.41. As can be observed in Fig. 3, we observed a highly significant spectral power peak at 10.2 Hz (p < 0.001, p = 0.004 when time course, not trial ID, was shuffled; norm. spectral amplitude = 3.47: 95%-CI of the surrogate distribution obtained by shuffling the n−1-SOA labels across trials = [−1.38 2.51]; spectral powers exceeding the upper limit of this CI were deemed significant). Notice that we controlled in this case for multiple comparisons across all frequency bins between 2 and 12 Hz because the observed peak was seen at an unexpected frequency bin10–12. We conducted an additional exploratory analysis, in which we visualized how foreperiod priming modulates the spectral power in behavioral performance time courses. That is, we sorted data by the foreperiod in trial n−1 and computed the whole performance time courses for each of those foreperiod bins. As we expect participants to allocate resources to good time points within the anchor oscillation which are close to the foreperiod in trial n−1, there may be modulations in the behavioral performance time courses surrounding the time point that coincides with the foreperiod in trial n−1. This should be visible in time-frequency analyzes of the time courses. We used the dataset from the Huang et al. study for this analysis because of the relatively large number of trials (54,400) and the length of the time window across which performance was tested in this study (0–980 ms)40,42. The analysis and the results are presented in the Supplementary Material (Supplementary Figs. 3 and 4). The result is consistent with the phase alignment model, as we observe a modulation of the spectral power associated with the ISI time bin in trial n−1 (Fig. S4). Discussion Numerous recent studies suggest that behavioral performance is rhythmic4–26. This has been shown by measuring performance in detection or discrimination tasks at various timepoints after a salient reference event that is assumed to reset perceptual oscillations11–20. The present results show that rhythmic temporal structure in behavioral performance time courses in such dense sampling studies can be explained, even when the temporal evolution of perceptual efficiency in each single trial is non-rhythmic. We show in a series of simulations that rhythmicity in overall performance time courses in dense sampling studies emerges when a singular peak in perceptual efficiency gets in each trial aligned to one of the peaks of an underlying anchor oscillation. In addition, we derive from these simulations the prediction that foreperiod priming (the temporal priming benefit when the time interval between reference event and the target is repeated from one trial to the next) alternates across time and we confirm the existence of such effects in three existing datasets from the dense sampling literature. This result is interesting because oscillatory changes in performance time courses from dense sampling studies relative to the reference have been interpreted as evidence for an underlying rhythmicity in perception, with continuously alternating phases of increased and decreased perceptual capacities13. This interpretation implies that perception is rhythmic: There are several equally spaced local maxima and local minima in PEtc in the sub-second range that are determined by peak and trough phases of some underlying oscillation11–13. Here, we argue that the relationship between brain oscillations and the PEtc may be less straightforward. In fact, assuming true rhythmicity of perceptual efficiency is not necessary to explain the observed rhythmic time course of perceptual performance in dense sampling. Even when the performance time courses in dense-sampling studies exhibit clear (maybe even nearly sinusoidal) rhythmic structure (see Fig. 1), the results can be explained by an alternative phase-alignment model: singular, short-lived peaks of perceptual efficiency align to one of the peaks of an underlying oscillation. Which of the peaks of the underlying oscillation is used depends on when participants expect relevant events. For this, the temporal occurrence of peaks and troughs in the PEtc does not have to mirror that of the underlying oscillation. Thus, perceptual resources can be temporally allocated, without being oscillatory themselves. Notice that such a model of temporal orienting mirrors what is known about the phase-alignment of eye movements. The onset of eye movements (saccades) might be statistically associated with certain phases within a neuronal oscillation (in the theta range)43–45, but it is clear that saccades do not occur in each cycle of the corresponding neuronal oscillations, as participants can easily withhold saccades for extended time periods even in demanding attentional tasks46,47. The simulations we present above corroborate the proposal that phase-alignment of short-lived peaks in the PEtc can lead to rhythmicity in performance time courses in typical dense sampling studies. We simulated datasets with empirically plausible trial counts and modeled such phase-aligned peaks of the PEtc. Rhythmicity in overall performance time courses emerged across a range of different model parameters. This confirms the predictions of the phase-alignment model. Importantly, we also propose a way of testing a strict model of true rhythmicity against the phase alignment model. Explaining rhythmic performance time courses in terms of phase alignment entails that the temporal focus of high processing efficiency within each trial changes between experimental trials. Under the assumption that these trial-to-trial shifts of the peak are, at least in part, orchestrated by priming, rhythmic effects should emerge in the time course of foreperiod priming27,28. The reason is that foreperiod priming is expected to work better when a “good” timepoint, to which peaks of perceptual resources can be aligned exactly, is primed. Only in these cases it is possible to allocate resources to the exact relative timepoint in the subsequent trial. When a timepoint is primed that falls in between two such phases, the perceptual resources cannot be exactly allocated to the corresponding timepoint. The magnitude of foreperiod priming should thus have several peaks and troughs across the observed time window. An additional series of simulations corroborates this claim (Fig. 2). We finally tested the hypothesis that foreperiod priming is rhythmic in three datasets from the dense sampling literature18,36–38,40,41. The hypothesis was confirmed in all three cases: For two studies, we observed a significant peak in the time course of foreperiod priming at 7–8 Hz with reaction time which is in accordance with prior estimates of the pace of the “attentional rhythm”13. In addition, we observed a significant 10 Hz peak in data from a recent study by Galas et al. with a detection task41. Notice that we also applied an AR surrogate method to corroborate the spectral amplitude peaks in the time courses of foreperiod priming (see Supplementary Note 1)22. This analysis shows that it is unlikely that non-rhythmic temporal structure accounts for the spectral amplitude peaks we observed in foreperiod priming. We attribute the discrepancy between the frequencies of the spectral amplitude peaks we find in the three studies to differences in the experimental design. In fact, the exact experimental procedure in the study by Galas et al. differed from the standard protocol of typical dense sampling studies in some ways, as there were several response-relevant stimuli in each trial. In addition, performance was in each trial probed at both locations, in the left visual hemifield and in the right visual hemifield11,12,41. In any case, the existence of rhythmicity in foreperiod priming together with the results from our simulations establishes the phase-alignment model as an alternative to the standard model of intrinsic/true rhythmicity. They also pave the way for future investigations. It is clear that the rhythmicity in foreperiod priming we observed is difficult to explain without assuming any form of phase alignment of peaks in processing efficiency to an underlying oscillation. First, no rhythmicity would be expected in priming if allocating resources to timepoints could be primed regardless of their temporal relationship to the phase of an underlying oscillation. One might argue that observing a stimulus in trial n−1 in a peak phase of an underlying rhythm just primes the whole rhythm also at other timepoints and rhythmic priming emerges in this way. We consider this interpretation unlikely: we computed the reference performance, against which we compare the performance in primed trials, across all trials so that the advantage in primed trials would be small, if a substantial number of trials in the reference elicited a similar priming effect. In addition, we would like to clarify that the rhythmic foreperiod priming effects we observed may only be the “tip of the iceberg”: It is unlikely that participants selected in each trial the exactly best matching peak of the anchor oscillation. It is possible that participants over- or underestimate the foreperiod from trial n−1 in a subset of trials so that a more temporally remote peak of the anchor oscillation gets selected. This is not at odds with the phase-alignment interpretation of the results, but it would decrease the magnitude of the foreperiod priming effect. Relationship to neuroimaging and electrophysiology It is important to consider how the phase-alignment model relates to converging lines of evidence from imaging and electrophysiology. There have been many studies that associate behavioral performance with the phase of pre-target brain oscillations as assessed by electroencephalography (EEG), magnetencephalography (MEG) or intracranial recordings with humans and non-human primates2–9,48–50. The results of such studies usually show a statistical relationship between performance in a behavioral task on the one hand and the phase of neuronal oscillations on the other hand. If the phase of neuronal oscillations goes along with performance, one might suspect that perceptual efficiency per se fluctuates along with the neuronal oscillation. However, an alternative interpretation in terms of an alignment between peaks of an underlying oscillation and single peaks of the PEtc is possible: If perceptual resources tend to peak at certain phases of an underlying oscillation as a result of temporal expectation/orienting, the behavioral benefits will become apparent at these phases even when peaks of the attentional resources do not consistently occur in each cycle. We think it is difficult to rule out such effects of temporal orienting with typical study designs4–10,48–50. We therefore think that major parts of the available evidence from imaging and electrophysiology are fully compatible with the phase-alignment model. Future directions The present work does not make any strong claims regarding the true profile of the PEtc. In our simulations, we modeled the PEtc as a Gaussian, but it is clear that similar results could be obtained with other, maybe physiologically more plausible parameters. Figure 4 illustrates one possibility (panels in the middle). What we call hybrid model in the middle column refers a condition where the PEtc is modeled as a product of a Gaussian and a sinusoid. In fact, it is possible to obtain similar results as in the simulations we present in Fig. 2 with such a modification (Supplementary Note 2, and Supplementary Fig. 1). The Fourier-transformed average performance time course then contains a peak at the anchor oscillation’s frequency and one corresponding to the frequency of the sinusoid in the PEtc. Thus, future studies have to investigate the true shape of the PEtc empirically: One could, for example, systematically assess the foreperiod priming effect at different ISIs relative to peaks and troughs of the presumed anchor oscillation. This should yield insights into the true profile of the PEtc (cf. Supplementary Fig. 2). Depicted is the sequence of five consecutive trials in a hypothetical dense sampling study. The black line represents in each trial, the time course of perceptual efficiency (PEtc) as conceptualized by the corresponding model. The dotted red line represents the anchor oscillation. According to the phase-alignment model (column on the left), the PEtc has a singular peak in each trial. Its temporal location is constrained by the phase of an underlying neuronal oscillation (dotted red line) and by when participants expect relevant events (here: a timepoint close to the relative timepoint of the target in the preceding trial). In contrast, according to the intrinsic rhythmicity model (column on the right), the PEtc is identified with the underlying oscillation. It unfolds across time with several peak and trough phases in each trial in a rhythmic way. Also, a hybrid model is possible: The PEtc may unfold across time like a ‘wavelet’ (hybrid model; column in the middle), that is, there may be several peaks and troughs in the PEtc within a trial, but the primary peak may still align with a specific phase within a cycle of an underlying anchor oscillation. In addition, we think that future neuroimaging and electrophysiology studies are necessary to reveal to which extent temporal orienting impacts on presumed neuronal correlates of perceptual efficiency. This could further elucidate effects of temporal orienting on spiking activity of single neurons in early sensory areas51,52. It should then be possible to investigate how these effects relate to ongoing oscillations in local field potentials. Such studies could provide further insights into which neuronal functions underly the PEtc. What is more, it should be possible to test the present model also with neuroimaging (EEG, MEG)2–9,48–50. In conditions in which the ISI (the time interval between a reference and a behaviorally relevant target) is varied across trials, we expect that a correlation between the phase of ongoing oscillations and the behavioral outcome emerges especially in trials, in which the ISI in trial n is primed by the preceding trial n−1, i.e., when the cue target interval is attended. Limitations There are several limitations of the present study that we list in this section. First of all, it is a limitation of the present study that we only used existing datasets to test the hypothesis that foreperiod priming is rhythmic18,40,41. Although we find this effect in three independent studies, future confirmatory work with new data is required. Secondly, we cannot answer the question to which extent particularities of stimulus timing may impact on the results in foreperiod priming. Interval timing follows Weber’s law over a wide range of time intervals (the discrimination threshold between time intervals is proportional to the length of the time interval)53. It is unclear how this influences the rhythmicity in foreperiod priming in dense sampling studies we present here. One may speculate that the fact that interval timing becomes less precise with increasing intervals could lead to an increase in the width of the Gaussian cone which describes the PEtc when the ISI gets longer. As a result, overall performance time courses in dense sampling studies may resemble dampened oscillations as it has been proposed recently23. Finally, it is important to note that several dense sampling studies in the literature used detection tasks in which targets were barely visible11,12. It is therefore at this point not clear how foreperiod priming could have worked in those studies, as participants might not be able to perceive ISIs in trials in which targets remain invisible. There are several possibilities. Firstly, it is possible that priming is in such studies restricted to transitions in which the target in trial n−1 is consciously perceived. In this case, the results could still be explained by priming if the peaks in the PEtc are large enough. Another possibility is that temporal priming can operate on a subliminal level, as priming works in many cases also with stimuli that are not consciously perceived46,47. Summary and conclusions The main contribution of the present work is twofold. Firstly, we present an alternative view on behavioral and electrophysiological results that have been interpreted as evidence for rhythmic perception2–26. We argue that a model of phase-alignment of perceptual efficiency is fully compatible with the existing evidence. Peaks in the temporal profile of perceptual efficiency may co-occur with peak phases of an underlying neuronal anchor oscillation, but these perceptual peaks do not need to occur in each cycle of the underlying anchor oscillation, so that perceptual resources do in fact not rhythmically alternate. This interpretation is, in our view, more than a subtle tweak. It has important implications for how we think about perception and related concepts, such as attention or working memory, as emerging from brain activity. Are these system properties of the brain strictly rhythmic, so that they could be identified with specific neuronal oscillations, or is their temporal profile evolving independently from these neuronal oscillations with only sporadic alignment? We think that the present work paves the way for future research to answer these questions. Secondly, we also present ways to test the phase-alignment model empirically. In particular, we argue that only the phase-alignment model, but not the intrinsic/true rhythmicity model, predicts rhythmicity in the time course of foreperiod priming in dense sampling studies. Importantly, our re-analyzes of public datasets also show such rhythmicity in foreperiod priming in three dense sampling studies from the recent literature18,40,41. Although future confirmatory work will have to corroborate this result, the present findings establish the phase-alignment model as a conceptual alternative to the view that perceptual resources are truly rhythmic11–26. Supplementary information Transparent Peer Review file Supplementary Information_PDF Reporting summary Acknowledgements This Study was supported by a grant to ST from the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation, https://www.dfg.de/en) – project number 436260547. The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript. Author contribution Tobias Schoeberl (T.S.) and Stefan Treue (S.T.) conceived of the idea and planned the analyses. T.S. analyzed the data and conducted the simulations. T.S. and S.T. wrote the manuscript. Peer review Peer review information Communications Psychology thanks the anonymous reviewers for their contribution to the peer review of this work. Primary Handling Editors: David Pascucci and Troby Ka-Yan Lui. A peer review file is available. Data availability The results of the present simulations are available on https://data.goettingen-research-online.de/dataverse/phase_align_202654–57. The datasets we re-analyzed in the present study can be found on public repositories from which they were retrieved (https://osf.io/b5q2u/; https://osf.io/69by5; https://osf.io/cxkbu)18,36–38,40,41. Code availability The codes used for the present simulations and for the analysis of the foreperiod priming effects are available under https://data.goettingen-research-online.de/dataverse/phase_align_202654–57. Matlab R2021a, Python (3.11.7; Spyder 5.4.3) were used for the simulations and data analysis. We used Matlab functions from https://chronux.org (Chronux version 2.12 v03)39. For conducting analysis with Brookshire’s AR surrogate method, we adapted publicly available Python code (https://github.com/gbrookshire/simulated_rhythmic_sampling). Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information The online version contains supplementary material available at 10.1038/s44271-026-00453-4. References Associated Data Supplementary Materials Transparent Peer Review file Supplementary Information_PDF Reporting summary Data Availability Statement The results of the present simulations are available on https://data.goettingen-research-online.de/dataverse/phase_align_202654–57. The datasets we re-analyzed in the present study can be found on public repositories from which they were retrieved (https://osf.io/b5q2u/; https://osf.io/69by5; https://osf.io/cxkbu)18,36–38,40,41. The codes used for the present simulations and for the analysis of the foreperiod priming effects are available under https://data.goettingen-research-online.de/dataverse/phase_align_202654–57. Matlab R2021a, Python (3.11.7; Spyder 5.4.3) were used for the simulations and data analysis. We used Matlab functions from https://chronux.org (Chronux version 2.12 v03)39. For conducting analysis with Brookshire’s AR surrogate method, we adapted publicly available Python code (https://github.com/gbrookshire/simulated_rhythmic_sampling).