Learning science glossary for parents: 60 terms in plain words
Spacing, retrieval, interleaving, mastery learning, knowledge tracing, effect sizes, ESSA tiers and more, each in one plain sentence with its source.
The science of learning has its own vocabulary, and most of it describes a few simple ideas: practise from memory, spread practice out, mix kinds of problem, aim for mostly-right, and judge results with a fair comparison. This glossary defines 60 terms in one plain sentence each, grouped by topic, with the study or source behind each number.
It is written for parents reading app marketing, school letters and articles like the ones on this blog. Peabody (peabody.kids) is a free AI-driven adaptive learning platform for kids, free through the end of 2026, and where a term describes something Peabody does, the last column says how. AI writes the stories; code decides the rest.
Memory and practice
| Term | What it means | Where it comes from |
|---|---|---|
| Retrieval practice | Learning by pulling an answer out of memory instead of looking at it again | Roediger and Karpicke 2006: 56% vs 42% recall a week later (PDF) |
| Testing effect | The same finding seen from the other side: a quiz teaches as well as measures | Same study; rereading felt better and was remembered worse |
| Spacing effect | Practice spread over days is remembered longer than the same practice in one go | Cepeda et al. 2008, 1,354 people (PubMed) |
| Optimal gap | The best wait between sessions, which grows with how long you need to remember | About 20–40% of a week's horizon, 5–10% of a year's (Cepeda 2008) |
| Spaced repetition | Software or flashcards that bring each item back on a growing schedule | Leitner boxes, SM-2, FSRS; Peabody uses gaps of 1, 3, 7 days, then three weeks |
| Expanding schedule | Each gap longer than the one before | Karpicke and Roediger 2007 found delaying the first test matters more than the expanding shape |
| Delayed first retrieval | Not testing a new item the moment it is taught | Same paper: the active ingredient of spacing |
| Relearning | Coming back to an item weeks later and getting it right again | Rawson and Dunlosky 2011: three right, then three spaced relearnings (ERIC) |
| Interleaving | Mixing kinds of problem in one session instead of one kind at a time | Rohrer et al. 2020: 61% vs 38%, 787 seventh graders (WWC) |
| Blocked practice | One kind of problem at a time; feels easier, teaches less for the long run | Taylor and Rohrer 2010: 77% vs 38% next day for interleaved (USF) |
| Desirable difficulty | A harder practice condition that helps later recall, if the learner can overcome it | Bjork and Bjork 2011 (PDF) |
| Performance vs learning | How well a child does during practice is a poor guide to what they keep | Bjork; interleaving looks worse in practice and better a day later |
| Automaticity | Knowing a fact well enough to answer without working it out | About 2 s is the classroom convention (ERIC) |
| Doer effect | Doing practice teaches more than reading or watching | About 6× reading, 3× video across 12,500 students (Koedinger et al. 2015/16, PDF); one replication found it for persistence only |
Difficulty and mastery
| Term | What it means | Where it comes from |
|---|---|---|
| Mastery learning | Move on when a skill is solid, not when the calendar says | Bloom; Kulik et al. 1990 put the effect around 0.5 SD, mostly on local tests |
| Bloom's two sigma | Bloom's 1984 claim that tutoring plus mastery lifts students two standard deviations | Not replicated at that size: pooled tutoring trials land near 0.3 (NBER) |
| Zone of proximal development | What a child can do with help but not yet alone | Vygotsky 1978 |
| Outer fringe | Knowledge Space Theory's formal version: the skills a learner is ready for next | Doignon and Falmagne (Springer) |
| 85% rule | A model result that simple learners learn fastest at about 85% right | Wilson et al. 2019, two-choice tasks, not children (PMC) |
| Target success rate | The chance of a right answer a system aims for when it picks a question | Math Garden aims at 75% (Hofman 2018); Peabody's rounds are about 85% known |
| Guess rate | How often a child gets it right by luck | One in three with three boxes, which is why one right retires a question in Peabody but passing a skill needs 80% of many questions solid |
| Wheel-spinning | Practising a skill without getting it: no three right in a row within ten tries | Beck and Gong 2013 |
| Gaming the system | Rapid guessing or hint abuse to get through practice | Baker et al. 2004: as strong a predictor of post-test as prior knowledge (PDF) |
| Productive failure | Letting students try a problem before teaching it | Sinha and Kapur 2021: g = 0.36, but instruction first wins for grades 2–5 |
| Misconception (bug) | A consistent wrong procedure, like taking the smaller digit from the larger | Brown and Burton 1978; with three boxes, each wrong box can name one |
| Elaborated feedback | Saying why an answer is wrong, not just that it is | Shute 2008 (PDF); feedback overall d = 0.48 (Wisniewski 2020) |
How software models a learner
| Term | What it means | Where it comes from |
|---|---|---|
| Adaptive learning | Software that changes what comes next based on a learner's answers | A learner model, a schedule and a difficulty target |
| Knowledge component | One skill the model tracks, like "7 × 8" or "short a" | Koedinger, Corbett and Perfetti 2012 |
| Q-matrix | The table saying which skills each question needs | Used by PFA and AFM |
| Knowledge tracing | Estimating, answer by answer, whether a skill is known | Corbett and Anderson 1995 onward |
| BKT | Bayesian Knowledge Tracing: known or not, with guess and slip rates | "Known" by convention at 95% probability |
| PFA | Performance Factors Analysis: a regression on past rights and wrongs | Pavlik, Cen and Koedinger 2009 |
| DKT | Deep Knowledge Tracing: a neural network over the answer history | Leads only above about a million interactions, by 0.06 AUC or less (Gervet 2020, JEDM) |
| Elo rating | A number per child and per question, nudged after every answer | Klinkenberg 2011 (Math Garden); Pelánek 2016 (PDF) |
| Item response theory | Models the chance of a right answer from ability and item difficulty | Rasch; the base of adaptive testing |
| Computerised adaptive testing | Picking each test question to learn the most about a child | A five-question placement can't get very precise, so it is a first guess |
| Learning curve | Error rate falling over practice tries for one skill | A flat or jagged curve means the skill is badly defined |
| FSRS | A modern spaced-repetition scheduler that predicts recall | Beats SM-2 at prediction on its makers' own benchmark (GitHub) |
Reading
| Term | What it means | Where it comes from |
|---|---|---|
| Systematic phonics | Teaching letter-sound patterns in a planned order | Ehri et al. 2001: d = 0.41, 0.55 when begun in K or 1st (SAGE) |
| Simple View of Reading | Reading comprehension = decoding × language comprehension | Gough and Tunmer 1986 |
| Orthographic mapping | Bonding a word's spelling to its sounds and meaning so it is read on sight | Ehri 2014 |
| Self-teaching | Each successful sounding-out of a new word helps the child learn it | Share 1995 (PubMed) |
| Decodable text | Text a child can sound out with the patterns taught so far | Cheatham and Allor 2012: one criterion among several |
| Leveled text | Text sorted by overall difficulty, often with pictures to guess from | Fountas and Pinnell letters A–Z |
| Heart word | A common word with a part that can't be sounded out yet | Taught after the phonics step it fits |
| Readability formula | A score from sentence length and word length or rarity | Flesch–Kincaid, Dale–Chall; noisy on short texts |
Evidence and studies
| Term | What it means | Where it comes from |
|---|---|---|
| Effect size (d, g) | How far apart two groups ended, in standard deviations | 0.20 is large for a scaled program on a standardised test (Kraft 2020) |
| RCT | A randomised controlled trial: chance decides who gets the program | ESSA Tier 1, "Strong" |
| Quasi-experiment | A comparison group matched but not randomised | ESSA Tier 2, "Moderate" |
| Active control | The comparison group does something else useful, not nothing | Shrinks effects; Sitzmann 2011 |
| Researcher-made measure | A test written by the study's authors, close to what was practised | Shows 2–3× the effect of standardised tests (Kim et al. 2021, PDF) |
| Fade-out | Gains shrinking after a program ends | Bailey et al. 2020: no meta-analysis shows full persistence |
| Regression to the mean | Low scorers rising on a retest with no learning at all | Why a low placement followed by "growth" proves little |
| Near and far transfer | Getting better at what was practised vs at something general | Near is real; brain-training's far transfer is near zero (Sala 2018) |
Games and motivation
| Term | What it means | Where it comes from |
|---|---|---|
| Intrinsic integration | The learning is the game's main action, not a quiz between levels | Habgood and Ainsworth 2011: children chose it 75.7 vs 10.3 minutes (PDF) |
| Chocolate-covered broccoli | Learning bolted onto play as a wrapper | Conventionally Bruckman 1999 |
| Self-determination theory | Motivation from competence, autonomy and relatedness | Ryan, Rigby and Przybylski 2006 for games |
| Undermining effect | Expected rewards for doing a task can lower a child's own interest in it | Deci, Koestner and Ryan 1999, stronger in children |
| Dynamic difficulty adjustment | A game quietly tuning its challenge to the player | Hunicke 2005; rubber-banding is the version players resent |
| Dark pattern | A design that works against the player's interest without consent | Zagal, Björk and Lewis 2013 |
How to use this glossary
Three terms do most of the work when you read a claim about a learning app. Effect size tells you how big a gain was; active control and researcher-made measure tell you how much to shrink it. Then ask the product which of the practice terms it actually does. Peabody's rules are written up rule by rule in how Peabody picks the next question, with the studies and their limits on the science page. Peabody has no efficacy study of its own, so none of the numbers above are claims about it.
Every definition here is one sentence on purpose. Several terms (Bloom's two sigma, the doer effect, the 85% rule) are contested or narrower than they are usually quoted, and the source column says so where it matters. Start free to see the practice side in a game, or open the free library for the reading side.
Learn more elsewhere
- Deans for Impact: The Science of Learning: how students learn, in a short free summary (second edition, 2026).
- The Learning Scientists: free posters and bookmarks on six study strategies, from cognitive psychologists.
- Daniel Willingham's articles: short columns by a cognitive scientist on memory, reading, motivation and the learning-styles myth.
For about seventy more, checked and sorted by who they're for, see where to learn about AI.
Questions parents ask
What is the science of learning?
The research on how people learn and remember, mostly from cognitive psychology and education trials. Its most reliable findings for home practice are retrieval practice, spacing, interleaving and feedback, and its strongest warning is that how practice feels is a poor guide to what sticks.
What is the difference between spaced repetition and retrieval practice?
Retrieval practice is answering from memory; spaced repetition is the schedule that decides when an item comes back. They work best together: recall it, wait, recall it again after a longer gap.
Are any of these findings disputed?
Some are. Bloom's two sigma has not replicated at that size, the doer effect failed on outcomes in one replication, the 85% rule comes from models rather than children, and the testing effect was not reliable for maths procedures in a 2025 meta-analysis. Spacing and retrieval for facts and words are the most solid.
Which three terms matter most for judging an app?
Effect size, active control and researcher-made measure. A big number on a test the company wrote, against a group that did nothing, is a much smaller number in real life.
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.