Retrieval Practice Generator

The size of this effect is well established. Rowland's (2014) meta-analysis of 159 experimental comparisons found a mean effect size of d = 0.50 for testing versus restudy — a meaningful advantage that holds across free recall, cued recall, and recognition formats. Critically, free recall (where no cues are provided) produces the largest gains because it demands the most effortful reconstruction, consistent with the desirable difficulties framework (Bjork, 1994).

Karpicke and Roediger (2008) demonstrated the mechanism in controlled conditions: students who were repeatedly tested retained 80% of material after one week compared to 36% for students who spent the same time re-studying. The testing group initially felt less confident — retrieval feels harder — yet their long-term retention was dramatically superior. This mismatch between perceived effort and actual learning is precisely why retrieval practice must be structured by the teacher rather than left to student choice.

These effects transfer to authentic classroom settings. Agarwal, Bain, and Chamberlain (2012) found that low-stakes quizzing in real school contexts improved both factual recall and higher-order application, with benefits persisting to end-of-term assessments. Dunlosky et al. (2013) rated practice testing as a high-utility learning strategy — one of only two techniques (alongside distributed practice) to earn that rating across their comprehensive review of ten common study strategies.

For Music Technology specifically, the implication is clear: students who are regularly asked to recall signal flow, parameter relationships, and process descriptions from memory will retain that knowledge more durably than students who re-read notes or watch demonstrations. The skill calibrates difficulty by adjusting the ratio of free recall to cued recall to recognition — newer or weaker topics receive more scaffolded cues, while consolidated topics demand unaided reconstruction.

Input Schema

Prompt

You are helping an A-Level Music Technology teacher generate retrieval practice questions for a 5-minute low-stakes starter activity. The questions must force genuine reconstruction from memory, not passive recognition.

Topic: {{topic}}
Student level: {{student_level}}
Number of questions: {{question_count}}
{{#if edexcel_section}}Edexcel spec section: {{edexcel_section}}{{/if}}
{{#if time_since_learning}}Time since learning: {{time_since_learning}}{{/if}}
{{#if known_misconceptions}}Known misconceptions: {{known_misconceptions}}{{/if}}
{{#if srs_box}}SRS box level: {{srs_box}}{{/if}}

Generate exactly {{question_count}} retrieval practice questions using three retrieval modes:

**Free recall** — No cues. The student must reconstruct the answer entirely from memory. These produce the strongest learning effect but are the hardest. Use for process descriptions, signal flow explanations, and parameter relationships.
Examples of the format:
- "Explain what a compressor's attack time controls and how it affects the output signal."
- "Describe the signal flow from a microphone to a DAW input, naming every stage."
- "What happens to a sound when you apply a high-pass filter? Explain the frequency response."

**Cued recall** — Partial information provided. The student fills gaps. Easier than free recall but still requires reconstruction. Use for effects chains, parameter tables, frequency ranges, and incomplete diagrams.
Examples of the format:
- "Complete this signal chain: mic → ___ → ___ → DAW input"
- "A parametric EQ band has three controls: frequency, ___, and ___."
- "Fill in the missing frequency ranges: Sub-bass: 20-60 Hz, Bass: ___, Low-mids: ___, Mids: 1-4 kHz"

**Recognition (MCQ)** — The answer is present among options. Lowest retrieval effort. Use sparingly and only when the alternatives are plausible distractors that reveal misconceptions.
Examples of the format:
- "Which EQ type allows you to adjust frequency, gain, AND bandwidth? A) Shelving EQ  B) Graphic EQ  C) Parametric EQ  D) High-pass filter"
- "A cardioid polar pattern rejects sound primarily from: A) The front  B) The sides  C) The rear  D) Above"

**Calibrating the mix:**
{{#if srs_box}}
- SRS Box 1-2 (recently learned or struggling): Weight toward cued recall (50%), with some free recall (25%) and recognition (25%). Students need scaffolding.
- SRS Box 3 (developing): Balanced mix — roughly equal free recall (40%), cued recall (40%), recognition (20%).
- SRS Box 4-5 (consolidating or secure): Weight toward free recall (60%), with cued recall (30%) and minimal recognition (10%). Students should be reconstructing without support.
Use the provided SRS box level ({{srs_box}}) to set the ratio.
{{else}}
{{#if time_since_learning}}
- If taught within the last week: more cued recall and recognition (students are still encoding).
- If taught 1-4 weeks ago: balanced mix.
- If taught over a month ago: weight toward free recall (retrieval effort should be high for consolidated material).
{{else}}
- Default to a balanced mix: approximately 40% free recall, 40% cued recall, 20% recognition.
{{/if}}
{{/if}}

{{#if known_misconceptions}}
**Targeting misconceptions:** Design at least two questions that specifically surface these known misconceptions: {{known_misconceptions}}. Use MCQ distractors or cued recall gaps that students holding these misconceptions would answer incorrectly. The question should reveal the misconception, not trick the student.
{{/if}}

**Output format:**

### Retrieval Practice: {{topic}}
**Spec alignment:** {{edexcel_section}}
**Target level:** {{student_level}}
**Recommended time:** 5 minutes (students write answers on mini-whiteboards or paper)

#### Questions

For each question, provide:
1. The question text
2. The retrieval mode label: [Free Recall], [Cued Recall], or [Recognition]
3. `spec_alignment`: the Edexcel specification reference(s) this question addresses (e.g., 1.9 Dynamic processing, 1.11 Equalisation)

#### Answer Notes

For each question, provide:
1. The correct answer or acceptable answer range
2. Common errors or misconceptions this question is designed to surface
3. A brief note on what a wrong answer likely indicates about the student's understanding

#### Spacing Recommendation

Based on the topic and student level, recommend when to re-use these questions or generate a follow-up set. Reference the spacing effect: the optimal gap between retrieval attempts increases as mastery develops.

#### 5-Minute Starter Script

A brief implementation guide for how to run this as a lesson-opening retrieval starter:
- How to present (board, printed slips, projected)
- Timing breakdown (e.g., 3 minutes writing, 1 minute peer discussion, 1 minute teacher review)
- What to do with students who finish early
- How to use the results formatively (which questions to discuss, what patterns to watch for)

**Use UK English throughout** (analyse, equalisation, colour, centre, programme, behaviour, defence, licence/practice distinction).

**Self-check before responding:** Verify that (1) every question maps to a specific Edexcel Component 4 specification reference, (2) no question can be answered by recognition alone unless explicitly labelled as Recognition/MCQ, (3) free recall questions genuinely require reconstruction without embedded cues that give away the answer, (4) all terminology matches Edexcel command words and subject vocabulary, and (5) no debunked frameworks (learning styles, VAK, left-brain/right-brain) are referenced anywhere in the output.

Example Output

Input: Topic: Equalisation (EQ), Edexcel section: 1.11, Student level: Year 12 (Lower Sixth, taught EQ fundamentals and practical application in Ableton Live), Question count: 8, Time since learning: 10 days, Known misconceptions: ["boosting EQ makes the overall mix louder", "cutting frequencies removes them permanently from the audio file"]

Output:

Retrieval Practice: Equalisation (EQ)

Spec alignment: 1.11 Equalisation Target level: Year 12 (Lower Sixth), 10 days post-teaching Recommended time: 5 minutes (mini-whiteboards or paper)

Questions

Q1. Explain the subtractive EQ approach and why it is generally preferred over additive EQ in mixing. [Free Recall] spec_alignment: 1.11 Equalisation — subtractive vs additive approaches

Q2. Describe the relationship between Q factor and bandwidth on a parametric EQ. What happens to the affected frequency range as the Q value increases? [Free Recall] spec_alignment: 1.11 Equalisation — parametric EQ parameters

Q3. Explain why a mix engineer might apply a high-pass filter to a vocal recording. What problem does it solve and at approximately what frequency would you typically set the cutoff? [Free Recall] spec_alignment: 1.11 Equalisation — filter types and applications

Q4. Complete the standard frequency band table:

[Cued Recall] spec_alignment: 1.11 Equalisation — frequency spectrum and band identification

Q5. A vocalist's recording sounds muddy and lacks clarity. Complete this EQ processing chain: Identify the problem frequency range → apply a ___ (cut/boost) in the ___ Hz region → optionally add a gentle ___ in the ___ kHz region for presence. [Cued Recall] spec_alignment: 1.11 Equalisation — corrective EQ techniques

Q6. A parametric EQ band has three adjustable parameters: frequency, ___, and ___. [Cued Recall] spec_alignment: 1.11 Equalisation — parametric EQ controls

Q7. Which type of EQ allows independent adjustment of frequency, gain, AND bandwidth for each band? A) Shelving EQ B) Graphic EQ C) Parametric EQ D) High-pass filter [Recognition] spec_alignment: 1.11 Equalisation — EQ types and their characteristics

Q8. The "mud" frequency range that engineers often cut to improve clarity sits approximately at: A) 80-120 Hz B) 200-500 Hz C) 1-3 kHz D) 8-12 kHz [Recognition] spec_alignment: 1.11 Equalisation — frequency ranges and common mixing decisions

Answer Notes

Q1. Subtractive EQ removes unwanted frequencies rather than boosting desired ones. It is preferred because: (a) cutting reduces the risk of clipping and distortion, (b) it creates space for other instruments in the mix rather than competing for headroom, and (c) the ear perceives cuts more naturally than boosts. Common misconception to surface: Students who write "boost the frequencies you want to hear" likely hold the "boost = louder = better" misconception. This reveals a misunderstanding of gain staging and headroom management.

Q2. Q factor (quality factor) is inversely related to bandwidth. As Q increases, the bandwidth narrows — a high Q value affects a very narrow range of frequencies (surgical EQ), while a low Q value affects a broad range (gentle, musical shaping). Common error: Students frequently describe Q as "how loud the EQ is" or confuse it with gain. If they cannot articulate the bandwidth relationship, they likely learned the parameter name without understanding its function.

Q3. A high-pass filter removes low-frequency content below the cutoff point — typically rumble, handling noise, proximity effect, and electrical hum. On a vocal, a cutoff around 80-120 Hz removes these problems without affecting the fundamental frequencies of the voice. Common error: Students who say "it makes the voice higher" are confusing filtering with pitch shifting. Students who suggest a cutoff above 300 Hz are likely to thin out the vocal body.

Q4. Bass: 60-250 Hz | Low-mids: 250 Hz-1 kHz | Upper-mids/Presence: 4-8 kHz | Brilliance/Air: 8-20 kHz. Accept reasonable variations in boundary frequencies (exact boundaries are conventions, not fixed rules). Common error: Students who leave blanks or swap ranges likely learned frequency bands by recognition (seeing them on an EQ display) rather than by recall. The gap is exposure without encoding.

Q5. Apply a cut in the 200-500 Hz region; optionally add a gentle boost in the 2-5 kHz region for presence. Misconception target: Students who write "boost" for the muddy region are demonstrating the additive misconception — reaching for the boost knob when the subtractive approach is more appropriate. Students who suggest "cutting frequencies removes them permanently from the audio file" should be corrected: EQ processing is non-destructive when applied as a plugin on a DAW channel.

Q6. Gain and Q (bandwidth). Accept "gain/level/amount" and "Q/bandwidth/width." Common error: Students who write "pan" or "volume" are confusing EQ parameters with channel strip controls.

Q7. C) Parametric EQ. Distractor analysis: (A) shelving EQ does not offer bandwidth control, (B) graphic EQ has fixed frequency bands and no bandwidth adjustment per band, (D) a high-pass filter is a filter type, not an EQ type with adjustable gain/Q.

Q8. B) 200-500 Hz. This is the region commonly described as "muddy" or "boxy." Distractor analysis: (A) is the sub-bass/rumble region, (C) is the presence range, (D) is the air/brilliance range. Students who select A are likely confusing mud with rumble.

Spacing Recommendation

This set was designed for use 10 days after initial teaching — an appropriate first retrieval interval for Year 12 students who have had one practical session with EQ. Based on expanding retrieval schedules:

• Next retrieval: 2-3 weeks from now. Generate a new set with more free recall and fewer cued recall questions. By this point, students should reconstruct frequency bands and EQ types without scaffolding.
• Third retrieval: 4-6 weeks after initial teaching. Shift to predominantly free recall. Interleave EQ questions with dynamic processing or synthesis questions to strengthen discrimination.
• Ongoing: If students are in SRS Box 3 or above for this topic, retrieval starters every 3-4 weeks are sufficient for maintenance. If any student drops to Box 1, schedule an immediate re-teach session before re-assessing.

5-Minute Starter Script

1. 0:00-0:30 — Setup. Project questions on the board (or distribute printed slips face-down). Instruct students to answer from memory — no notes, no DAW, no devices. Mini-whiteboards work well; paper is fine.
2. 0:30-3:30 — Silent writing. Students write answers individually. Circulate and observe but do not give hints. Note which questions produce blank responses — these indicate retrieval failure, not laziness, and are the most diagnostically useful.
3. 3:30-4:15 — Peer check. Students compare answers with a neighbour. This generates vocabulary: students who wrote a correct but vague answer hear their neighbour's more precise version. Where answers disagree, both students mark the question for class discussion.
4. 4:15-5:00 — Teacher review. Address 2-3 questions only — prioritise those where you observed the most blank responses or disagreements. Reveal the answer, name the misconception it was designed to surface, and briefly explain why. Do not review all eight questions; the unresolved ones become motivation for the next retrieval session.
5. Early finishers. Students who complete all questions before 3:30 should add detail to their free recall answers — "Can you add an example of when you would use a narrow Q versus a wide Q?" Adding detail is productive retrieval; sitting idle is not.

Classroom Notes

I have been running retrieval starters with my Lower Sixth and Upper Sixth classes since early March 2026. These are observations from using AI-generated retrieval questions in real lessons.

• Five-minute retrieval starters at lesson opening have become the most consistent routine in my teaching. Students now expect them and settle faster because there is something to do immediately. The transition from corridor to focused work happens in under a minute.

• Free recall questions on signal flow produce the best class discussions by a significant margin. When I asked "Describe the signal path from microphone to DAW input," the range of responses revealed exactly who understood gain staging and who was guessing. Three students omitted the audio interface entirely. That single question told me more than ten minutes of Q&A would have.

• MCQ questions feel too easy for Upper Sixth students who have been studying the subject for over a year. They can often eliminate distractors by recognition even when they cannot reconstruct the knowledge from memory. For Year 13, I now weight heavily toward free recall and cued recall — recognition questions are only useful when the distractors are specifically designed to surface a known misconception.

• Students who struggle with free recall often have vocabulary gaps rather than knowledge gaps. They know what a compressor does — they can demonstrate it on a DAW — but they cannot articulate "the attack time controls how quickly the compressor begins to reduce gain after the signal exceeds the threshold." The retrieval practice surfaces this gap, and the peer discussion phase helps them acquire the precise language. This is particularly important for Component 4, where the written exam demands specific terminology.

• Eight questions in five minutes is tight. Six is more realistic for a genuine retrieval exercise where students are writing full sentences for free recall. I now generate eight but tell students to attempt at least six. The final two act as extension for faster workers rather than a source of time pressure.

• The cued recall format — "Complete this signal chain: mic to ___ to ___ to DAW input" — works exceptionally well for lower-confidence students. It gives them enough scaffolding to attempt the question without making it trivially easy. I have observed students who leave free recall questions blank successfully complete the cued version of the same knowledge, which confirms the cue is activating a trace they could not access unaided.

• Students who finish early should add detail, not sit idle. I now explicitly instruct: "If you have finished, go back to your free recall answers and add one more sentence of detail to each." This keeps the retrieval effort going without requiring new questions.

• The most common adjustment I made after first use was reducing the number of recognition questions. My initial sets had three MCQs in a set of eight. I dropped to one or two and replaced them with cued recall. The MCQs were not generating the retrieval effort I wanted — students were pattern-matching rather than reconstructing.

Integration Guidance

Feeds into:

• Assessment module: Generated questions can be exported directly as markdown quiz files compatible with the assessment module. Deploy using node scripts/deploy-assessment.js "path/to/quiz.md" --live for timed digital delivery. Note that free recall questions work better on paper than as timed digital assessments — see Known Limitations.
• SRS system: Each question's spec_alignment maps to SRS topic keys. The topic key mapping is: eq (1.11), dynamic-processing (1.9), synthesis (1.3), sampling (2.4), delay (1.12), automation (1.14), pitch-rhythm-correction (1.15), cables-signals (1.1-1.2), numeracy (2.1-2.3). Quiz results feed back into the SRS schedule to update box placement.

Feeds from:

• SRS box data: The srs_box input parameter can be populated directly from the srs_schedule table. Box 1 = recently failed or newly introduced (more cued recall), Box 2 = developing, Box 3 = secure, Box 4-5 = consolidated (more free recall). This automates the difficulty calibration.
• Assessment results: The known_misconceptions input can be populated from assessment module data — specifically, from common wrong answers in previous quizzes on the same topic. If three students selected "shelving EQ" when asked to identify a parametric EQ, that misconception should be fed into the next retrieval set.

SRS connection: Output questions are tagged with topic keys that map directly to the SRS srs_schedule table. When a retrieval starter is used in class, the results (correct/incorrect per student per topic) can update box placement via the compute_srs_schedule() RPC. The spacing recommendation in the output aligns with Leitner box intervals: Box 1 topics appear within days, Box 5 topics appear at 3-4 week intervals.

Assessment connection: The question format is designed to be compatible with the assessment module's markdown quiz format. Recognition (MCQ) questions translate directly. Cued recall questions can be formatted as short-answer with model answers for auto-marking. Free recall questions require teacher marking and are better suited to in-class paper use.

Known Limitations

• Cannot verify against a specific Edexcel spec edition. The skill generates questions aligned to spec section numbers (e.g., 1.9, 1.11) based on the current Edexcel Component 4 specification, but it cannot confirm that the wording or scope of a spec reference has not changed in a new edition. The teacher must verify alignment against their current copy of the specification.

• Practical DAW skills cannot be assessed through retrieval questions. Music Technology has a significant practical component — students must operate a DAW, apply processing, record audio, and produce a mix. Retrieval practice questions assess declarative knowledge (what a compressor does, what parameters it has) but not procedural fluency (can the student actually set up a compressor in Ableton Live and adjust it by ear). Retrieval starters complement practical work; they do not replace it.

• Free recall questions work poorly as timed digital assessments. When deployed through the assessment module as a timed digital quiz, students rush free recall answers and produce bullet fragments rather than coherent explanations. The retrieval benefit comes from the effort of reconstruction, which requires thinking time that a countdown timer undermines. Free recall questions are best used in-class on paper or whiteboards, where the teacher controls pacing and the peer discussion phase adds value. Reserve digital deployment for cued recall and recognition formats.

• Question quality depends on input specificity. Vague inputs like "topic: EQ" without a student level or time-since-learning produce generic questions that may not match the class's actual position in the curriculum. The more context provided, the better the output. Providing known_misconceptions from previous assessment data significantly improves targeting.

• Cannot adapt mid-session. The skill generates a fixed set before the lesson. If a teacher discovers during the starter that students are struggling more (or less) than expected, the questions cannot be adjusted in real time. The teacher must use professional judgement to skip questions, add verbal scaffolding, or extend discussion on a single question rather than covering the full set.