Skip to content

Game-Based Mastery Learning

Pattern · All patterns
Evidence · 24 claims (13 for, 10 mixed, 1 against) · 30 studies (9 causal, 9 review, 7 quant-synthesis, 3 theoretical, 1 qualitative, 1 design), q1–q4 · 5 of 30 report an effect size · 18 claims rest on one study

Description and scope

A reusable policy for practice-heavy, cumulative skills in which progress through levels depends on reaching a stated criterion, and the game layer (levels, a progress map, points or XP, streaks) reports that progress rather than paying for activity: short lessons of attempt → immediate feedback → retry of missed items; a level opens only when a check on the target is passed; earlier levels come back in later sessions; points and badges are given for accuracy and improvement. The intended change is in what learners can do on a task outside the app (immediate-performance, delayed-retention, general-achievement), with continued practice over weeks (persistence) as a second, separately measured outcome. It suits verbal-association targets (vocabulary, notation), fact fluency and procedures drilled to accuracy; it does not suit open-ended, whole-task judgement.

It sits inside two converted pages and does not repeat their models. The Mastery Learning pattern owns the gate-correct-recheck loop: the unit check, the corrective step, the parallel recheck and the attempt limit. The Gamification principle owns what a game layer added to instruction should reward, how scores are shared and how theme and decoration are kept away from the moment of practice. This page owns their combination: the gate as the game's level structure, so that a level, a badge or a point means a criterion passed. Learning through a whole game, where the game is the activity, belongs to the Game-based Learning principle.

No claim in this wiki tests this combination as a whole, against the same practice without a game layer or against conventional pacing. The claims that bear on it are of four kinds: a meta-analysis of mastery programmes without a game layer (mastery learning improves outcomes [+M]); a meta-analysis of game layers without a mastery gate (gamification has small positive effects [+M]); one analysis of a commercial gamified language app's level system, reported as four claims from one study (the Duolingo case study, below); and the reward experiments that limit what points should be given for (expected tangible rewards reduce later free-choice engagement [~S]). Each is carried to this pattern by extrapolation, and the page says how far. The sequence below is the earlier page's (short attempt-feedback-retry cycles, visible progression, mastery thresholds, rewards kept aligned) made concrete; each step states its evidence status, and the response-dependent branches are untested design proposals.

Inputs: establish the design brief

Field Record or elicit
Learner and starting observation A short placement check on the first two or three levels' targets, scored item by item, before any lesson. Record accuracy, response time and which items were missed; record whether the learner has used a gamified app before and how they speak of points and streaks. Use novice, intermediate or advanced only where the check shows it; keep unknowns.
The skill and its levels Whether the skill is cumulative (later items need earlier ones), whether it breaks into items that can be scored instantly (a word, a sum, a keystroke sequence), and the order of levels. If the target is an open performance (a conversation, an essay, a clinical judgement), record which parts are itemisable and which are not.
Context and activities Setting (online-self-paced, classroom, online-instructor-led, workplace-clinical), device (phone, shared computers), session length and frequency, who sees progress (the learner, a teacher, peers, an employer), whether a teacher or facilitator is present and for how long, and any game layer already in place (an app's XP, streaks or leagues; a class points chart).
Learner-valued goal Ask what the learner wants to be able to do and by when (order a meal on holiday, pass a certification, keep up with classmates), and what they think of streaks and rankings. Record disagreement with the designer objective.
Designer objective The capability each level stands for, the criterion that opens the next level, and which later levels depend on it. Name the instrument outside the app that will judge the course.
Outcome Accuracy on new items of passed levels at a stated delay; performance on a task with no game around it; continued practice after week 2 and after any points are hidden; time per level. Each with its instrument and horizon, recorded separately.

A request to "gamify the course with levels" does not specify these. Ask what the levels stand for, what opens a level and what the outcome outside the app is before choosing points, streaks or a criterion.

Sequence and conditional policy

The default sequence across a course. Quantities are labelled proposals where no claim gives them; set them locally and record them.

  1. Map the skill into short cumulative levels and state each level's criterion. A level is 3–6 lessons of 10–15 scorable items on one target (proposal). The criterion is a check on that target without hints, for example 8 of 10 correct on new items (proposal); tell learners what it is. Evidence status: mastery effects vary with the standard, pacing and test type [+M]: in a moderator regression over 103 studies of mastery programmes (no game layer), programmes with a higher unit mastery standard had larger effects. The size of standard that suits a given app is not on the claim page.
  2. Place before the first lesson. Run the placement check; a learner who meets a level's criterion skips it (tests out). Untested proposal; from the earlier page's "learners at different speeds through the same progression map".
  3. Run each lesson as attempt → immediate feedback → retry. Each item is answered and marked at once, with the correct answer and one informational sentence ("finished action, so past tense"); missed items return at the end of the lesson until each is answered correctly once. A lesson lasts 5–10 minutes (proposal). Evidence status: untested as a step here; the earlier page's first requirement. The informational sentence is the Gamification principle's step 3, resting on verbal positive feedback raising later free-choice engagement in the rewards meta-analysis [+S] (laboratory tasks).
  4. Make a level mean the skill passed, and repeat lessons at harder settings. A level opens on the step 1 check, taken on a later day than the last lesson (proposal), not on lessons completed. Inside a level, the same lesson can be redone at a harder setting (fewer word banks, typed rather than chosen answers). Evidence status: in one analysis of a commercial language app's data (Duolingo case study [+M]), a regression discontinuity found that learners who had studied a lesson twice (two levels) answered later review items on it correctly more often than learners who had studied it once, at least for the first level-up; completing more levels went with higher post-test accuracy on items first answered wrongly (levelling up and checkpoint accuracy [+W]), and levelling up earlier lessons in a skill went with better review accuracy on later ones (transfer across lessons [+W]). One study, no effect size printed, in-app measures only.
  5. Give points for accuracy and improvement, never for opening the app. XP for correct first attempts on new or harder items and for beating the learner's own previous score on a lesson; none for log-ins, time or lessons finished regardless of accuracy. The level map shows criteria passed, not points accumulated. Evidence status: expected tangible rewards for doing or finishing a task [~S] lowered later free-choice engagement in laboratory experiments (d = −0.28 to −0.40 across engagement-, completion- and performance-contingent rewards), more for children; undermining was reported weaker for well-defined tasks with a clear quality standard. Carried from laboratory puzzles to points in an app.
  6. When a learner fails the level check, change the correction and cap the attempts. After a failed check, give a different explanation or a worked example on the missed items, then a parallel check; after a third failed check (proposal), advance with the gap recorded and re-probe it in the next level's first lesson, or route the learner to a person. Evidence status: in one product's logs [~W], students who needed all three post-quiz attempts on a quest reached the 80% criterion only 14% of the time and were promoted after making a video instead. The corrective loop itself is the Mastery Learning pattern's steps 4–7.
  7. Bring earlier levels back. Open each session with 2–3 minutes of review items from passed levels, weighted toward items missed before (proposal); a passed level whose review accuracy falls below the criterion reopens for one lesson. Evidence status: untested here; the earlier page's "retention through revisit". The Spaced Learning principle owns the schedule.
  8. Prompt the next level; plan for drop-off. Show the next level on the map and offer it at the end of each lesson. In a self-paced course, set a weekly target of lessons and a soft deadline per level; a streak, if used, tolerates one missed day. Evidence status: in randomised A/B tests in the same app, a visual stand-in for the level system and a pop-up prompting the next level raised lessons completed per level by more than 10% and overall study activity (level prompts [+M]); this measures practice, not learning. Self-paced mastery courses lost more students: in 23 of 32 college studies completion was higher in the conventional class (self-paced mastery reduces completion [-M]). The streak rule is an untested proposal.
  9. Judge the course outside the app. Test the capability on a task with no game around it, at the horizon the objective sets; record practice volume, streaks and enjoyment separately. Evidence status: in the mastery meta-analysis, effects on standardised tests were near zero (−0.05 to 0.09) in the four studies that used both kinds of test, against up to 0.64 on locally developed tests (effects vary with test type [~M]); a review of tutoring systems argues that platform interaction and task completion cannot by themselves establish cognitive engagement (behavioural engagement [~M]).

Dose and progression across sessions (all proposals; the only durations on the claim pages are the gamification meta-analysis's finding that duration did not moderate cognitive outcomes, and the Duolingo data windows of two and four months):

Phase Session Support and the source of items Moving on
Week 1 10–15 minutes, 5 days a week; two lessons a session Short lessons (about 10 items), choices from a word bank or options, feedback and an informational sentence after every item, an easy first level so every learner passes a check early Level check on day 3 or later; 8 of 10 on new items
Weeks 2–4 10–20 minutes, 4–5 days a week Each lesson redone once at a harder setting (typed answers, fewer hints); 2–3 minutes of review items from passed levels at the start; points weighted to harder and previously missed items Check on a later day than the last lesson; at most three checks, then advance with the gap recorded
Week 5 onward 15–20 minutes, 3–4 days a week Mixed review across levels replaces most new-item drill; points move from every item to every level; personal records and the learner's own targets replace streak messages Level criterion as before; one item set per month checked outside the app
Before the course ends One week Points and streaks hidden; practice continues on the level map alone Outside-app test at the stated delay; the drop in practice, if any, is recorded
Observation Discriminating probe Proposed action, with its uncertainty
Levels and points rise; the outside test does not Give the same items once in the app's format and once outside it; check what share of points came from items already passed. If accuracy holds in the app only, the format is carrying the answer (word banks, recognition): move to typed or spoken answers before the check. If points come from easy items, change what earns them. Neither branch is tested.
Passes the level check, fails review items on it a week later Re-probe two items from the level with changed surface details. Reopen the level for one lesson and add its items to review; a one-day check may be measuring short-term retention. Untested.
Practice drops after week 2, or when points are hidden Compare practice before, during and after; ask what the learner did instead. If the drop starts with hiding points, points may have displaced interest the learner had: give fewer, informational awards. If it starts before, check difficulty and the learner's goal. Untested.
Many quick answers, falling accuracy, a long streak Compare response time with accuracy for one week; remove the speed element for one session. If accuracy recovers, speed or volume is what earns points: pay for accuracy on new items only. Untested.
Repeated failed checks on one level Give a prerequisite item from the level before; ask the learner to explain one wrong answer. If the prerequisite fails, reopen that level; if not, change the correction's form (worked example, a person) rather than cycling the same lesson. Untested.

Fitting the design to a situation

The three facts that most change the decision:

  • Whether the skill breaks into items scored on the spot. The loop depends on instant marking and on a check that stands for the skill; a whole performance does not break that way. If unstated, ask: can each practice attempt be marked right or wrong at once, and does passing the items predict doing the real task?
  • Whether learners already want to do the practice. Points for doing a task risk lowering later engagement most where interest already exists; a dull, necessary drill carries less of that risk. If unstated, ask: do learners practise this now without being made to?
  • Who paces and who sees progress. Self-paced mastery courses lose more learners; progress seen by a teacher, peers or an employer is a different design from progress only the learner sees. If unstated, ask: is there a class timetable or deadline, and who will see levels, streaks or rankings?
If the brief says… Then change… Basis
Learners are novice in the domain Word banks and recognition items first, typed or spoken answers only after the first level is passed; feedback with the correct answer on every item; a criterion check on day 3 or later Untested here; the feedback and gate steps rest on the claims in steps 3–4, none tested with novices separately
Learners are advanced, or already accurate Placement check and test-out by level; harder settings (typed answers, mixed review) from the start; no points for easy items Untested proposal; the earlier page's "difficulty should rise in response to growing skill"
Learners are children (under 12) No tangible prizes for finishing lessons; points and badges for accuracy and effort, worded as information; no public individual ranking; a teacher sees the level map Rewards claim [-S]: tangible rewards more detrimental for children than college students, laboratory tasks
adults in a workplace-clinical setting (onboarding, compliance) Levels tied to job tasks passed at a criterion; no badges for completion or seat time; no ranking among colleagues; the manager sees criteria passed, not streaks Untested proposal; mastery evidence is from school and college courses (mastery learning improves outcomes [+M])
Goal is verbal-association (vocabulary, terms, notation) The full loop fits; check recall on new sentences a week after the level, outside the app Duolingo level analysis [+M] (in-app items, one study); gamified vocabulary, reported in a review [+W]
Goal is a complex-skill or whole-task performance the brief wants kept whole (a conversation, a case, a project) Use levels only for the itemisable parts (vocabulary, a calculation); the whole performance stays one task, judged on the brief's own rubric, outside the level map; do not split it into items to fit the gate Untested; the earlier page's constraint "weak fit for complex open-ended performance"
online-self-paced, no teacher The software runs feedback, retries, checks and review; a weekly lesson target and a soft deadline per level; a streak that survives one missed day Self-paced mastery reduces completion [-M] (college PSI courses, carried to apps); level prompts [+M] (practice volume only)
classroom, one teacher and many learners Learners work the levels on devices; the class keeps a common pace with a weekly level target; the teacher spends the session's last 10 minutes with learners who failed a check twice Mastery effects vary with pacing [~M] (group-based programmes had larger effects, no game layer); group-based mastery and attrition [+M] (one synthesis)
Mixed modes: some learners remote, some in the room, or phone-only One tool reached the same way by everyone; lessons short enough for a phone (about 10 items); no speed-ranked live rounds; level checks run in the tool, not on paper in the room Untested proposal
Scarce facilitator time (volunteers, a busy shift, a tutor who checks in briefly) Automate feedback, retries and checks; the facilitator spends about 5 minutes a week on a list of learners who failed a check twice or stopped practising, and contacts those Untested proposal; the attempt cap rests on eSpark logs [~W]
Time is a single-session Do not build levels, streaks or a points economy; one short lesson with feedback and retry of missed items, one check at the end; claim nothing about persistence Untested; no claim here tests a single session
A fixed short time budget per session (for example 15 minutes) 2 minutes review, 10 minutes of one lesson with feedback and retries, 3 minutes for a level check when one is due; never two checks in one session Untested proposal
Learners already practise willingly, or find the task interesting Keep the level map and informational feedback; drop expected points, prizes and streak rewards Rewards claim [~S] (laboratory tasks, free-choice time; carried to apps)
Learners have low confidence or low literacy Private progress only; a first level everyone passes; feedback with icons and audio; a failed check shown as "one more lesson", not a lost life Untested proposal
Progress or answers are private or sensitive (health, a second-chance course) The learner controls who sees levels and streaks; no shared boards; a teacher or employer sees only what the learner releases Untested proposal
Levels or points feed a grade, or stakes are high Keep points formative; grade on an aligned task outside the app at the delay that matters; never grade on streaks or practice volume Effects vary with test type [~M] (near-zero effects on standardised tests); behavioural engagement [~M]
The brief names its own routine (an app's XP, streaks or leagues; a class points chart; its own level criterion or rubric) Keep it; change only what the evidence says matters: points for accuracy and improvement rather than log-ins or completion, a level that opens on a check rather than on lessons finished, and an outside check beside the app's own Rewards claim [~S]; mastery standard [~M]; the combination is untested
The brief gives its own categories or skill map (CEFR bands, a job's competency list) Use those as the levels and their criteria; do not substitute a level scheme of your own Untested proposal
The criterion is set by an automated rule ("N correct in a row") Prefer a check on new items over a streak of correct answers; if a streak rule is kept, record how many items learners need to pass Simulated mastery learning [~W]: with 30 simulated students per condition, a streak rule gave more problems than knowledge tracing (about 17 against 11 for one problem type); simulated, not real learners
The brief reports an error learners actually made Make it a feedback item: a near-miss answer shown beside the correct one, with the sentence that separates them, in the lesson and in the check Untested proposal

Ten of the twenty rows link a claim (children, adults at work, vocabulary, self-paced, a classroom with one teacher, scarce facilitator time, already-interested learners, high stakes, the brief's own routine, and the automated criterion), and each says where it carries the claim beyond the learners or settings it tested; only the vocabulary and self-paced rows draw on a gamified app, and that is one study. The other rows are untested proposals or the earlier page's guidance.

Choosing configurations from evidence

  • Whether to gate at all. Mastery learning programmes, 108 controlled evaluations [+M]: across 103 studies of examination performance, mastery programmes (Keller's self-paced PSI and Bloom's group-based Learning for Mastery) raised scores by 0.52 SD on average against conventionally paced classes, at college, high-school and upper-elementary levels; the claim's table gathers 7 distinct studies behind the question. No study had a game layer. It supports a level that opens on a criterion as a candidate configuration, not a gamified app.
  • What the gate's effect depends on. Mastery effects vary with the standard, pacing and test type [~M]: larger effects with a higher unit standard, group-based rather than self-paced programmes, locally developed tests, and comparison groups receiving less quiz feedback; standardised-test effects near zero where both were used. Its table gathers 6 distinct studies. For an app this means: a high criterion, some common pacing where possible, and an outside test that is not the app's own.
  • Whether a game layer adds anything. Gamification has small positive effects [+M]: a meta-analysis of 40 experiments (2013–2017, school, university, work and informal settings) found cognitive g = .49, motivational g = .36 and behavioural g = .25, all heterogeneous; among the more rigorous designs only the cognitive effect held (g = .42). It pools many combinations of elements and does not show that levels, points or streaks carry the effect, nor that a game layer adds to a mastery gate.
  • What a level structure does inside one app. The Duolingo case study (Portnoff et al. 2021), one study read as four claims: a regression discontinuity on about two months of review data (levelling up and later accuracy [+M]); a regression over four months of checkpoint quiz data controlling for item, user, course and prior proficiency (levels and checkpoint accuracy [+W]); the same review analysis for earlier lessons (transfer across lessons [+W]); and randomised A/B tests of level prompts (level prompts [+M], more than 10% more lessons per level). No effect size is printed for any of them, every outcome is measured inside the app, and none compares the app with or without its game layer. It says that studying a lesson again at a higher level goes with better later in-app accuracy and that prompts raise practice; it does not say the game layer caused learning.
  • What points should pay for. Expected tangible rewards reduce later free-choice engagement [~S]: two competing meta-analyses of laboratory experiments (128 and 96 studies), the 1971 experiment and two reviews. Deci et al. (1999) found engagement-, completion- and performance-contingent tangible rewards lowered later free-choice engagement (d = −0.28 to −0.40) and verbal positive feedback raised it (d = 0.33); Cameron and Pierce (1994) found no overall undermining and a minimal effect only for expected tangible rewards for doing a task. The claim's table gathers 8 distinct studies. This is the limit on the pattern: a game layer that pays for opening the app or finishing lessons is the configuration most at risk.
  • The cost of self-pacing. Self-paced mastery reduces completion [-M]: in 32 PSI studies, 23 had higher completion in the conventional class (h = −0.14 on attrition), while the examination advantage held either way. Group-based mastery classes reported lower attrition in 7 of 8 disciplines in one evaluation (group-based mastery and attrition [+M]). Most gamified apps are self-paced; the drop-off risk is in the brief, not a side issue.
  • What the platform's own data can show. Behavioural engagement indicators do not establish cognitive engagement [~M] (a review's argument about tutoring platforms) and gamification's effect on engagement in adaptive platforms is mixed [~M] (one review citing one study of large impact and another of minimal effect). Streaks, XP and lessons completed are observations about practice, not about the capability.

Do not rank this pattern against an ungamified mastery course or a conventional course from these claims: none compares them. A precise abstention names the missing comparison (the same levelled practice with and without the game layer, on an outside test at a stated delay, with practice after the layer is removed) and the local observation that would stand in for it.

The evidence by question

Further evidence, not yet read against this model

Claims found while converting this page that bear on it but are not part of the model above. Each says what it is and how far it bears on the page.

Illustrative design instance and observation record

Invented illustration, not a tested course. Adult learners in a hospital's housekeeping team (adult, workplace-clinical, phone-only) learn 200 English workplace phrases over eight weeks, in 15-minute slots at the start of shifts, with a supervisor who can spare about 5 minutes a week. The brief names its own skill map (the hospital's list of phrases by task: greeting a patient, reporting a spill, asking for a colleague) and wants no ranking among staff. The table changed the default: levels follow the hospital's task list; each level is four lessons of about 10 items with audio, feedback and retry of missed items; a level opens on 8 of 10 on new items, taken on a later shift; points come only for correct first attempts and personal bests; progress is private, and the supervisor sees only a list of who failed a check twice or has not practised for a week. A learner says she wants to understand patients' requests on the ward; the designer's objective is producing the phrases. They agree to add a listening item to each level check, and in week 8 points are hidden and a supervisor asks each learner three ward questions face to face. The level-analysis claims come from a language app's in-app data, not from this population, setting or outcome; these are local choices to observe.

Record: level, its target and criterion → placement result → lessons taken, items, attempts and feedback received → points earned and for what → level check form, date, attempt number and response → correction given after a failed check → review accuracy on passed levels → practice per week, before and after points were hidden → outside-app test and horizon → candidate interpretations (skill passed, app format carrying the answer, short-term retention, points displacing interest) and their basis, alongside the learner's stated goal and any change in it. An agent may summarise this record but must not fabricate missing inputs or assign a probability that a learner "has mastered" a level without a calibrated model.

Elements and limits

Adaptive mastery learning, adaptive difficulty, feedback, practice, reassessment, formative assessment, progress tracking and gamification. Principles the pattern draws on: Mastery Learning, Gamification, Game-Based Learning, Immediate Feedback, Spaced Learning and Self-monitoring.

This pattern is scoped to cumulative skills that break into items scored on the spot. Open-ended performances, contested judgement and programme-level certification need other configurations (competency-based learning, epistemic games). The policy supports observation and design reasoning; whether a gamified mastery course produces better outside-app performance than the same levelled practice without the game layer, for which learners, at what cost in drop-off and at what horizon, remains to be tested.

Claims

Examples

  • Scaffolding via level-gated progression and hearts in Duolingo — a design that uses this pattern
  • Duolingo-style language progression with retries, review lessons and a level map; the level analysis [+M] comes from one such app's data
  • Mastery-based digital math practice in which a skill opens only after a check on new items
  • Duolingo-style language progression with retries, streaks, and review loops.
  • Mastery-based digital math practice with adaptive item difficulty.
  • Technical skill trainers that unlock later levels only after demonstrated competence.
  • My Math Academy adaptive game design — early math games with in-game scaffolding and a personalised path between skills

Key Sources

  • Gee, J. P. (2003). What video games have to teach us about learning and literacy. Palgrave Macmillan. doi:10.1145/950566.950595
  • Kulik, C.-L. C., Kulik, J. A., & Bangert-Drowns, R. L. (1990). Effectiveness of mastery learning programs: A meta-analysis. Review of Educational Research, 60(2), 265–299. doi:10.3102/00346543060002265
  • Sailer, M., & Homner, L. (2020). The gamification of learning: A meta-analysis. Educational Psychology Review, 32(1), 77–112. doi:10.1007/s10648-019-09498-w
  • Portnoff, L., Gustafson, E., Bicknell, K., & Rollinson, J. (2021). Methods for language learning assessment at scale: Duolingo case study. Proceedings of the 14th International Conference on Educational Data Mining (EDM21). https://educationaldatamining.org/edm2021/
  • Kapp, K. M. (2012). The gamification of learning and instruction. Pfeiffer.
  • Van Eck, R. (2006). Digital game-based learning: It's not just the digital natives who are restless. EDUCAUSE Review, 41(2), 16-30.
Page metadata
Field Value
type pattern
id game-based-mastery-learning
aliases game-based-mastery-learning-eg-duolingo-pattern, game-based-mastery-learning-duolingo-pattern
title Game-Based Mastery Learning
description A reusable policy in which progress through short, cumulative levels depends on passing a check on the target, with attempt-feedback-retry lessons, review of earlier levels and points paid for accuracy and improvement rather than activity; expected to raise performance outside the app for itemisable, practice-heavy skills, but no claim tests the combination as a whole, only a mastery gate without a game layer, a game layer without a gate, and one app's level system on in-app measures.
status review
grain_size lesson
author gamified mastery learning tradition
canonical True
generated by claude/unspecified on 2026-10-15
trust unverified — no one has confirmed this page's content

Frontmatter is the page's machine-readable record — see the Schema & Guide.