Mastery learning explained, and why grade levels are only a start
What mastery learning is, what Bloom's two sigma did and didn't show, how software decides a skill is known, and why a child's grade should only set the first step.
Mastery learning means a child moves on when a skill is solid, not when the calendar says so. Reviews put its effect at roughly half a standard deviation, mostly on local tests; Bloom's famous two-sigma figure has not replicated. The hard part is deciding when a skill is known, and a grade level is a guess.
Peabody (peabody.kids) is a free AI-driven adaptive learning platform for kids, free through the end of 2026, built around a mastery ladder per subject. AI writes the stories; code decides the rest. This post covers what the research says and how Peabody's ladder decides.
Mastery learning in one table
| Calendar-paced | Mastery-paced | |
|---|---|---|
| What moves a child on | The week, the unit, the grade | Evidence the skill is solid |
| What is fixed | Time | The bar to pass |
| What varies | How much each child learns | How long each child takes |
| What happens after a miss | The class moves on anyway | More practice, a different explanation, then a re-check |
| The risk | Gaps that compound | Getting stuck on one skill for too long |
What Bloom found, and what happened to it
Benjamin Bloom (1984, Educational Researcher) reported that one-to-one tutoring combined with mastery learning lifted students about two standard deviations above a conventional class, and mastery learning alone about one. It is the most quoted number in education technology. It did not hold up at that size.
Von Hippel (2024, Education Next) traced Bloom's figure to two dissertations with about 20 students per group, narrow tests, and a feedback-and-retest routine that alone accounted for about 1.1 standard deviations (Education Next). Kulik, Kulik and Bangert-Drowns (1990, Review of Educational Research) put mastery learning at roughly half a standard deviation, larger for weaker students and mostly on tests made locally. Pooled randomised trials of tutoring land at 0.37 SD over 96 trials (Nickow, Oreopoulos and Quan 2020) and 0.29 in the 2024 version; none reached two sigma. On Kraft's (2020) scale for programs at scale, 0.2 is already large.
So the idea survives and the number shrinks. Moving on when a skill is solid is sound; it does not double anyone's learning.
How software decides a skill is known
This is the part that matters for an app, and the part with the most traps.
| Approach | The rule | The trap |
|---|---|---|
| Three in a row | Pass after three consecutive right answers | Easy to hit by luck with multiple choice, and short-term |
| Bayesian Knowledge Tracing | Pass when the model's probability the skill is known reaches 95% | Assumes a guess rate under 30%; three answer boxes give 33% by blind guessing |
| Counts and shares | Pass when enough has been seen and most of it is solid | Cruder than a model, but readable |
| Retention checks | Pass, then confirm days and weeks later | Slower; the only one that measures keeping |
The best recent warning comes from a two-year trial of Khan Academy with Khanmigo (Oreopoulos and Low 2026): its mastery rule (a video plus three in a row) raised how many students reached the rule but not their delayed learning (working paper). Hitting a bar is not the same as keeping a skill. Rawson and Dunlosky (2011) found that coming back to items on three widely spaced later occasions mattered far more for retention at one to four months than how many times they were right the first day (ERIC).
The opposite trap is wheel-spinning: practising a skill without getting it. Beck and Gong (2013) defined it as no three right in a row within ten tries, and found it common; the advice that follows is to change the approach rather than serve an eleventh identical question.
Why grade levels are only a start
A grade is a fair first guess and a poor ceiling. Children in one class are spread across several years of skill, and a fixed grade path either repeats what a child already has or skips what they don't. Mastery learning's whole point is to let the evidence, not the label, decide. The catch is that the first evidence is thin: the research report behind this post notes a five-question placement cannot narrow a child's position much, so it should be a starting point that practice quickly corrects.
How Peabody's ladder decides
Each subject is a ladder. Maths climbs counting, making ten, adding, taking away, number sense, times tables, dividing; reading climbs first sounds, rhymes, phonics, sight words, spelling, word meanings. Every number below is a constant in the code, written up in how Peabody picks the next question.
- A grade sets only the first rung. Kindergarten starts at counting and first sounds, first and second grade at adding and phonics, third grade and up at times tables and word meanings.
- A rung passes on counts, not a streak. Enough of it seen (24 sums, 30 times facts, 30 phonics words and so on), 80% of what was seen solid, and only a few items still shaky.
- Or a child tests out. Twelve or more questions seen and 88% of all answers right passes the rung whatever the day-by-day levels say. This was added after a child who was past adding kept seeing sums everywhere.
- The next rung shows up early. Rounds rotate between the lowest two rungs not yet passed once a child is halfway up the current one, and about one round in four reviews a passed rung.
- Passed stays passed. Rungs are saved per child and apply in every game, so one slip in a review doesn't send a child back.
- Known items come back. After 1, 3 and 7 days, then three weeks; a miss puts an item back into learning.
Two honest limits. Peabody's ladder counts rather than fitting a model, and "right once, never again" retires a question long before Rawson and Dunlosky's three spaced relearnings; adding spaced confirmation checks is on Peabody's own list of changes. And Peabody has no efficacy study, so none of the effect sizes above are claims about it. Start free, or see the reading ladder at work in the free library.
Learn more elsewhere
- EEF toolkit: mastery learning: a UK research charity rates mastery learning by impact, cost and strength of evidence.
- J-PAL: teaching at the right level: 13 randomized trials of teaching children at their level, not their grade.
For about seventy more, checked and sorted by who they're for, see where to learn about AI.
Questions parents ask
What is an example of mastery learning?
A child practises adding within 10 until nearly all of it is right and quick, then moves to adding within 20, whatever week of the school year it is. A classmate who already has it skips ahead; one who needs longer gets it.
Is mastery learning better than traditional teaching?
On average, moderately: reviews put it around half a standard deviation, mostly on tests made by the schools themselves, and larger for weaker students. The two-sigma figure is from small studies and has not replicated.
Can a child get stuck in mastery learning?
Yes, which is why good systems watch for it. No three right in a row within ten tries is a common warning sign. The fix is a different explanation, an easier step, or a grown-up, not more of the same question.
Why does Peabody use a grade at all?
Only as a starting point. It decides the first rung so a new child doesn't begin with things far too easy or far too hard. After that, passing is decided by the child's answers.
Checked against how Peabody works on 18 October 2026. Written with help from AI. Peabody is a reading app, not a reading assessment: if you're worried about your child's reading, talk to their teacher.