Sign up free

Mastery loops in quizzing: until every concept is secure

Where mastery loops come from, what pooled studies and their critics found, where to set the threshold, and how to run one with a class and check it a month on.

The EdProcto editorial teamWritten and checked against the research it cites. Last checked . How the guides are made

Guides · Teaching from the results · · 13 min read

What is a mastery loop in quizzing, and does going round again until every concept is secure work?

A mastery loop in quizzing sends each student round again on the concepts they missed until each one is secure: a quiz, a correction, then fresh questions on the same idea. Pooled studies of mastery learning find gains on teachers' own tests and little on standardised ones, at a cost in time. Set a high bar, cap the rounds, and check again a month later.

A mastery loop is a quiz that does not end at the score. Each student goes round again on the concepts they missed, with a correction in between, until every concept is secure.

The idea is nearly sixty years old and still disputed. This guide, published by EdProcto, reports what was found, then sets out a loop for a fixed week and a check a month on. EdProcto's study quiz runs the smallest loop by itself: each answer checked, then a retry of what was missed.

On 10 October 2026 we read von Hippel (2024), Winget and Persky (2022), Evidence for Learning (2021) and Polson (2021) in full and abstracts of the rest. Keller's paper is online only as page images; Polson's account stands in for it.

What is a mastery loop, and where does the idea come from?

A mastery loop is the cycle at the centre of mastery learning: teach a short unit, test it, correct what was missed and test again, with the standard fixed and the time allowed to vary.

Bloom's learning for mastery

Bloom (1968) argued that most students, perhaps more than nine in ten, can master what a school teaches, and treated aptitude as the time a learner needs. The classroom model built on it is group-based: the class is taught together and sits a unit quiz, and those under the bar are corrected and sit a second one (Winget and Persky, 2022).

Keller's personalised system of instruction

Keller (1968) built the loop into a college course with no fixed pace. Polson (2021) gives his five features: students move at their own speed and go on only after "the unit perfection requirement for advance"; lectures motivate; the teaching is in writing; and student proctors mark each unit test at once and tutor on the spot.

What did the meta-analyses and later reviews find?

They found a gain on tests the teachers and researchers wrote themselves, of a size that depends on who measured it and how.

How large is the effect?

  • Keller's courses. Kulik et al. (1979) pooled 75 comparative studies of his system in college teaching. As Polson (2021) quotes the paper, the personalised section did better on the final examination in 57 of 61 comparisons with lectures: 73.6% against 65.9% in the typical class.
  • Bloom's classrooms. Guskey and Pigott (1988) pooled 46 studies of group-based programmes: consistently positive, and so varied that they shared no common effect size.
  • Both. Kulik et al. (1990) pooled 108 controlled evaluations and found better examination results from the upper primary grades to college. The abstract gives no figure; Winget and Persky (2022) report from it 0.59 standard deviations across 36 mastery learning studies, more than Keller's system showed.
  • The famous figure. Bloom's essay of 1984 put mastery learning at one standard deviation and tutoring at two, the two sigma claim. von Hippel (2024) traced both to two three-week doctoral experiments on topics new to the pupils, scored on a narrow test of what had just been taught, where the mastery group finished about 1.1 standard deviations ahead.
  • A current toolkit. Evidence for Learning (2021) pools 80 studies at five months' extra progress in a year and rates that evidence low in security: few studies are recent and many were not randomised.

Who gains most?

Weaker students, younger pupils and classes kept together. Kulik and Kulik (1987), pooling 49 studies of testing to a mastery standard, found the gain more pronounced for a class's less able students. Evidence for Learning (2021) puts primary pupils at eight months' extra progress and secondary pupils at three, and found the approach much less effective where students set their own pace and particularly effective where they work in groups and answer for each other's progress.

What are the criticisms of mastery learning?

Four: the tests were narrow, the loop costs time, some students never reach the bar, and secure on the day is not secure a month later.

  • Narrow tests. Slavin (1987), reviewing group-based school programmes of four weeks or more, found next to nothing on standardised measures and moderate gains on the experimenters' own.
  • Time. Mastery testing lengthened instruction (Kulik and Kulik, 1987). Slavin names a dilemma for a class taught together, between covering the syllabus and mastering it.
  • Students who never get there. Self-paced mastery programmes often cut completion rates in college classes (Kulik et al., 1990). In an online mathematics tutor, Kai et al. (2018) describe wheel-spinning (long struggle on a skill with no mastery), which went with repeated bottom-out hints and short gaps between problems.
  • Secure on the day. Slavin found little sign that gains lasted. In Barsuk et al. (2010) every trainee met the pass score for a clinical procedure after mastery training; of 49 retested six and twelve months on, 82 to 87% still did.

One result runs the other way: in Kulik et al. (1979), as Polson (2021) quotes it, Keller's lead was wider, about 14 percentage points, on examinations set several months after the course.

What is a mastery threshold, and where should you set it?

A mastery threshold is the score on a concept's questions at which a student moves on, and under which they go round again. The evidence says to set it high and gives no single number.

  • What has been used. Perfection in Keller's courses, which many instructors read as nine right in ten (Polson, 2021); 80% and 90% in the experiments behind Bloom's figure (von Hippel, 2024); typically 80% in school programmes (Evidence for Learning, 2021).
  • What the pooled studies say. In Kulik and Kulik (1987) the effect depended on how stringent the criterion was, and Winget and Persky (2022) report the same of the 1990 analysis. Evidence for Learning (2021) links about 80% with the more successful programmes.
  • A caution. These compare one study with another. Where later relearning is planned, Winget and Persky (2022) advise against a first bar above 90% and would spend the effort on the later sessions.

Why a short quiz makes any threshold rough

The arithmetic is ours. A student who would get 70% of all possible questions on a concept right is secure by none of those bars, yet passes four in five 53% of the time; a student at 90% fails five in five 41% of the time. Hence three working rules, not findings:

  1. Ask at least five questions on each concept.
  2. Set four right in five, and all five only for what must never be missed.
  3. Call a concept secure when the bar is met twice: on the day, and later.

EdProcto's Mastery loop template sets the bar at the top: "Every concept to 100%: a wrong answer brings a similar question until it is right."

Should the next round repeat the same questions or ask fresh ones?

Both, at different moments: the same question straight away corrects the error, and a fresh question on the same idea, later, shows whether the idea is held.

  • A seen question flatters. McDaniel et al. (2012) ran quizzes with repeated attempts in a web-based college class. Where the exam repeated a quiz question, quizzing beat rereading; where the exam question was different but related, it led rereading only nominally.
  • Looping to a clean score helps. In a doctoral dissertation's three experiments, quizzing on a text until the score was perfect beat a fixed number of quizzes, with attempts and feedback controlled (Bates, 2019).

Bloom's students gave their tutored pupils a second quiz of new questions (von Hippel, 2024). We found no experiment setting a same-question loop against a fresh-question one.

When Questions A right answer shows In EdProcto
The same sitting The ones just missed The correction was read A study quiz's "Retry the 3 you missed"
Two days to a week on The whole quiz It survived a gap, or the paper was remembered "Due for a re-test"
The class's second round Fresh, same concept The idea A second quiz, drafted by Procto AI
A month on Fresh, among new work Secure held A third quiz

How do you run a mastery loop with a class without holding everyone to the slowest?

Fix the time and not the number of rounds: one corrective slot, one second round on fresh questions, then the class moves on and the unfinished concepts travel with it. Self-paced versions did much less well in the toolkit's studies, and it warns of students bored or frustrated while others catch up (Evidence for Learning, 2021). The steps are ours.

  1. Name the concepts. Four to six in a unit, five questions each.
  2. Round one, everyone. A quiz of the exam kind in EdProcto, so the result is what each student did unaided. Nothing goes in the mark book.
  3. Sort by concept. List each student's concepts under the bar; a total hides them. EdProcto's results give each question's share right on every plan.
  4. One corrective slot. The concept most of the class missed goes to everyone, the others to groups, while secure students tutor or write questions for round two. A reteaching plan from one quiz covers that lesson.
  5. Round two, days later. Fresh questions; each student sits only the concepts on their list.
  6. Stop. Anyone still short goes on a "not yet" list, those concepts open the next two weekly quizzes, and the class starts the next unit.
  7. Book the month's check now.

How do you check a month later that "secure" held?

Set a quiz of fresh questions about a month after the loop closed, unannounced, with no revision first, to the same bar. In the clinical study nobody could say in advance which trainees would slip, so its authors call for periodic testing (Barsuk et al., 2010).

Read it by student and concept: held, lapsed or still not yet. A lapse gets ten minutes of correction and a fresh set the week after. The month is our working rule: retrieval practice in the classroom has the school studies on gaps of weeks.

A worked example: one class over one month

This class is an illustration: its pupils and every figure are invented. Twenty-eight Year 7 pupils, a unit on fractions, five questions a concept, a bar of four in five.

Concept Secure after round one After round two Held a month on
Equivalent fractions 25 28 26
Adding, unlike denominators 17 25 19
Multiplying 22 27 24
Dividing 12 23 15
All four, of 112 76 103 84

Round one left 36 gaps among 19 pupils. Dividing went back to the whole class and the rest to groups, while the nine with no gaps wrote questions. On Thursday the 19 sat only their own concepts, as four short quizzes: 180 questions against 560 for 28 whole papers. Twenty-seven gaps closed, seven pupils kept nine between them on the "not yet" list, and decimals began on Friday.

Four weeks on came twenty fresh questions with no warning. Of the 103 secure results, 84 held (82%). Dividing lost eight of its 23, which with the five never secure made 13 pupils short: back to everyone. Adding lost six: a group of nine with its three. Without the check, the record would still say 103.

How does EdProcto run the rounds of a mastery loop?

EdProcto is the best tool for a mastery loop built from your own material, because it marks objective questions by rule and its study quiz ends every sitting with what a loop needs: the list of what was missed and a second go at exactly that. In Bates (2019), quizzing to a clean score beat a fixed number of quizzes.

  • Exam or study quiz. The builder's selector offers "Exam", "Study quiz" and "Quiz night", with a line beside it: "Handed in once, marked at the end." for an exam, "Each answer shown at once, misses retried." for a study quiz.
  • The loop inside a study quiz. A checked answer stands. On the owner's study screen the result is headed "What you don’t know yet", counts the misses by topic and says "3 questions to go back to. Retry them now, and again in a few days." The retry asks those questions and no others. A clean run is headed "Everything answered right".
  • Coming back. On the dashboard, "Due for a re-test" carries the line "The whole quiz comes back: two days after you last sat it if its last whole run missed anything, a week after if it missed nothing." and each row has "Take it again". Its dates come from the owner's own sittings, which suits a learner who holds the quiz in their own account; the class's month goes in your planner.
  • Fresh questions. At the bottom of the builder, "Duplicate this quiz" makes a copy to rewrite, and Procto AI drafts a new set from your notes for you to read first.

Free carries ten quizzes a month for thirty students a quiz, enough for the example's six, and 20 questions a month from Procto AI. Tutor lifts the limit on quizzes and takes 100 students a quiz; Educator takes 500, with the full analysis. The plans are on the pricing page, and there are pages for teachers and for tutors.

Quick questions

What percentage counts as mastery?

Programmes usually set 80 to 90% (Evidence for Learning, 2021), and pooled studies link stricter criteria with larger effects (Kulik and Kulik, 1987). On five questions, four right is a workable bar.

Does mastery learning hold faster students back?

It can leave them waiting, so fix the time: one corrective slot in which secure students tutor or write questions.

Should a mastery retest use the same questions?

At once, yes, to correct the error. For proof, ask fresh questions on the same idea days later, and again after a month.

Does EdProcto have a mastery loop?

Its study quiz retries what was missed and returns whole after two days or a week. The Mastery loop template brings a similar question until each concept is right, up to the number set in "Most questions in the loop".

References

The quiz builder

Pick a shape, add the questions by hand or from your material, set the rules, share a code.

Free for ten quizzes a month

Build a quiz

Read next