At 60, I started playing piano again. My biggest frustration was not knowing how to practice. Which passage to work on, how many times, and when to come back to it. Every session felt like starting from scratch.

That changed when I came across a lecture by Dr. Molly Gebrian, a professional violist and music researcher who explains how motor learning actually works in musicians. That was the spark. I spent two years reading research on cognition, motor memory, and spaced repetition — not as an academic, but as someone trying to solve a real problem.

ModusPractica Pro is the result. It is not a scientific instrument. It is a practical tool built on scientific principles — an honest indicator of when to practice what, so that musicians can spend their time playing instead of guessing. Treat it as a guide, not absolute truth: it does not replace a teacher, and it does not measure musical artistry — only a structured estimate of when to return to a passage.

The ideas behind this app, the design of how it works, and all decisions about what to include and what to leave out are entirely mine. I have a background in technical education and programming, and have worked on this app for two years — the scheduling engine alone took the longest to get right.

The text on this page was refined with the help of AI. The code was written by AI under my full supervision and control. All scientific sources referenced here are peer-reviewed and publicly available — I make no claims beyond what those sources support. Citation of a researcher or teacher is not a collaboration, partnership, or endorsement; see About — Disclaimer.

ModusPractica Pro is an experimental software project. While its core scheduling and learning intervals are inspired by peer-reviewed cognitive and motor-learning research, the specific calibration values, thresholds, and adaptive mechanisms — including the version 1.4.2 feedback loop — are empirical solutions born from personal practice and real-world testing. The tool is designed to solve a practical musical problem, not to serve as a rigid academic instrument.

Two jobs are easy to confuse at the instrument, and they ask for opposite tools. Acquisition / encoding is writing a motor programme: choosing fingerings, isolating a two- or three-note transition, organising a shift. That is the early, highly attentive stage of skill learning described by Fitts & Posner (1967) — error is expected, because the movement has not been designed yet. Retention / retrieval is testing and reinforcing a programme that already exists: can you produce it after time away, without warming it first?

The testing effect — attempting retrieval rather than restudying — reliably improves later retention in verbal memory (Roediger & Karpicke, 2006). The same logic, applied with caution to motor skill, is the case Schmidt & Bjork (1992) and later reviews make: conditions that make practice feel harder (spacing, a delayed test) often produce better learning than conditions that make practice look fluent. Bjork’s work on metamemory (1994) is the other half: learners routinely mistake easy, current performance for lasting knowledge.

That is why this app keeps Analysis (time only, no counters) and Training (counted retrieval) apart. Strict repetition targets and FrustrationGuard are useful once you are checking a formed programme: they keep the test honest and stop a failing retrieval from turning into an error-filled grind. The same tools are counterproductive during first encoding. Trial-and-error is mandatory while you design the movement; treating those necessary misses as a “failed session” adds pressure without giving the scheduler a meaningful signal. The recommended musician workflow is therefore: encode without counting (Analysis), postpone the start date until a full start-to-finish attempt is realistic, let one or more nights of sleep consolidate the emerging trace, then retrieve cold with the counter. “Day 1 / Day 2” in the Manual is a teaching sketch, not a 24-hour deadline: schema construction and motor-programme encoding (Fitts & Posner, 1967; Sweller, 1988) do not run on a fixed clock. Intricate coordination often needs several sittings and several nights of sleep-dependent consolidation (Walker et al., 2002; Korman et al., 2007) (debated — see Sleep & Motor Consolidation above) before a stable, retrievable sequence even exists. The practical steps are in the Manual, The Two-Phase Workflow: Acquisition vs. Retention.

Entering the counted retrieval loop too early violates what the DSR-inspired scheduler assumes. Training treats each registered repetition as retrieval of a complete motor trace for the defined section. Scoring fractional fragments, a single bar of a multi-bar unit, or a half-run as “reps” feeds the model a retrieval signal for a programme that is still being assembled. Stability and the next interval then describe something the musician has not yet attempted as a whole, and FrustrationGuard may fire on ordinary encoding error. Stay in Analysis until the full section — including the transition note — can be attempted uninterrupted, even slowly.

This is a pedagogical distinction and a product design choice, not a laboratory measurement of Fitts & Posner stages. The app cannot tell whether a motor programme has been “written.” It can only stay out of the way during Analysis and treat Training as a delayed retrieval check. See also Memory Zones below.

After a failed cold start, the tempting response is to keep drilling until the passage works. That massed continuation produces a warm, transient performance that feels like progress and is a poor predictor of the next morning. In motor learning this is a documented metacognitive trap: blocked or massed practice inflates judgements of learning relative to later retention (Simon & Bjork, 2001). More generally, performance during practice is not a reliable index of learning (Soderstrom & Bjork, 2015; Schmidt & Bjork, 1992).

Sleep-dependent motor consolidation is a better place to put the remaining work than another warmed block of repetitions. Piano-type sequence studies show overnight (and even daytime nap) gains without further practice (Walker et al., 2002; Korman et al., 2007) (debated — see Sleep & Motor Consolidation above). A short diagnostic deconstruction — isolating the single failing transition at a slow tempo, then stopping — gives sleep a clean encoding episode. A long massed drill after a failed retrieval does the opposite: it encodes fluency-under-warmth and overwrites the information the cold test actually provided.

The app does not enforce this recovery. Analysis after a weak Training session logs time only; it does not rewrite the schedule a second time the same day. Whether you stop after a few slow corrections, or grind until the passage feels safe, is a practice decision. The research cited here is why the Manual recommends the former.

Waiting until one fragment is fully consolidated before opening the next is blocked practice: high fluency in the moment, weaker later retention. Practising several distinct sections of a work in a staggered or mixed sitting — cold retrieval on what is due, then a new fragment in Analysis — is closer to the contextual interference effect first shown for motor skills by Shea & Morgan (1979) and reviewed by Magill & Hall (1990). The extra effort of switching tasks during practice tends to feel worse and to retain better (Schmidt & Bjork, 1992). That is a reason to use the idle days that spacing creates on a later fragment, not to keep drilling the first one. It is not a claim that this app implements a published random-practice protocol; the Interleave screen remains a convenience that logs Analysis time only.

The opposite mistake is to start uncounted encoding on every bar of a long span in one afternoon. Cognitive load theory treats working memory as sharply limited during the construction of new schemas (Sweller, 1988). Applying that idea to first motor encoding is an extrapolation from instructional research, not a musician-specific trial — but it matches the practical limit: one or two new fragments per day can be designed with attention; eight novel two-bar units cannot. Staggering initial encoding protects that limited capacity while older fragments take their spaced retrievals and overnight consolidation (Walker et al., 2002; Korman et al., 2007) (debated — see Sleep & Motor Consolidation above). As intervals lengthen, Due Today thins and there is room to feed the next section into the pipeline. The musician-facing steps are in the Manual, Parallel & Staggered Sections.

This section keeps two layers apart on purpose. The direction of scheduling — spaced repetition, sleep consolidation, challenge-point thinking, shortening the interval after a weak retrieval — is inspired by the peer-reviewed research cited on this page. The exact numbers and some adaptive engineering around them are not: no published study hands down a ready-made motor-practice formula. They are pragmatic design choices, listed here so nothing hides behind the word "algorithm" and so this page does not read as pseudoscientific marketing.

• 85% success threshold (difficulty only). Since coldstart-v1 the session success ratio steers difficulty, not the interval: at least 85% of registered attempts correct lowers difficulty (by 0.05, or 0.02 after a long session); 60–85% raises it slightly (+0.03); below 60% it rises faster (+0.08). An unfinished session keeps difficulty unchanged in the top band. The repetition interval instead follows the cold start (next item). Research on optimal training difficulty independently identifies roughly 85% success as an effective learning zone (Wilson et al., 2019). Wilson et al. derive that figure from stochastic gradient-descent learning models on binary classification tasks (shown on artificial and biologically plausible neural networks), not from human motor practice or music; applying it to instrumental practice is therefore an extrapolation, not a direct empirical finding for musicians. Sessions saved between v2.1.0 and this fix under-count correct repetitions in accuracy history, because only the final streak was stored.
• Cold-start grades (interval). A clean first rep (entryCost 0) grows the interval fully (at least ×1.1); one error before the first clean rep (entryCost 1) caps growth (×1.0–×1.2, never shorter); two or more errors first (entryCost ≥ 2) shortens the interval like a weak session (×0.85 setback). An unfinished session caps a clean cold start to the middle grade, and so does a clean cold start followed by a weak rest of session (below 60% accuracy): a single first rep is a noisy observation — in real logs most cold starts begin with an error — so growth stays cautious when the rest of the session disagrees. Sessions without a usable entryCost (older history) keep the previous ratio-based bands. All grades and caps are design choices, not published values.
• Why the cold start decides — an honest note. A check on the developer's own data (one person, 126 consecutive session pairs) found no predictive advantage for either entryCost or the session ratio. The choice rests on (1) consistency with retention research and with what the Manual tells users about cold starts, and (2) entryCost being the only measure whose definition stayed the same across versions. Planned re-evaluation around early January 2027 with the same analysis on sessions carrying the "coldstart-v1" scheduling marker.
• Interval growth of roughly ×1.4–1.7 per clean-cold-start session at high accuracy. Deliberately more conservative than flashcard systems such as SM-2/Anki (ease factor ≈ 2.5). Motor skills are more fragile than declarative recall, and practicing slightly too often costs far less than practicing too late. In spaced-repetition scheduling the risk of a late review is forgetting before it happens; in research on gaps between study sessions, later retention drops more steeply when the gap is too short than when it is too long. This comes from verbal-learning research (Cepeda et al., 2008).
• Base setback of ×0.85 on weak sessions. Milder than a flashcard "lapse": a bad day does not erase a motor programme, so the interval is shortened, never reset. The idea of a mild setback follows motor-learning and spaced-repetition principles; the specific multiplier ×0.85 is a design choice.
• Partial success never shortens the interval — except a failed cold start. A session between 60% and 85% success keeps the interval growing slowly (at most ×1.2) and never shortens it; only a weak session (below 60%) or a cold start with two or more errors shortens the interval. This keeps the response monotone within each grade: a better session never ends with a shorter interval than a worse one. It follows the general behaviour of spaced-repetition schedulers, where a successful recall does not shorten the interval and only a lapse does. The ×1.2 cap and the 60-85% band are design choices, not published values.
• An unfinished session is not a strong session. Success is measured as accuracy (correct reps over all registered attempts), so three clean reps out of a target of seven score 100%. When a Training session ends below its rep target with 85% or higher accuracy, a clean cold start still grows the interval only as for partial success (at most ×1.2, never shorter) and difficulty stays unchanged; in the first two acquisition days difficulty is not lowered, but the day still counts. Reasoning: performance during a session is a weak indicator of what is retained (Schmidt & Bjork, 1992; Kantak & Winstein, 2012; Soderstrom & Bjork, 2015), and fewer reps than planned means less practice. Treating such a session like the partial-success band is a design choice, not a published value. A session that ends early after a failed cold start (entryCost ≥ 2) shortens the interval instead.
• The trend badge needs a clear and relevant difference. The badge compares the last 5 sessions with the previous 5 via a two-sided test for two proportions (alpha 0.05) and appears only when the difference is also at least 5 percentage points, because statistical significance is no measure of the size of an effect (Wasserstein & Lazar, 2016) and a pre-chosen smallest relevant difference is needed (Lakens, 2017). Repetitions within a session are not independent, so the test is optimistic rather than strict. The 5 percentage points are a design choice, not a published value. In the Progress view the badge compares the last 5 reached targets with the 5 before, using the same rules.
• Adaptive setback scaling (engineering, not a researched phenomenon). With enough recent training history (at least ten of the last thirty sessions), the engine compares your average success rate (same ratio as scheduling) to the ~85% learning target and softens the setback (toward roughly ×0.90–0.92) when you have been below that zone, or slightly tightens it when success is consistently very high. The same scaling adjusts FrustrationGuard stability penalties. Fewer than ten measurable sessions keep the default ×0.85. See the note below.
• Difficulty steps of ±0.02 to 0.08. Difficulty is treated as a slowly changing property of the passage; a single session never rewrites it.
• Long-session damping. Applies only in the top difficulty band (success ratio ≥ 85%) in the spaced phase (initialDaysDone ≥ 2), when elapsed time exceeds Math.max(600, targetReps × 100) seconds (ten-minute floor; about 100 s per planned repetition). Difficulty then falls by 0.02 instead of 0.05. High time cost is read as high cognitive effort, in the spirit of the challenge-point framework (Guadagnoli & Lee, 2004). The clock does not reverse stability, shorten the interval, change BPM, demote the session to a failure, or trigger FrustrationGuard. Acquisition days and sessions below 85% success are unaffected. Design choice, not a published time-on-task rule.
• Sessions without registered attempts. A training session in which no repetition was registered (no running total of correct repetitions, no errors) carries no information about how well the passage is remembered, so it is treated as missing data: it is saved for time statistics, but it does not change difficulty, stability or the due date, and the section stays due. This follows the general design of spaced-repetition schedulers, which update their memory estimate from observed recall outcomes, so an absent observation changes nothing. How the app applies that principle to counted motor practice is a design choice, not a published rule. A session with registered errors but no correct repetition is not missing data: it is scored as weak.
• FrustrationGuard limits of 5 and 8. Discussed above — design choices, not biological constants.
• Memory Zone labels (Exploration, Consolidation, Mastery, and related). Heuristic UI stages derived from the app’s scheduling lifecycle — not a validated measurement of Fitts & Posner phases, competence stages, or automaticity. See the dedicated Memory Zones section below.
• Exponential retrievability (not the later FSRS power-law curve). Retrievability is computed with a half-life exponential of elapsed days and stability (R = 0.5^t/S). Later FSRS revisions — still building on Ye et al. (2022) — switched to a power-law forgetting curve that fits large-scale review data better. ModusPractica keeps the simpler exponential form as an intentional simplification for this motor-practice engine; it is not claimed to match current FSRS mathematics.

A note on the adaptive feedback loop (version 1.4.2). While the core scheduling intervals and difficulty scaling are inspired by established models (such as FSRS and the challenge-point framework), the self-calibrating feedback loop introduced in version 1.4.2 is a pragmatic engineering choice rather than a directly researched psychological phenomenon. No published study prescribes adjusting a motor-learning setback factor based on a sliding window of one's last 30 sessions. This mechanism was added purely to bridge the gap between rigid mathematical models and human psychological friction — ensuring that a temporary, stressful dip in performance does not trigger overly punitive intervals, while still respecting the underlying principles of motor consolidation.

Two design rules keep these choices safe. First, every value is bounded: stability is clamped between 1 and 365 days, difficulty between 0.01 and 0.99 — a mis-calibrated parameter can produce a slightly suboptimal schedule, never an absurd one. Tempo is not an engine output. Second, wherever the evidence is uncertain the model errs on the cautious side: growth is damped, penalties are mild, and the worst realistic failure mode is practicing a passage slightly more often than strictly needed.

Many musicians practice the same section more than once a day — and that is entirely compatible with this app. The scheduling algorithm updates once per section per calendar day as a design choice that leaves time (including a night) for stabilisation. But a second session on the same day is not wasted.

The Analysis mode (🔍) is designed precisely for this: free exploration without performance pressure, without affecting the scheduled learning cycle. Use it to explore a passage you are not yet ready to count (the Day 1 encoding step in the two-phase workflow), to work on phrasing and expression after a cold training session, or simply for the pleasure of playing. This supports the mental representation of the music without interfering with the motor consolidation process (Chaffin et al., 2002; Gabrielsson, 2003). Do not use Analysis before Training on a due day — that warms the passage and turns the counted session into a performance check rather than a retrieval test.

The app shows Memory Zones such as Exploration, Consolidation, and Mastery (plus Maintenance and Tracking where relevant). These labels are a heuristic UI approximation of where a section sits in the app’s own scheduling lifecycle. They are not a scientifically validated diagnosis of motor-learning stage, competence level, or “how automated” a passage has become in the brain.

Broad ideas from motor learning — early acquisition versus later spaced practice, and the value of sleep consolidation — inform the product language. Classic stage models (for example Fitts & Posner’s cognitive / associative / autonomous phases, or the four stages of competence) are useful teaching metaphors; ModusPractica does not claim to detect those stages from click counters. The zone is derived from simple app state (chiefly acquisition progress via initialDaysDone, and a consolidated / Mastery flag when you mark long-term retention), not from dual-task tests, movement variability, expressive transfer, or other laboratory markers of automaticity.

Treat the zone as an honest coaching hint for the section’s place in this app’s plan. Session outcome and correct / failed counts describe how a given practice session went; the zone does not score that session and must not be read as proof of fluency or mastery in the scientific sense. See also the Manual chapter on Memory Zones.

This section matters most for readers with an academic background. The app measures correct and failed attempts, tempo relative to a target, and entry cost — the number of failed attempts before the first correct repetition. These are valid indicators for the learning phase this app is designed for. Tempo is a benchmark you set: the app does not measure whether a faster speed is ready, and it does not push BPM upward.

• The next practice tempo — whether a passage is biomechanically ready for a faster metronome mark. That decision stays with the musician’s physical awareness (see Manual Tempo Management above).
• Temporal consistency — the regularity of timing within a passage. This requires MIDI input or audio analysis and is beyond the scope of a browser-based tool without hardware integration.
• The exact state of motor memory — no algorithm can measure what happens at the synaptic level. Stability and difficulty are mathematical approximations, not direct measurements.
• True motor-learning stage (cognitive / associative / autonomous, unconscious competence, fluency, mastery in the laboratory sense). Memory Zones are heuristic scheduling labels, not a validated classification of skill acquisition.
• Expressive quality — phrasing, dynamics, musical intention. These are outside the domain of motor repetition counting entirely.
• Effort or “struggle” from the clock. Elapsed time is a stopwatch for logging and, on strong spaced Training sessions only, a conservative dampener on difficulty (see Long-session damping above). It is not a failure criterion: time pressure is deliberately absent from counting. The clock does not trigger FrustrationGuard, does not cut stability, and does not recast a successful session as a weak one.

The stability and difficulty values are a structured, evidence-informed estimate. Entry cost is similar: an app-defined proxy — a downward trend across sessions is a useful indicator, not a direct measurement of consolidation. Memory Zones are coarser still: named stages for orientation in the app, not peer-reviewed stage detection. They are useful precisely because they are consistent and honest about what they are.

• Bjork, R. A. (1994). Memory and metamemory considerations in the training of human beings. In J. Metcalfe & A. P. Shimamura (Eds.), Metacognition: Knowing about knowing (pp. 185–205). MIT Press.
• Bock, O., Thomas, M., & Grigorova, V. (2005). The effect of rest breaks on human sensorimotor adaptation. Experimental Brain Research, 163(2), 258–260.
• Bönstrup, M., Iturrate, I., Thompson, R., Cruciani, G., Censor, N., & Cohen, L. G. (2019). A rapid form of offline consolidation in skill learning. Current Biology, 29(8), 1346–1351.e4. https://doi.org/10.1016/j.cub.2019.02.049
• Brooks, E., Hendrikse, J., Buch, E. R., Censor, N., Cohen, L. G., & Coxon, J. (2026). Offline contributions to early skill learning. Trends in Cognitive Sciences.
• Carter, C., & Grahn, J. A. (2016). Optimizing music learning: Exploring how blocked and interleaved practice schedules affect advanced performance. Frontiers in Psychology, 7, 1251. https://doi.org/10.3389/fpsyg.2016.01251
• Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T., & Rohrer, D. (2006). Distributed practice in verbal recall tasks: A review and quantitative synthesis. Psychological Bulletin, 132(3), 354–380.
• Cepeda, N. J., Vul, E., Rohrer, D., Wixted, J. T., & Pashler, H. (2008). Spacing effects in learning: A temporal ridgeline of optimal retention. Psychological Science, 19(11), 1095–1102.
• Chaffin, R., Imreh, G., & Crawford, M. (2002). Practicing Perfection: Memory and Piano Performance. Lawrence Erlbaum.
• Czyż, S. H., Wójcik, A. M., & Solarská, P. (2024). The effect of contextual interference on transfer in motor learning: A systematic review and meta-analysis. Frontiers in Psychology, 15, 1377122. https://doi.org/10.3389/fpsyg.2024.1377122
• Das et al. (2025). Micro-offline gains do not reflect offline learning during early motor skill acquisition in humans. PNAS. https://doi.org/10.1073/pnas.2509233122
• Donovan, J. J., & Radosevich, D. J. (1999). A meta-analytic review of the distribution of practice effect. Journal of Applied Psychology, 84(5), 795–805.
• Duke, R. A., Simmons, A. L., & Cash, C. D. (2009). It's not how much; it's how: Characteristics of practice behavior and retention of performance skills. Journal of Research in Music Education, 56(4), 310–321.
• Fitts, P. M., & Posner, M. I. (1967). Human Performance. Brooks/Cole.
• Gabrielsson, A. (2003). Music performance research at the millennium. Psychology of Music, 31(3), 221–272.
• Gebrian, M. (2024). Learn Faster, Perform Better: A Musician's Guide to the Neuroscience of Practicing. Oxford University Press. Link: mollygebrian.com
• Guadagnoli, M. A., & Lee, T. D. (2004). Challenge point: A framework for conceptualizing the effects of various practice conditions in motor learning. Journal of Motor Behavior, 36(2), 212–224.
• Gupta, M. W., & Rickard, T. C. (2022). Dissipation of reactive inhibition is sufficient to explain post-rest improvements in motor sequence learning. npj Science of Learning. https://doi.org/10.1038/s41539-022-00140-z
• Gupta, M. W., & Rickard, T. C. (2026). Spaced practice and reactive inhibition have limited or no effects on motor sequence learning. Scientific Reports. https://doi.org/10.1038/s41598-026-52702-5
• Joëls, M., et al. (2006). Learning under stress: how does it work? Trends in Cognitive Sciences, 10(4), 152–158.
• Kantak, S. S., & Winstein, C. J. (2012). Learning–performance distinction and memory processes for motor skills: A focused review and perspective. Behavioural Brain Research, 228(1), 219–231.
• Keetch, K. M., & Lee, T. D. (2007). The effect of self-regulated and experimenter-imposed practice schedules on motor learning for tasks of varying difficulty. Research Quarterly for Exercise and Sport, 78(5), 476–486. https://doi.org/10.1080/02701367.2007.10599447
• Korman, M., et al. (2007). Daytime sleep condenses the time course of motor memory consolidation. Nature Neuroscience, 10(9), 1206–1213.
• Lakens, D. (2017). Equivalence tests: A practical primer for t tests, correlations, and meta-analyses. Social Psychological and Personality Science, 8(4), 355–362.
• Lee, T. D., & Genovese, E. D. (1988). Distribution of practice in motor skill acquisition: Learning and performance effects reconsidered. Research Quarterly for Exercise and Sport, 59(4), 277–287.
• Lee, T. D., & Magill, R. A. (1983). The locus of contextual interference in motor-skill acquisition. Journal of Experimental Psychology: Learning, Memory, and Cognition, 9(4), 730–746.
• Magill, R. A., & Hall, K. G. (1990). A review of the contextual interference effect in motor skill acquisition. Human Movement Science, 9(3–5), 241–289.
• Pan, S. C., & Rickard, T. C. (2015). Sleep and motor learning: Is there room for consolidation? Psychological Bulletin, 141(4), 812–834.
• Porter, J. M., & Magill, R. A. (2010). Systematically increasing contextual interference is beneficial for learning sport skills. Journal of Sports Sciences, 28(12), 1277–1285.
• Rickard, T. C., Cai, D. J., Rieth, C. A., Jones, J., & Ard, M. C. (2008). Sleep does not enhance motor sequence learning. Journal of Experimental Psychology: Learning, Memory, and Cognition, 34(4), 834–842.
• Roediger, H. L., III, & Karpicke, J. D. (2006). Test-enhanced learning: Taking memory tests improves long-term retention. Psychological Science, 17(3), 249–255.
• Rosenbaum, D. A. (2010). Human Motor Control (2nd ed.). Academic Press.
• Schmidt, R. A., & Bjork, R. A. (1992). New conceptualizations of practice. Psychological Science, 3(4), 207–217.
• Shea, J. B., & Morgan, R. L. (1979). Contextual interference effects on the acquisition, retention, and transfer of a motor skill. Journal of Experimental Psychology: Human Learning and Memory, 5(2), 179–187.
• Simon, D. A., & Bjork, R. A. (2001). Metacognition in motor learning. Journal of Experimental Psychology: Learning, Memory, and Cognition, 27(4), 907–912.
• Soderstrom, N. C., & Bjork, R. A. (2015). Learning versus performance: An integrative review. Perspectives on Psychological Science, 10(2), 176–199.
• Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science, 12(2), 257–285.
• Walker, M. P., et al. (2002). Practice with sleep makes perfect. Neuron, 35(1), 205–211.
• Wasserstein, R. L., & Lazar, N. A. (2016). The ASA's statement on p-values: Context, process, and purpose. The American Statistician, 70(2), 129–133.
• Wilson, R. C., Shenhav, A., Straccia, M., & Cohen, J. D. (2019). The Eighty Five Percent Rule for optimal learning. Nature Communications, 10, 4646.
• Ye, J., et al. (2022). A Stochastic Shortest Path Algorithm for Optimizing Spaced Repetition Scheduling. KDD '22.