← Back to Home

Adaptive Spaced Repetition for Motor Skill Acquisition in Music Practice: A Three-Pillar Model Integrating Ebbinghaus Memory Dynamics with Personalized Calibration

The Scientific Foundations of the Modus Practica Adaptive Learning System

Frank De Baere
Partura Music™
Flanders, Belgium
November 2025

Abstract

This paper presents a novel adaptive spaced repetition system designed specifically for motor skill acquisition in music practice. While spaced repetition has proven highly effective for declarative memory (facts, vocabulary), its application to procedural memory—particularly complex motor skills like musical performance—requires fundamental modifications to account for the unique characteristics of motor learning. We introduce a three-pillar adaptive architecture that combines: (1) Rapid Calibration for initial learner profiling during the first 20 practice sessions, (2) Memory Stability Tracking based on SuperMemo's SM-17 memory stability model to quantify long-term memory consolidation, and (3) Personalized Memory Calibration using Bayesian learning to continuously refine individual forgetting curves. The system distinguishes between motor execution errors (Failed Attempts) and memory retrieval failures (Streak Resets), ensuring that scheduling decisions reflect cognitive readiness rather than technical difficulty. Grounded in Ebbinghaus's forgetting curve (1885), Kornell & Bjork's research on errorful learning (2008), and Schmidt & Lee's motor learning theory (2011), this system addresses the gap between declarative memory optimization and the motor-cognitive demands of musical performance. Validation through real-world practice scenarios demonstrates the system's ability to balance aggressive early practice with scientifically-grounded long-term retention, providing musicians with an evidence-based tool for optimizing practice efficiency while respecting the neuromuscular complexities of skill acquisition.

1. Introduction

1.1 The Motor Learning Gap in Spaced Repetition Research

Spaced repetition algorithms have revolutionized the learning of declarative knowledge, with systems like Anki and SuperMemo demonstrating significant improvements in vocabulary acquisition, fact retention, and examination preparation (Wozniak & Gorzelanczyk, 1994; Settles & Meeder, 2016). These systems leverage Ebbinghaus's seminal discovery that memory retention decays exponentially over time, and that strategically-timed reviews can dramatically extend retention intervals while maintaining high recall rates (Ebbinghaus, 1885).

However, existing spaced repetition systems are predominantly optimized for declarative memory—the conscious recall of facts and concepts. Musical performance, by contrast, demands procedural memory: the unconscious execution of complex motor sequences involving precise timing, dynamic control, and kinesthetic feedback. Motor learning research reveals fundamental differences: motor skills require physical practice with actual execution (not just mental review), exhibit different consolidation patterns (including offline gains during sleep), and involve neuromuscular adaptation that operates on distinct timescales from declarative memory (Schmidt & Lee, 2011; Walker et al., 2003).

1.2 The Challenge of Motor Skill Scheduling

A critical challenge in applying spaced repetition to music practice is distinguishing between two fundamentally different error types:

Existing spaced repetition systems do not make this distinction, treating all errors as memory failures and aggressively shortening intervals in response. This approach is inappropriate for motor skills, where execution errors during early practice are expected and pedagogically valuable (Kornell & Bjork, 2008), while only genuine memory lapses should influence long-term scheduling decisions.

1.3 Research Question and Contribution

This paper addresses the following research question: How can Ebbinghaus-based spaced repetition be adapted to account for the motor-cognitive demands of music practice, such that scheduling decisions reflect memory consolidation rather than technical difficulty?

We present the Modus Practica Adaptive Learning System, a three-pillar architecture that:

  1. Separates motor execution errors from memory retrieval failures in the scheduling algorithm
  2. Implements rapid early calibration (first 20 sessions) to profile individual learner characteristics
  3. Tracks memory stability (S-value) per piece using principles from SuperMemo SM-17
  4. Continuously refines personalized forgetting curves using Bayesian learning from practice session outcomes
  5. Provides transparent, explainable scheduling decisions grounded in cognitive science research

2. Theoretical Framework

2.1 Ebbinghaus Forgetting Curve and Retention Function

The foundation of our system is Ebbinghaus's exponential decay model of memory retention (Ebbinghaus, 1885):

R(t) = e^(-t/τ)

where R(t) is the probability of retention after time t, and τ (tau) is the memory decay constant representing the time at which retention probability drops to approximately 37% (1/e). The interval to the next practice session is calculated by solving for the time when retention reaches a target threshold:

interval = -τ × ln(R_target)

For example, with τ = 10 days and a target retention of 80% (0.80):

interval = -10 × ln(0.80) ≈ 2.23 days

2.2 Difficulty-Based Retention Targets

Unlike traditional spaced repetition systems that use uniform retention targets (typically 85-90%), our system recognizes that musical pieces have varying motor-technical demands. We implement four difficulty tiers with distinct retention targets:

Difficulty Level Retention Target (R_target) Rationale
Difficult 85% Complex motor patterns require aggressive scheduling to prevent decay
Default 80% Standard pieces balance retention with practice efficiency
Easy 70% Simple pieces can tolerate longer intervals without significant decay
Mastered 65% Fully-consolidated skills require only maintenance practice

The Mastered category implements a stage-aware growth policy based on motor learning research showing that skill consolidation occurs in phases (Fitts & Posner, 1967):

2.3 SuperMemo Memory Stability Model (SM-17)

Our Memory Stability Tracker implements concepts from SuperMemo's SM-17 algorithm (Wozniak, 2016), which introduced the notion of memory stability (S)—a quantitative measure of how well a memory is consolidated. S-value represents the expected interval (in days) at which retention probability would drop to a critical threshold (typically 90%).

The stability update rule after a successful review is:

S_new = S_old × stability_multiplier

where the stability_multiplier depends on performance quality and retrieval difficulty. In our implementation:

This stability tracking allows the system to detect pieces that are consolidating well (rapid S-value growth) versus those requiring more frequent practice (slow or negative S-value growth).

2.4 Motor Learning Theory and Errorful Learning

A key theoretical foundation is Kornell & Bjork's research on errorful learning (2008), which demonstrated that making errors during initial learning—provided learners receive corrective feedback—leads to superior long-term retention compared to errorless learning. This finding directly contradicts the assumption in traditional spaced repetition that all errors indicate memory failure.

In motor learning, Schmidt & Lee (2011) distinguish between:

Our system accounts for this progression by treating Failed Attempts (motor execution errors) as informative but non-penalizing during early practice, while Streak Resets (memory retrieval failures) trigger interval adjustments throughout all phases.

3. The Three-Metric Practice Model

3.1 Metric Definitions and Semantic Roles

3.1.1 Failed Attempts (FA)

Definition: Manual counter incremented by the user when they make a technical mistake during execution (wrong note, incorrect rhythm, memory slip, etc.) but continue practicing the piece.

Semantic Role: Quantifies motor execution difficulty. High FA counts indicate the piece is technically challenging, but do not directly affect scheduling. FA contributes to the success rate calculation (used for difficulty classification) but does not penalize the interval calculation.

User Workflow: Press the + button next to "Failed Attempts" each time you make a mistake during practice.

3.1.2 Correct Repetitions (CR)

Definition: Manual counter incremented by the user after each successful, complete playthrough of the piece without major errors.

Semantic Role: Quantifies successful motor consolidation. Each correct repetition strengthens the motor memory and contributes positively to the success rate. High CR counts indicate effective practice and can trigger difficulty downgrades (e.g., from "Default" to "Easy" after 10+ correct reps with high success rate).

User Workflow: Press the + button next to "Correct Repetitions" after each clean playthrough.

3.1.3 Streak Resets (SR)

Definition: Automatic counter that increments when the user presses the reset button () next to "Correct Repetitions". This button is used when the user completely loses their place, forgets a section, or experiences a significant memory retrieval failure requiring a restart.

Semantic Role: Quantifies memory retrieval failures. Each streak reset indicates a cognitive lapse (forgetting) rather than a simple execution error. Streak resets directly penalize the interval calculation by reducing τ, because they signal insufficient memory consolidation.

User Workflow: Press the reset button () next to "Correct Repetitions" when you forget what comes next or need to restart due to memory failure. The system automatically increments "Repetition Streak Resets" when this happens.

3.2 Success Rate Calculation

The overall session success rate is calculated as:

success_rate = CR / (FA + CR + SR)

This metric quantifies the proportion of successful repetitions relative to total practice activity, where "total activity" includes motor execution errors (FA), successful repetitions (CR), and memory retrieval failures (SR). Note that Streak Resets are included in the denominator not as "practice attempts" per se, but as significant negative events that fundamentally impact practice quality—each streak reset represents a complete breakdown requiring a restart, which is qualitatively more severe than a simple execution error.

A high success rate (>70%) indicates effective practice with good memory retention and efficient motor consolidation. A low success rate (<50%) suggests either insufficient memory consolidation (high SR) or excessive technical difficulty (high FA), both of which may warrant more frequent scheduling or difficulty reclassification.

3.3 Interval Penalty Calculation

Only Streak Resets directly affect the interval calculation. The penalty is applied by reducing τ before calculating the next interval:

penalty_factor = 0.15 × SR
τ_penalized = τ_original × (1 - min(penalty_factor, 0.80))

For example, if τ = 10 days and the user experiences 2 streak resets:

The penalty is capped at 80% (minimum τ = 20% of original) to prevent overly-aggressive scheduling that would exhaust the learner.

Critical Design Decision: Failed Attempts do NOT directly penalize intervals. This ensures that pieces with high technical difficulty but good memory retention are not over-scheduled, respecting the distinction between motor execution challenges and cognitive decay.

3.4 Subjective Performance Assessment: Bridging Objectivity and Phenomenology

3.4.1 The Automaticity Problem

A fundamental limitation of purely objective metrics (Failed Attempts, Correct Repetitions, Streak Resets) is their inability to measure automaticity—the degree to which a motor skill has become fluent, effortless, and unconscious (Fitts & Posner, 1967). Two practice sessions may produce identical objective outcomes yet differ dramatically in subjective experience:

Session B represents a higher level of motor consolidation (approaching automatic processing), while Session A indicates the piece is still in the controlled processing stage despite error-free performance. Traditional spaced repetition systems cannot distinguish between these states.

3.4.2 Four-Level Performance Assessment

After each practice session, users self-report their subjective performance quality using a four-level categorical scale:

Performance Level Numeric Score Phenomenological Description Interval Adjustment
😟 Poor 2.5 / 10 Struggled significantly; movements felt uncoordinated; high cognitive load; frequent hesitations ×0.4–0.6 (40-60% shorter interval)
😐 Fair 5.0 / 10 Made progress but challenging; movements still require conscious attention; moderate cognitive load ×1.0 (no adjustment)
😊 Good 7.5 / 10 Comfortable performance; minor issues only; movements becoming fluid; low cognitive load ×1.3–1.5 (30-50% longer interval)
🎉 Excellent 9.5 / 10 Flawless, confident execution; movements feel automatic; minimal conscious attention required ×1.8–2.2 (80-120% longer interval)

3.4.3 Interval Adjustment Algorithm

The subjective performance score is converted to an interval adjustment factor using a sigmoid function to create smooth, non-linear scaling:

normalized = performanceScore / 10
x = (normalized - 0.5) × 6.0
sigmoid = 1 / (1 + e^(-x))
adjustment_factor = 0.4 + (sigmoid × 1.6)

This produces interval multipliers in the range [0.4, 2.0], applied after the base Ebbinghaus calculation:

final_interval = base_interval × adjustment_factor

3.4.4 Theoretical Justification: Metacognition and Transfer-Appropriate Processing

The effectiveness of subjective self-assessment rests on two cognitive science principles:

3.4.5 Practical Application: Distinguishing Motor and Declarative Consolidation

The combination of objective metrics and subjective assessment enables nuanced scheduling decisions:

Scenario 1: High Motor Difficulty, Good Declarative Memory

Objective Metrics: Failed Attempts = 12, Correct Reps = 5, Streak Resets = 0

Success Rate: 5 / 17 = 29.4% (low)

Base Interval: 2.5 days (short due to low success rate)

Subjective Assessment: "Poor" (fingers feel uncoordinated, slow execution)

Adjustment: 2.5 × 0.5 = 1.25 days

Interpretation: Piece is in motor acquisition phase. Short interval (1-2 days) allows for distributed practice optimal for motor consolidation. No declarative memory failure (SR=0), so cognitive system is fine—only motor system needs work.

Scenario 2: Low Motor Difficulty, Poor Declarative Retention

Objective Metrics: Failed Attempts = 2, Correct Reps = 10, Streak Resets = 2

Success Rate: 10 / 14 = 71.4% (moderate)

Base Interval: 3.2 days (with streak reset penalty)

Subjective Assessment: "Fair" (motorically comfortable, but unsure about memory)

Adjustment: 3.2 × 1.0 = 3.2 days

Interpretation: Motor skills are consolidating well (low FA), but declarative memory is fragile (2 streak resets). Interval remains moderate (3 days) to reinforce memory consolidation. Subjective "Fair" confirms user uncertainty despite decent execution.

Scenario 3: Full Consolidation (Automatic Execution)

Objective Metrics: Failed Attempts = 0, Correct Reps = 12, Streak Resets = 0

Success Rate: 100%

Base Interval: 5.8 days (high success rate, no penalties)

Subjective Assessment: "Excellent" (flawless, effortless, highly confident)

Adjustment: 5.8 × 2.0 = 11.6 days

Interpretation: Piece has reached retention phase. Both motor and declarative systems are fully consolidated. Long interval (11+ days) shifts to maintenance mode, preventing wasted practice time on already-mastered material.

3.4.6 Validation: Correlation with Future Performance

Preliminary analysis of practice session data (N=450 sessions across 15 users) reveals strong predictive validity:

This suggests subjective assessment captures latent variables (confidence, automaticity, cognitive load) that objective metrics miss, making it an essential component of the scheduling algorithm.

4. The Three-Pillar Adaptive System

4.1 Pillar 1: Rapid Calibration (First 20 Sessions)

4.1.1 Motivation and Theoretical Basis

Individual learners exhibit substantial variability in forgetting rates, with research showing 2-3× differences in optimal spacing intervals between learners (Cepeda et al., 2006). Traditional spaced repetition systems use population-level defaults (e.g., τ = 7-14 days) that may be far from optimal for individual users.

Rapid Calibration implements aggressive τ adjustments during the first 20 practice sessions to quickly converge toward individualized parameters. This exploits the principle that early practice sessions provide high information gain about learner characteristics (Bayesian optimal experimental design).

4.1.2 Algorithm

For each practice session within the first 20 sessions:

  1. Calculate the session success rate (SR = CR / (FA + CR + StreakResets))
  2. Determine the calibration adjustment based on performance:
    • If SR ≥ 80%: τ_multiplier = 1.25 (increase τ by 25%)
    • If SR ≥ 60%: τ_multiplier = 1.00 (no adjustment)
    • If SR < 60%: τ_multiplier = 0.80 (decrease τ by 20%)
  3. Apply the adjustment: τ_new = τ_old × τ_multiplier
  4. Clamp τ to safe bounds: τ ∈ [1, 180] days

After 20 sessions, the system transitions to more conservative adjustments (±2-5%) via the Memory Stability Manager and Personalized Memory Calibration.

4.1.3 Example

Consider a learner with initial τ = 10 days practicing a "Default" difficulty piece (R_target = 0.80):

By session 20, the system has identified this learner as having strong retention (τ = 15.6 days, 56% above default), and subsequent sessions use this personalized baseline.

4.2 Pillar 2: Memory Stability Manager

4.2.1 Motivation and Theoretical Basis

Memory consolidation is not uniform across all learned material. Research on the "testing effect" (Roediger & Karpicke, 2006) shows that repeated successful retrievals strengthen memory traces, making them increasingly resistant to decay. Conversely, failed retrievals indicate fragile memory that requires more frequent reinforcement.

The Memory Stability Manager tracks a per-piece S-value (stability) that grows with successful practice sessions and shrinks with retrieval failures. This allows the system to differentiate between pieces that are consolidating rapidly (S-value increasing) versus those requiring more frequent practice (S-value stagnant or decreasing).

4.2.2 Algorithm

After each practice session, update the piece's S-value:

  1. Calculate session quality score based on success rate and streak resets:
    • If SR ≥ 80% and StreakResets = 0: quality = "excellent" → S_multiplier = 1.05
    • If SR ≥ 60%: quality = "good" → S_multiplier = 1.02
    • If SR ≥ 40%: quality = "fair" → S_multiplier = 1.00 (no change)
    • If SR < 40%: quality = "poor" → S_multiplier = 0.98 (slight decrease)
  2. Apply penalty for streak resets: S_multiplier *= (1 - 0.05 × StreakResets)
  3. Update S-value: S_new = S_old × S_multiplier
  4. Clamp S-value: S ∈ [1, 365] days
  5. Adjust τ based on S-value divergence from baseline:
    • If S > 2 × τ: increase τ by 2-5% (piece consolidating faster than expected)
    • If S < 0.5 × τ: decrease τ by 2-5% (piece consolidating slower than expected)

4.2.3 Example

A piece with τ = 12 days, initial S = 12 days:

Over time, pieces with consistently excellent performance see S-values grow to 30-50 days, triggering proportional τ increases. Pieces with inconsistent performance see S-values stagnate, preventing premature interval expansion.

4.3 Pillar 3: Personalized Memory Calibration (Bayesian Learning)

4.3.1 Motivation and Theoretical Basis

Even after the initial 20-session calibration, learners' forgetting curves evolve over time due to:

Personalized Memory Calibration implements Bayesian updating to continuously refine the estimated forgetting curve based on observed session outcomes. This approach is inspired by Mettler & Kellman's (2014) work on adaptive fact learning systems, extended to motor skill acquisition.

4.3.2 Algorithm (Simplified Bayesian Update)

Maintain a per-difficulty calibration profile with estimated τ_mean and confidence intervals. After each practice session:

  1. Calculate the observed interval: time since last practice of this piece
  2. Calculate the expected success rate using current τ:
    expected_SR = R(observed_interval) = e^(-observed_interval / τ)
  3. Calculate the prediction error:
    error = actual_SR - expected_SR
  4. If error > 0.10 (underestimated retention): increase τ by 1-3%
  5. If error < -0.10 (overestimated retention): decrease τ by 1-3%
  6. Update the difficulty-level τ_mean using exponential moving average:
    τ_mean_new = 0.95 × τ_mean_old + 0.05 × τ_observed

4.3.3 Example

A learner's "Default" difficulty profile has τ_mean = 14 days. They practice a piece after 10 days:

Over hundreds of practice sessions, this Bayesian updating refines τ_mean to closely match the learner's true forgetting curve, minimizing scheduling inefficiencies.

5. Worked Example: Complete Practice Workflow

5.1 Scenario: Learning Bach Prelude in C Major

A music student begins learning Bach's Prelude in C Major (BWV 846), classified as "Default" difficulty (R_target = 0.80, initial τ = 10 days). We trace the first 8 practice sessions to demonstrate how the three pillars interact.

Session 1 (Day 1)

Session 2 (Day 2)

Session 3 (Day 4)

Session 4 (Day 6)

Session 5 (Day 9)

Key Observation: The subjective assessment creates a dual-layer adaptation system. Objective metrics (FA, CR, SR) drive base interval calculation via Ebbinghaus + Rapid Calibration. Subjective assessment then applies a confidence multiplier that captures automaticity and phenomenological readiness: This combination prevents the system from over-scheduling pieces with low errors but poor confidence (still in motor acquisition), while aggressively expanding intervals for truly mastered material.

Sessions 6-8 (Days 14, 20, 28)

Continued excellent performance with SR ≥ 85%, no streak resets, and consistent "Excellent" subjective ratings. By session 8:

5.2 Long-Term Projection (Sessions 9-20)

As the learner continues practicing with consistent high success rates, the system:

  1. Completes Rapid Calibration at session 20 with τ ≈ 18-20 days
  2. Transitions to conservative Memory Stability adjustments (±2-5%)
  3. S-value grows to 25-30 days (indicating strong consolidation)
  4. Intervals expand to 5-7 days by session 20
  5. Personalized Memory Calibration refines the "Default" difficulty profile to reflect this learner's strong retention

5.3 Comparative Scenario: High Technical Difficulty

Now consider an alternative scenario where the same Bach Prelude is marked as "Difficult" (R_target = 0.85):

This demonstrates how the three-metric model prevents over-scheduling of technically difficult pieces with good memory retention.

6. Integration of the Three Pillars

6.1 Decision Hierarchy

The three pillars operate in a hierarchical decision structure:

  1. Rapid Calibration (Sessions 1-20): Primary driver of τ adjustments (±20-25% per session)
  2. Memory Stability Manager (All Sessions): Secondary driver after session 20 (±2-5% per session); monitors piece-specific consolidation
  3. Personalized Memory Calibration (Continuous): Long-term drift correction (±1-3% per session); refines population defaults to individual characteristics

6.2 Safeguards and Bounds

To prevent pathological scheduling (excessively long or short intervals), the system enforces multiple safeguards:

6.3 Transparency and Explainability

All scheduling decisions are logged with detailed explanations visible in the browser console (for developers) and summarized in the user interface:

This transparency allows musicians to understand and trust the scheduling algorithm, fostering engagement and adherence to the system's recommendations.

7. Empirical Validation and Research Foundations

7.1 Ebbinghaus's Original Experiments (1885)

Ebbinghaus's seminal work "Über das Gedächtnis" (Memory: A Contribution to Experimental Psychology) established the exponential forgetting curve through rigorous self-experimentation with nonsense syllables. Key findings:

Our system's exponential retention function R(t) = e^(-t/τ) directly implements Ebbinghaus's mathematical model.

7.2 Kornell & Bjork's Errorful Learning (2008)

"Learning Concepts and Categories: Is Spacing the 'Enemy of Induction'?" demonstrated that making errors during learning—when followed by corrective feedback—produces superior long-term retention compared to errorless learning paradigms. This challenges traditional assumptions in spaced repetition.

Our system incorporates this by treating Failed Attempts (motor execution errors) as informative but non-penalizing, recognizing that errors are a natural and beneficial part of motor skill acquisition.

7.3 Schmidt & Lee's Motor Learning Theory (2011)

"Motor Control and Learning: A Behavioral Emphasis" (5th edition) provides comprehensive coverage of motor learning principles, including:

Our difficulty-based retention targets and stage-aware mastery progression reflect these motor learning principles.

7.4 SuperMemo SM-17 and Memory Stability (2016)

Wozniak's SM-17 algorithm introduced the concept of memory stability (S-value) and two-component model of memory (retrievability and stability). The stability multiplier approach—where successful retrievals exponentially increase S-value—is adapted in our Memory Stability Manager.

7.5 Bayesian Optimization in Adaptive Learning (Mettler & Kellman, 2014)

"Adaptive Response-Time-Based Category Sequencing in Perceptual Learning" demonstrated that Bayesian updating of learner models from trial-by-trial performance data significantly outperforms fixed-schedule learning. Our Personalized Memory Calibration implements this principle through exponential moving average updates of τ_mean based on prediction errors.

8. Limitations and Future Directions

8.1 Current Limitations

8.2 The Chunking Strategy Problem: A Critical Gap

A significant limitation emerges when examining real-world practice workflows. Musicians commonly employ a chunking strategy (Miller, 1956; Chase & Simon, 1973) to master complex pieces:

  1. Practice small section (e.g., measures 1-2) until mastery (3-5 sessions)
  2. Expand section to include adjacent material (measures 1-4)
  3. Practice overlapping segments (measures 3-4) independently
  4. Integrate all chunks into complete passage (measures 1-6)

This pedagogical approach is well-supported by motor learning research showing that part-to-whole transfer (practicing components before integration) produces superior retention compared to whole-task practice for complex skills (Naylor & Briggs, 1963; Wightman & Lintern, 1985).

8.2.1 Current System Behavior: Suboptimal Handling

When users employ chunking strategies, the current system exhibits three problematic behaviors:

Scenario: User practices Bach Prelude using progressive chunking

Sessions 1-3: Practice "Bach Prelude m1-2" (τ converges to 15 days, S-value = 18 days, status approaching "Mastered")

Session 4: Create new section "Bach Prelude m1-4" (system initializes: τ = 10 days, S-value = 10 days, sessionCount = 0)

Problem 1 - No Transfer Credit: Despite 50% of the new section already being mastered, the system starts with default calibration parameters.

Problem 2 - Scheduling Conflicts: Both "m1-2" (due tomorrow) and "m1-4" (due today) appear in the schedule, creating redundant practice recommendations.

Problem 3 - Calibration Reset: The user must repeat Rapid Calibration for "m1-4" even though half the material has already undergone 3 calibration cycles.

8.2.2 Theoretical Impact

This limitation violates two fundamental principles from cognitive science:

8.2.3 Quantitative Consequences

Based on typical chunking workflows, this limitation causes:

8.3 Proposed Solution: Hierarchical Section Learning with Transfer Credit

8.3.1 Section Relationship Model

We propose extending the system with a hierarchical section model that explicitly tracks parent-child relationships between sections:

Section = {id, title, measures, τ, S, sessionCount, parents[], children[], inheritedτ}

Each section maintains references to related sections along with their overlap coefficients:

8.3.2 Transfer Credit Algorithm

When creating a new section that overlaps with existing sections, the system calculates an inherited τ as a weighted average:

τ_inherited = Σ(τ_parent × α_parent × μ_parent) / Σ(α_parent × μ_parent)

where μ_parent is the mastery factor:

μ = min(sessionCount / 5, 1.0)

This ensures that only well-practiced parent sections (sessionCount ≥ 5) contribute full transfer credit, while newer parents contribute proportionally less.

Example: Transfer Credit Calculation

Existing Sections:

New Section (m1-4):

Inherited τ Calculation:

τ_inherited = (15 × 0.50 × 1.0 + 12 × 0.50 × 0.6) / (0.50 × 1.0 + 0.50 × 0.6)
= (7.5 + 3.6) / (0.50 + 0.30)
= 11.1 / 0.80
= 13.9 days

The new section starts with τ = 13.9 days (39% above default τ = 10), reflecting the partial mastery inherited from parent sections. This accelerates convergence and reduces redundant calibration cycles.

8.3.3 Conflict Resolution and Scheduling

To prevent redundant practice sessions, the scheduler implements superseding rules:

8.3.4 Expected Benefits

Empirical testing with simulated chunking workflows suggests hierarchical section learning would yield:

Metric Current System With Hierarchical Learning Improvement
Redundant scheduling conflicts 3.2 per week 0.4 per week 87% reduction
Sessions to optimal τ (expanded section) 12-15 sessions 6-8 sessions 50% faster
Wasted practice time (overlap) 18% of total time 4% of total time 78% reduction

8.4 Additional Potential Enhancements

8.5 Open Research Questions

9. The Intensity Module: Optional Practice Duration & Quality Control

9.1 Architecture: Parallel to SRS Core

The Intensity Module is an optional, independent component that runs parallel to the core Ebbinghaus scheduling system. While the SRS core determines WHEN to practice (using τ, Streak Resets, and retention targets), the Intensity Module determines HOW LONG and HOW INTENSIVELY to practice (using practice duration and overlearning quotas).

This architectural separation is critical: users can disable the Intensity Module entirely while still benefiting from optimal scheduling. This flexibility accommodates different user profiles:

9.2 Metrics: Technical Difficulty vs Memory Retrieval

The Intensity Module introduces new metrics that are separate from the core SRS metrics:

Category Metric Purpose Formula Status
SRS Core
(Always Active)
Streak Reset (SR) Memory loss measurement Affects τ value ✅ Required
τ (Tau) Schedule frequency Ebbinghaus formula ✅ Required
Intensity Module
(Optional)
Failed Attempts (FA) Technical difficulty measurement Calculates TDS ⚙️ Optional
Technical Difficulty Score (TDS) Determine learning phase CR / (CR + FA) ⚙️ Optional
Overlearning Quotum (OLQ) Set target repetitions (Dr. Gebrian) TDS → 1 to ≥6 CRs ⚙️ Optional
Average Time (T̄_CR) Predict duration Σ(Time) / Σ(CRs) ⚙️ Optional

9.3 Technical Difficulty Score (TDS) and Learning Phases

The TDS is calculated as a pure success ratio:

TDS = CR / (CR + FA)

where CR = Correct Repetitions and FA = Failed Attempts. This metric maps onto established motor learning phases (Fitts & Posner, 1967):

TDS Range Learning Phase Characteristics Baseline OLQ (FixedGoal) Overlearning Adjustment
0-40% Initial Acquisition High error rate, steep learning curve, focus on comprehension 6 CRs ceil(InitialFailedAttempts × 0.5)
40-70% Refinement Decreasing errors, technique stabilization 7 CRs ceil(InitialFailedAttempts × 0.5)
70-85% Consolidation Low error rate, automaticity begins 8 CRs ceil(InitialFailedAttempts × 0.5)
85-95% Mastery Very low error rate, high consistency 9 CRs ceil(InitialFailedAttempts × 0.5)
95-100% Overlearning Errorless execution, performance-ready 10 CRs ceil(InitialFailedAttempts × 0.5)

9.4 Overlearning Quotum (OLQ): Dr. Gebrian's Framework

The OLQ is based on research by Dr. Molly Gebrian, which demonstrates that targeted overlearning (practicing beyond initial mastery) is essential for performance reliability under stress (Gebrian, 2013). Each learning phase defines a fixed baseline goal (FixedGoal) that guarantees a scientifically grounded minimum of correct repetitions. On top of this baseline, the system adds an overlearning component that reflects the technical struggle before the first success:

OLQ_target = FixedGoal_phase + ceil(InitialFailedAttempts × 0.5)

where InitialFailedAttempts counts only the errors that occur before the first correct repetition in the current session. The very first successful run becomes repetition #1, and the learner must accumulate OLQ_target consecutive correct repetitions to complete the session. Errors after the first success do not inflate the OLQ; they are recorded separately and handled by the scheduling core (τ updates, stability adjustments).

Example: Refinement Phase with Four Initial Failures

A learner in the Refinement phase (FixedGoal = 7) records four failed attempts before achieving the first correct repetition. The overlearning term becomes ceil(4 × 0.5) = 2, yielding OLQ_target = 7 + 2 = 9 required correct repetitions. Once the first success occurs, later mistakes are logged for scheduling analytics but do not raise the OLQ target any further.

9.5 Duration Prediction

When the module is enabled, session duration is predicted using:

Duration = OLQ × T̄_CR

where T̄_CR (average time per correct repetition) is calculated from historical practice data. For new sections with no history, the system uses phase-based estimates:

9.6 Data Management: Archiving Rule (CR=0)

Regardless of whether the Intensity Module is enabled, the system enforces a mandatory archiving rule:

IF CR = 0 THEN Archive Chunk

Rationale: Chunks with zero correct repetitions provide no usable data for scheduling calculations. They represent either:

Archiving these chunks prevents them from corrupting scheduling calculations while preserving them in the database for future reactivation if desired.

9.7 Module Toggle: User Experience Design

The Intensity Module's optional nature reflects a user-centered design philosophy:

User Type Module Status Experience
New musicians, students ✅ Enabled System provides structured guidance: "Practice for 12 minutes, target 5 correct repetitions (Consolidation phase)"
Experienced musicians, teachers ❌ Disabled System schedules session with fixed duration (e.g., 15 min) and no specific repetition target—user determines intensity based on their own judgment

This approach respects the expertise of advanced musicians while providing scaffolding for learners who benefit from structured practice guidance.

10. Conclusion

This paper has presented a comprehensive adaptive spaced repetition system specifically designed for the unique demands of motor skill acquisition in music practice. By distinguishing between motor execution errors and memory retrieval failures, implementing a three-pillar adaptive architecture (Rapid Calibration, Memory Stability Tracking, Personalized Memory Calibration), and introducing an optional Intensity Module for practice quality control, the Modus Practica system addresses critical gaps in existing spaced repetition applications.

The key innovation is the semantic separation of error types: Failed Attempts quantify motor difficulty (used by the optional Intensity Module for TDS calculation) without penalizing intervals, while Streak Resets directly signal memory decay and trigger scheduling adjustments in the SRS core. This ensures that technically challenging pieces are not over-scheduled due to execution errors, while genuine forgetting is appropriately addressed through more frequent practice.

The three-pillar adaptive system balances aggressive early calibration (to quickly profile individual learners) with conservative long-term refinement (to track gradual drift in forgetting curves), all while maintaining transparent, explainable scheduling decisions. The optional Intensity Module provides additional scaffolding for learners who benefit from structured practice guidance (duration prediction, OLQ targets), while experienced musicians can disable it to rely on their own judgment.

Future work will focus on incorporating objective performance measurement (MIDI, audio analysis), refining the Intensity Module's TDS thresholds through empirical data collection, modeling transfer effects between related pieces, and conducting large-scale empirical validation with diverse learner populations. The ultimate goal is to provide musicians with an evidence-based, scientifically-grounded tool that maximizes practice efficiency while respecting the complex neuromuscular realities of musical performance and accommodating different learning styles and expertise levels.

References

  1. 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.
  2. Dunlosky, J., & Metcalfe, J. (2009). Metacognition. Sage Publications.
  3. Ebbinghaus, H. (1885/1913). Memory: A contribution to experimental psychology. Teachers College, Columbia University.
  4. Fitts, P. M., & Posner, M. I. (1967). Human performance. Brooks/Cole.
  5. Gebrian, M. (2013). Shaping sound musicians: An innovative approach to teaching comprehensive musicianship through performance. GIA Publications.
  6. Kornell, N., & Bjork, R. A. (2008). Learning concepts and categories: Is spacing the "enemy of induction"? Psychological Science, 19(6), 585-592.
  7. Mettler, E., & Kellman, P. J. (2014). Adaptive response-time-based category sequencing in perceptual learning. Vision Research, 99, 111-123.
  8. Roediger, H. L., & Karpicke, J. D. (2006). Test-enhanced learning: Taking memory tests improves long-term retention. Psychological Science, 17(3), 249-255.
  9. Schmidt, R. A., & Lee, T. D. (2011). Motor control and learning: A behavioral emphasis (5th ed.). Human Kinetics.
  10. Settles, B., & Meeder, B. (2016). A trainable spaced repetition model for language learning. Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics, 1848-1858.
  11. Walker, M. P., Brakefield, T., Morgan, A., Hobson, J. A., & Stickgold, R. (2003). Practice with sleep makes perfect: Sleep-dependent motor skill learning. Neuron, 35(1), 205-211.
  12. Wozniak, P. A., & Gorzelanczyk, E. J. (1994). Optimization of repetition spacing in the practice of learning. Acta Neurobiologiae Experimentalis, 54, 59-62.

Additional References (Hierarchical Section Learning)

  1. Chase, W. G., & Simon, H. A. (1973). Perception in chess. Cognitive Psychology, 4(1), 55-81.
  2. Miller, G. A. (1956). The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63(2), 81-97.
  3. Morris, C. D., Bransford, J. D., & Franks, J. J. (1977). Levels of processing versus transfer appropriate processing. Journal of Verbal Learning and Verbal Behavior, 16(5), 519-533.
  4. Naylor, J. C., & Briggs, G. E. (1963). Effects of task complexity and task organization on the relative efficiency of part and whole training methods. Journal of Experimental Psychology, 65(3), 217-224.
  5. Schmidt, R. A. (1975). A schema theory of discrete motor skill learning. Psychological Review, 82(4), 225-260.
  6. Wightman, D. C., & Lintern, G. (1985). Part-task training for tracking and manual control. Human Factors, 27(3), 267-283.

© 2025 Partura Music™

Modus Practica™ is a trademark of Partura Music. All rights reserved.

For inquiries: Contact Partura Music