--- title: "How children learn through play" description: "What research says about how children learn through surprise, self-led play and rewards, how Professor Oops fits it, and where the research pushes back." source_name: "Game Tailors" source_url: "https://professoroops.com/en/inside/research-play/" site_home: "https://professoroops.com/" source_description: "The drawing app where the paper is real. No ads. No accounts. No personal data: only anonymous counts, which you can switch off." locale: "en" legal: - "Game Tailors · Schieweg 15Y-40, 2627 AN Delft (NL)" - "KvK 92437702 · VAT NL866044474B01" - "info@gametailors.com · +31 6 4227 5919" copyright: "© 2026 Game Tailors" published_at: "2026-11-10T12:02:42.230Z" modified_at: "2026-11-10T12:02:42.230Z" --- # How children learn through play ## A world that answers back, and nobody keeping score Research on young children keeps pointing the same way: they learn from surprises they can follow up, in play they lead themselves. Professor Oops gives them a sheet of paper where everything really reacts, with no goals, no scores and no way to fail. We built that toy first and read the research afterwards. Much of it fits what we built. Some of it pushes back, and this page says where. [For engineers](<#hood-engineers>) ![Painted illustration of the painter finger-painting a big sheet covered in hand prints, paint splats, a toothpaste squiggle, glitter and a cracked egg.]() ## We built a toy first and read the research afterwards Professor Oops started as a toy, not a lesson. The first design says it in one line: “There are no goals, scores or failure states. A child can make art or just make a mess.” Those rules came from a review of what children already play with, among them Kid Pix and its dynamite eraser, Tux Paint with its big simple tools, and falling-sand games, where a few simple rules make endless stories. That review also set the bar for the simulation: “The simulation has to be real enough that combinations surprise people, even the designers.” The learning research came later, when we wrote pages for parents and teachers. That order matters for how you read this page. The game was not derived from the papers below. We checked it against them, and we report what we found, including where they disagree with us. One more thing to say plainly: we have never measured whether children learn anything from the game. There is no study of our own, no control group and no test score. What follows is what the design draws on, never what it proves. A world that answers back, and nobody keeping score For engineers ### A world that answers back, and nobody keeping score The released game has no goal state, no score and no failure state \(design spec, 2026-10-17\). Its value for play rests on one property of the engine: 97 materials react through the same shared per-cell layers, so an experiment nobody planned still gets a consistent, repeatable answer. Undo keeps 4 to 100 steps inside a fixed memory budget, storing only what changed from one touch to the next. We have run no learning study; every source below is labelled by how it relates to the game you can play today. [Under the hood](<#hood-history>) We built a toy first and read the research afterwards Under the hood ### We built a toy first and read the research afterwards **No goal, in code.** There is no goal state, score variable or failure screen to switch off. A search of the released interface code for badge, achievement, streak, leaderboard, high score, trophy, level up and game over finds only lines that deny them, such as “No timer, no streak, nothing to come back for” at the top of the break screen's code. **The labels on this page** describe each source's relation to the released game: - **Applied directly**: a concrete rule in the game traces to it, and a design document says so. - **Informed the design**: read while the game was designed; it shaped the approach, with no one-to-one rule. - **Found later**: read after the design was set, and consistent with it. - **Pushes back**: found later, and it challenges part of the design or of what we may claim. No learning paper is labelled "applied directly". The released game's rules came from product precedents, and we would rather say so than borrow authority. 16 Oct 2026 ### A review of earlier creative apps and sandbox games No rules, no failure, no wrong move; destruction as play; an undo the child can trust; no ads, after a press report of an app study [Meyer 2019](). 17 Oct 2026 ### The first design "No goals, scores or failure states." Materials must interact with each other, and it must run on a mid-range tablet. 20 Oct 2026 ### The business rules Nothing locked or teased in the child's view, written after a review of consumer law, regulators' guidance and a study of manipulative design in children's apps [Radesky 2022](). 24 Oct 2026 ### A daily play-time limit for grown-ups Built because Brazil's ECA Digital asks for parental time tools. 29 Oct 2026 ### The first learning papers Cited on the Professor Oops website, on a page for parents asking whether children learn anything from it. ![The game: a page after free play, with a ketchup splat, an egg broken on the paper with bits of shell round it, a blue watercolour wash gone soft and pale along one edge where water was poured on it, and a small burnt hole with a dark rim and the table showing through.]() **What "no goal" looks like.** A page after a few minutes of free play. Every mark on it is still a material that can go on reacting: where water was poured on the blue wash, the paint has run out soft and pale, and the egg can still be smeared or cooked. 0 Goals, scores or ways to fail. A child can make art or just make a mess. 97 Materials in the released game, all reacting through the same shared layers. 4 to 100 Steps undo keeps, as many as fit in its memory budget. 0 Learning studies of our own. We say what the design draws on, not what it proves. ## Children learn from surprises they can test, and here every surprise holds up Even babies do it. Eleven-month-olds who saw an object break their expectations learned more about it, explored it more, and tested it in ways that fit what had gone wrong [Stahl 2015](). With 3- to 6-year-olds the same researchers found better learning after a surprising event, and only about the thing that had behaved surprisingly [Stahl 2017](). Preschoolers go further. Given a toy whose causes were muddled, they chose to explore it and sorted the causes out in free play [Schulz 2007](). Given ambiguous evidence, they picked and designed little experiments that would tell them something [Cook 2011](). A review in *Science* sums it up: young children test hypotheses and learn causes from statistics, experiments and other people [Gopnik 2012](). A surprise only helps if it holds up when you poke it again. That is the one thing the game is engineered for. No line of code says where a fire must stop. Every material writes into a few shared layers over the paper, such as heat, water, what's dissolved in it and the air, and reacts to what it finds there by its own properties. So a child who paints a line of water and lights the paper beside it sees the fire held at the line until its heat has boiled the water off, because the fire reads how wet each spot of paper is, whatever made it wet. If they try again, it's held again. If they make the line thinner, the fire finds the thin part. Each new try gets an answer that follows from the same rules. The research also sets limits. A surprise is not a lesson on its own: scientists and science students have seven ways to respond to evidence that doesn't fit, and only one of them is changing their mind [Chinn 1993](). And the strongest evidence comes from children younger than most of our players. [Under the hood](<#hood-surprise>) Children learn from surprises they can test, and here every surprise holds up Under the hood ### Children learn from surprises they can test, and here every surprise holds up The demo below runs the released game's paper-and-fire rule, per cell at 60 Hz, on a 120 × 90 grid instead of up to 512 × 384: Δ*T* ∝ \(mean of the four neighbours − *T*\), faster through more fibrous paper - **Heat spreads** by plain diffusion between neighbouring cells. - **A cell lights** above paper's real ignition point of about 230 °C, if it still has fuel and is close to dry. - **A burning cell** eats its fuel, is pushed toward flame heat, and loses heat to the room by Newton cooling. - **Wet paper** at the release starts losing its water well below boiling point and damps its own heat while it does, with no latent heat yet. A burnt-out cell is a hole and conducts nothing. No cell knows what a firebreak is. The hold, the leak at the thin part and the burn-back from behind all come from these few rules. An automated test checks that soaked paper will not light. Try it: a fire meets a line of water The game's own ruleA fire is already burning on the left. The water line is thick at the top and bottom and thin in the middle. **What to look for:** the fire is held at the thick water but slips through the thin part, then spreads out behind the line and can burn back into it from the other side. In the game, given long enough, the fire boils even the thick water dry and burns on. Once the sheet has burnt away, a fresh one is laid. Pick **Water** or **Match** and draw on the paper to try your own idea, or press the buttons. Paper fibre changes how fast heat travels and how fast the paper burns. ![A fire burning across paper towards a line of water, drawn as a grid of cells]() Draw withWaterMatch Water per pass0.60 Paper fibre0.50 Speed3× hottest— burning— burnt through— water in the paper— time— ![The game: a fire has burnt the left half of the paper down to the table. At the top and bottom it stops at grey patches of wet paper where the water line is thick; in the middle, where the line was thin, the burning edge bulges on past it.]() **The same thing in the game.** A line of water poured down the paper holds the fire back where it's thick \(the grey wet patches top and bottom\) and lets it through where it's thin, in the middle. Nobody scripted it; it's the rule in the demo above, on the full sheet. ## Being shown how narrows play, but play alone teaches less, so a grown-up guides In a well-known experiment, preschoolers were given a new toy. When a teacher showed them one thing it could do, they stuck almost entirely to that one thing. Children who weren't taught explored widely. Even overhearing the teaching given to another child narrowed their play; watching it given to an adult did not [Bonawitz 2011](). The game sets no tasks and has no right answers. A child picks up the ketchup and finds out. But there is a strong other side, and a jury may know it. Across 164 studies, explicit instruction beat unassisted discovery, while discovery with feedback, scaffolding or prompts to explain beat other kinds of teaching [Alfieri 2011](). Third and fourth graders learned to design a fair experiment far more often from direct instruction [Klahr 2004](). Reviews argue that guided discovery beats pure discovery [Mayer 2004]() and that minimal guidance does not work for teaching content [Kirschner 2006](). A sandbox on its own is unassisted discovery. So we do not claim the game teaches science by itself, and we never call it the teacher. The best-supported middle is *guided play*: the child leads, in a prepared setting, with light help from an adult [Weisberg 2016](). A review of 39 studies, 17 of them pooled in a meta-analysis, with 3,893 children aged 1 to 8, found it beat direct instruction on early maths, shape knowledge and task switching, and beat free play on spatial vocabulary, with no difference on other key outcomes [Skene 2022](). Play can be seen as a spectrum from free play through guided play to games [Zosh 2018](). The game sits at the free end, on purpose. The light help comes from people. The Professor Oops website pairs its experiments, where that's safe, with a version to do for real in the kitchen, ends them with questions that ask what will happen next, and gives teachers a plan: a hook, a prediction, a try in the app, the real thing, then the explanation. That fits the research on simulations, where the evidence is strongest for simulations used to enhance ordinary lessons, especially lab work [Rutten 2012](). And paediatricians recommend play *with* parents and peers [Yogman 2018](). [Under the hood](<#hood-guided>) Being shown how narrows play, but play alone teaches less, so a grown-up guides Under the hood ### Being shown how narrows play, but play alone teaches less, so a grown-up guides **Effect sizes.** *d* and *g* are standardised differences between two groups: 0.2 is usually called small, 0.5 medium. The chart puts both on one axis; they come from different studies, outcomes and ages, so compare signs and rough sizes only. **A yardstick we score ourselves against, not one we were built on:** educational apps should be active, engaged, meaningful and socially interactive, within a supported learning goal [Hirsh-Pasek 2015](). Active and engaged: by design. Meaningful: arguable. Socially interactive: weak for a child playing alone. A learning goal: absent on purpose. None of it measured. **What "guided play" means in the literature:** a learning goal, a scaffolded setting and the child in control [Weisberg 2013](), with an adult giving the scaffolding [Weisberg 2016](). The game has no goal, so we don't call it guided play. Discovery alone loses to teaching; guided play wins on some outcomesFrom the left: unassisted discovery vs explicit instruction \(580 comparisons\) and enhanced discovery vs other instruction \(360\), Alfieri 2011; guided play vs direct instruction on early maths, task switching and shapes, and vs free play on spatial vocabulary, Skene 2022. Unassisted discovery against explicit instruction is the one negative bar \(d −0.38\); enhanced discovery \(d \+0.30\) and guided play on early maths \(g 0.24\), task switching \(0.40\), shapes \(0.63\) and spatial vocabulary \(0.93\) come out ahead. Data table Discovery alone loses to teaching; guided play wins on some outcomes | | effect size | | --- | --- | | Discovery alone | −0.38 | | Enhanced discovery | 0.30 | | Early maths | 0.24 | | Task switching | 0.40 | | Shapes | 0.63 | | Spatial words | 0.93 | Source: Effect sizes from the abstracts of [Alfieri 2011]() \(164 studies\) and [Skene 2022]() \(17 studies, ages 1 to 8; no differences on other key outcomes\). ![Painted illustration in two halves: a girl doing magic milk on a tablet, and the same girl doing it for real with a dish of milk while a grown-up's hand helps.]() **On screen, then for real.** The help the research favours comes from a grown-up nearby. The website gives each experiment a kitchen version to do together. ## No points, stars or streaks, because promised rewards can dull the wish to play In a classic study, preschoolers who drew in order to earn an award they had been promised later showed less interest in drawing than children who got an unexpected award or none [Lepper 1973](). A review of 128 experiments found the same pattern: rewards for taking part, for finishing or for performing lowered the wish to do the activity freely, and tangible rewards were worse for children. Positive feedback did the opposite and raised it, though less for children [Deci 1999](). A rival review argues the harm is narrower: it shows up on interesting tasks when rewards are tangible, expected and loosely tied to how well you did [Cameron 2001](). Professor Oops has no points, stars, badges, streaks, levels or "game over". Nothing is locked or teased in the child's view: tools the family hasn't bought are hidden from the child, and the door to them is a small, plain tile for grown-ups. There are no ads. We decided that after reading that 95 % of apps for children under five carried ads. The study behind that figure also found that apps labelled "educational" were no better [Meyer 2019](). More on that in [Designed with respect for children](). We went further than the evidence. Research says positive feedback can raise motivation, and some researchers find the harm of rewards much narrower; our rule leaves out rewards of every kind anyway. That is a choice, not a finding. Play also ends calmly. A grown-up can set a daily limit of 15, 30 or 45 minutes or an hour. It is off unless they pick one. Two minutes before the end, the professor says so once. At the end the game rests behind "Time for a break\!", never in the middle of a stroke, with the drawing kept. The break screen has no timer, no streak and nothing to come back for. The limit wasn't drawn from screen-time research. It exists because Brazil's ECA Digital asks for parental time tools. Paediatric advice itself now looks beyond screen limits, at how the media is designed [Munzer 2026](), and the WHO's limits cover children under five [WHO 2019](). [Under the hood](<#hood-rewards>) No points, stars or streaks, because promised rewards can dull the wish to play Under the hood ### No points, stars or streaks, because promised rewards can dull the wish to play The play clock is kept on the device only: - A grown-up picks 15, 30, 45 or 60 minutes a day, or no limit, the default. - Only time with the game in front counts, and one tick can only count a few seconds, so a device that slept doesn't eat the time. - The two-minute warning is said once; it comes again only when a grown-up changes the limit. - "15 more minutes" sits behind the parental gate and counts from now, however long the break screen was up. - The count starts again at local midnight. Each of these is pinned by an automated test, such as the one that checks a device that slept doesn't eat the time. Expected rewards lower the wish to play; feedback raises itRewards for taking part \(engagement-contingent\), for finishing \(completion-contingent\) and for doing well \(performance-contingent\); positive feedback, less so for children. The game has none of these rewards, and no scores either. Rewards for taking part \(d −0.40\), finishing \(−0.36\) and doing well \(−0.28\) lowered the wish to play freely; positive feedback raised it \(d \+0.33\). Data table Expected rewards lower the wish to play; feedback raises it | | effect size d | | --- | --- | | For taking part | −0.40 | | For finishing | −0.36 | | For doing well | −0.28 | | Positive feedback | 0.33 | Source: Meta-analysis of 128 experiments: [Deci 1999](). Disputed in size and reach by [Cameron 2001](). Try it: a day with the play-time limit The game's own ruleA child is drawing, and the clock runs fast. **What to look for:** the break waits for the stroke to finish; putting the game in the background stops the count; a device that sleeps for an hour costs only a few seconds; and a grown-up's 15 minutes count from now. Set the limit to Off and the game never ends. ![A drawing in progress above a bar showing the day's play time against the limit]() - In the background: not counting - played today - \+ 15 min - last 2 min: warned once - no limit: the game never ends - The professor says it's nearly time - Time for a break\! - The drawing is kept. Daily limitOff15 min30 min45 min1 hour Clock speed30× played— left— state— game in front— slept 1 h, counted— ## Every mess can be taken back, so the extreme thing is safe to try A toy for setting paper on fire only works if the fire never costs a child their drawing. The review that shaped the game found a designer describing a child in tears after erasing a drawing by accident. Its conclusion became a rule: destruction must be undoable. In Professor Oops, undo always goes back at least four touches, and usually far more. A step is everything that changed from one touch to the next, including the paint drying and soaking in. If a child leaves a fire burning and skips ahead three days, the fire stops during the wait, so they never come back to a page that burned away while they weren't looking. This came from product experience, not from a paper. We know of no study on undo buttons and children's learning, so we don't cite one. What we can say is that it keeps the extreme thing safe to try, which is the whole point of a game about doing what you're normally told not to do with paper. [Under the hood](<#hood-undo>) Every mess can be taken back, so the extreme thing is safe to try Under the hood ### Every mess can be taken back, so the extreme thing is safe to try Undo keeps one full copy of the page and, for each older step, only the parts of the page that changed. When a new step comes, the oldest steps let go until what's held fits a fixed memory budget, never below four steps and never above a hundred. On the budget tablet's 384 × 288 sheet we measured about 0.6 MB for a stroke on a dry page, about 3.5 MB for one on a wet, still-running page, about 1 MB for a pour and 1 to 2.5 MB for a time skip. The four full copies undo held before it saved only the changes came to about 80 MB. Automated tests pin both promises: undo puts back exactly what full copies did, and the oldest steps are the ones let go. Another checks that a fire stops during a time skip. Try it: how far back can a child go? The game's own ruleA pretend child is playing; each block is one undo step, oldest on the left, sized by its memory: a stroke on a dry page 0.6 MB, a stroke on a wet page 3.5 MB, a pour 1 MB, a time skip 2.5 MB. The dark line under the oldest blocks marks the 4 steps always kept. **What to look for:** dry strokes are small, so dozens fit; a run of wet strokes fills the budget fast and the oldest steps drop off, but never below four. Add steps yourself, tap Undo, or shrink the budget to the paint board's smaller one \(its sheet is smaller, so its steps are too\). ![A row of undo steps sized by memory, with the oldest dropping off the left when the budget is full]() - memory budget - ← oldest - newest → - dry stroke - wet stroke - pour - time skip - always kept Memory budget40 MB steps you can undo— held— oldest let go— How many steps the table’s memory budget holds, by kind of stepStrokes on a dry page, about 0.6 MB a step; pours, about 1 MB; time skips, 1 to 2.5 MB \(the bar uses 2.5\); strokes on a wet, still-running page, about 3.5 MB. "About 65" and "about 11" are the measured figures. Undo never keeps fewer than 4 steps, the floor it always keeps. About 65 strokes on a dry page fit, 40 pours, 16 time skips and about 11 strokes on a wet, still-running page; undo never keeps fewer than 4. Data table How many steps the table’s memory budget holds, by kind of step | | steps | | --- | --- | | Dry strokes | 65 | | Pours | 40 | | Time skips | 16 | | Wet strokes | 11 | | Never fewer than | 4 | Source: Step sizes measured on the budget tablet's 384 × 288 sheet. Counts: the table's memory budget divided by the step size, as the game's rule gives. ## The research mostly studied other children, so we claim little Few of these studies looked at children of our players' age, and none tested learning in a sandbox on a tablet. The surprise studies tested infants and preschoolers. The teaching study tested preschoolers with a toy. The classic reward study tested preschoolers. Our public age range is about 4 to 10, and on Apple's App Store the game sits in the 6 to 8 band. The chart shows how little overlap there is. So we read this research as direction, not proof. Here is what that rules out. - **We don't say the game teaches.** No study, no control group, no learning measure. A child who just plays gets nothing explained, and the discovery research says that alone teaches content weakly. - **We don't say "educational app".** The label means little: apps sold as educational were no better on ads [Meyer 2019](), and the science of learning sets a much higher bar for an app than the label does [Hirsh-Pasek 2015](). - **We don't say drawing helps children learn science.** Drawing as a learning strategy looks largely ineffective before secondary school [Brod 2021](). Art is the reason to play here, not a proven route to learning. - **We don't say "play builds" a skill.** Even for pretend play, the evidence doesn't support strong causal claims [Lillard 2013](), and the paediatric case for play is about play with people [Yogman 2018](). - **We don't say rewards always kill motivation.** The size and reach of that effect are disputed [Cameron 2001](). Our no-rewards rule is a design choice. Who the studies tested, against our playersStahl 2015: 11 months. Cook, Schulz, Bonawitz: mean ages of 54, 57, and 58 and 61 months. Not shown: Klahr & Nigam \(grades 3 and 4\), Lepper \(preschoolers\), Deci \(children and adults\). Only Skene 2022 \(ages 1 to 8\), Stahl 2017 \(3 to 6\), Radesky 2022 \(3 to 5\) and the WHO \(under 5\) reach into our players’ 4 to 10; the infant and preschool means sit at 11 months to about 5 years. - youngest age studied \(or the mean\) - oldest age studied \(or the mean\) Data table Who the studies tested, against our players | | youngest age studied \(or the mean\) | oldest age studied \(or the mean\) | | --- | --- | --- | | Stahl 2015 | 0.9 | 0.9 | | Cook 2011 | 4.5 | 4.5 | | Schulz 2007 | 4.8 | 4.8 | | Bonawitz 2011 | 4.8 | 5.1 | | Stahl 2017 | 3.0 | 6.0 | | Skene 2022 | 1.0 | 8.0 | | Radesky 2022 | 3.0 | 5.0 | | WHO 2019 | 0.0 | 5.0 | Source: Ages from each source's abstract or full text \(see the list below\). Our range: "about 4 to 10" on the Professor Oops website; 6 to 8 is Apple's App Store age band. ## A ball blows out a candle and ash turns cabbage juice blue, because the systems share the page Every surprise a child can follow up on this page comes from systems that share their layers. These are some of the chains a child can find, each held in place by a test. Shared: wet \+ temp ### Paint a line of water in front of a fire The paper won't light until the water has gone. *[Heat and fire]()* · pinned by a test Shared: carbonate \(ash\) ### Burn paper, then drip red cabbage juice on the ash The purple dye turns blue in a thin ring round the ash. *[Chemistry and pH]()* · pinned by a test Shared: air speed ### Roll a football fast past a lit candle It blows out; rolled slowly, it doesn't. *[Air that really moves]()* · pinned by a test Shared: heavy gas \(CO₂\) ### Pour vinegar on baking soda near a candle The candle goes out with no gust; further off it stays lit. *[Built for surprises]()* · pinned by a test Shared: every layer, saved as changes ### Make a mess, then tap undo Exactly the page you had, a few touches back. *[How we keep it right]()* · pinned by a test Shared: play clock ### The day's play time runs out mid-stroke The professor says so two minutes before, then the game rests, with the drawing kept. *[Designed with respect for children]()* · pinned by a test ## Any mess can be undone and no grown-up guide is built in, each a deliberate choice Some of these are simplifications of the world; some are choices that go further than the research. Each one is deliberate. **In real life or in the research.** Burnt or soaked paper stays ruined. **In the game.** Undo keeps 4 to 100 steps within a fixed memory budget, saving only what changed; a fire stops during a time skip. **Why that's enough.** The extreme thing has to be safe to try. What's lost: real consequences. **In real life or in the research.** Wet paper in a fire is held near 100 °C until its water has boiled off, by latent heat. **In the game.** At the release, wet paper starts drying well below 100 °C and simply damps the heat, with no latent heat. next update holds wet paper near boiling with a latent-heat term built from water's real values. **Why that's enough.** A child sees the same thing: the fire is held at the water and goes on when it's dry. The physics is now being made right as well. **In real life or in the research.** Guided play has a learning goal and an adult who gives the scaffolding. **In the game.** No goal is shown and no adult is built in. Help lives outside the app, on the website's kitchen versions and teacher plans. **Why that's enough.** The child leads. What's lost is the guidance the research says helps most, so we claim no teaching effect. **In real life or in the research.** Positive feedback can raise motivation \(*d* = \+0.33 across studies\). **In the game.** No rewards, scores, stars or streaks at all. **Why that's enough.** Stricter than the evidence. A reward for an activity children already love is close to the case where rewards do harm. It's a choice, and we say so. **In real life or in the research.** Screen-time guidance differs by age and country, and the AAP's latest statement, as reported, sets no fixed limit. **In the game.** An optional daily limit of 15 to 60 minutes, off by default; only time in front counts, and a sleeping device can't eat the time. **Why that's enough.** A tool for grown-ups, not a judgement. It answers a legal requirement in Brazil, not a guideline. **In real life or in the research.** The studies tested infants, preschoolers, older pupils and adults, in labs and classrooms. **In the game.** We apply their direction to children of about 4 to 10 on a tablet. **Why that's enough.** Nobody has tested a tablet sandbox at these ages. We claim fit, never effect. ## Each study is listed, with whether it shaped the game or was found afterwards Each bibliographic entry was checked against its DOI record and its abstract, or the page cited. The label says how the source relates to the game you can play today \(see the definitions near the top\). ### Surprise and exploring - Stahl, A. E., & Feigenson, L. \(2015\). Observing the unexpected enhances infants' learning and exploration. Science 348\(6230\), 91–94.[doi:10.1126/science.aaa3799]()Found later110 eleven-month-olds learned better about objects that broke their expectations, explored them more, and tested them in ways that fit the violation. Infants, far from our players. - Stahl, A. E., & Feigenson, L. \(2017\). Expectancy violations promote learning in young children. Cognition 163, 1–14.[doi:10.1016/j.cognition.2017.02.008]()Found laterAges 3 to 6, four experiments: learning was better after a surprising event, and only about the thing that behaved surprisingly. Closer in age than the infant study. - Schulz, L. E., & Bonawitz, E. B. \(2007\). Serious fun: Preschoolers engage in more exploratory play when evidence is confounded. Developmental Psychology 43\(4\), 1045–1050.[doi:10.1037/0012-1649.43.4.1045]()Found laterMean age 57 months: children preferred to explore toys whose causes were confounded, and sorted the causes out in free play. - Cook, C., Goodman, N. D., & Schulz, L. E. \(2011\). Where science starts: Spontaneous experiments in preschoolers' exploratory play. Cognition 120\(3\), 341–349.[doi:10.1016/j.cognition.2011.03.003]()Found laterMean age 54 months: given ambiguous evidence, children chose and designed informative interventions. - Gopnik, A. \(2012\). Scientific thinking in young children: Theoretical advances, empirical research, and policy implications. Science 337\(6102\), 1623–1627.[doi:10.1126/science.1223416]()Found laterA review: preschoolers test hypotheses, make causal inferences and learn from statistics, experiments and other people. - Chinn, C. A., & Brewer, W. F. \(1993\). The role of anomalous data in knowledge acquisition: A theoretical framework and implications for science instruction. Review of Educational Research 63\(1\), 1–49.[doi:10.3102/00346543063001001]()*Pushes back.* Seven responses to anomalous data, only one of which is changing the theory. About scientists and science students. A surprise on paper may simply be shrugged off. ### Teaching, discovery and guided play - Bonawitz, E., Shafto, P., Gweon, H., Goodman, N. D., Spelke, E., & Schulz, L. \(2011\). The double-edged sword of pedagogy: Instruction limits spontaneous exploration and discovery. Cognition 120\(3\), 322–330.[doi:10.1016/j.cognition.2010.10.001]()Found laterPreschoolers \(85 and 64 children, mean ages 58 and 61 months\) shown one function of a toy by a teacher stuck to it; untaught children explored widely. Overheard teaching narrowed play too. First cited on the website, after the game's design. - Alfieri, L., Brooks, P. J., Aldrich, N. J., & Tenenbaum, H. R. \(2011\). Does discovery-based instruction enhance learning? Journal of Educational Psychology 103\(1\), 1–18.[doi:10.1037/a0021017]()Found later164 studies. Explicit instruction beat unassisted discovery \(*d* = −0.38, 580 comparisons\); enhanced discovery beat other instruction \(*d* = \+0.30, 360 comparisons\). Cited on the website. The clearest reason a sandbox alone shouldn't be called a teacher. - Klahr, D., & Nigam, M. \(2004\). The equivalence of learning paths in early science instruction. Psychological Science 15\(10\), 661–667.[doi:10.1111/j.0956-7976.2004.00737.x]()*Pushes back.* 112 third and fourth graders: many more learned experimental design from direct instruction than from discovery. - Mayer, R. E. \(2004\). Should there be a three-strikes rule against pure discovery learning? American Psychologist 59\(1\), 14–19.[doi:10.1037/0003-066X.59.1.14]()*Pushes back.* Guided discovery beat pure discovery. Pushes back on any claim for the sandbox alone. - Kirschner, P. A., Sweller, J., & Clark, R. E. \(2006\). Why minimal guidance during instruction does not work: An analysis of the failure of constructivist, discovery, problem-based, experiential, and inquiry-based teaching. Educational Psychologist 41\(2\), 75–86.[doi:10.1207/s15326985ep4102\_1]()*Pushes back.* Argues that minimally guided instruction is less effective for teaching content. Pushes back on any claim that a sandbox teaches by itself. - Weisberg, D. S., Hirsh-Pasek, K., & Golinkoff, R. M. \(2013\). Guided play: Where curricular goals meet a playful pedagogy. Mind, Brain, and Education 7\(2\), 104–112.[doi:10.1111/mbe.12015]()Found laterDefines guided play as midway between free play and instruction: a learning goal, a scaffolded setting, the child in control. The game presents no learning goal. - Weisberg, D. S., Hirsh-Pasek, K., Golinkoff, R. M., Kittredge, A. K., & Klahr, D. \(2016\). Guided play. Current Directions in Psychological Science 25\(3\), 177–182.[doi:10.1177/0963721416645512]()Found laterAutonomy within a prepared environment, with adult scaffolding. - Skene, K., O'Farrelly, C. M., Byrne, E. M., Kirby, N., Stevens, E. C., & Ramchandani, P. G. \(2022\). Can guidance during play enhance children's learning and development in educational contexts? A systematic review and meta-analysis. Child Development 93\(4\), 1162–1180.[doi:10.1111/cdev.13730]()Found later39 studies, 17 in the meta-analysis, 3,893 children aged 1 to 8. Guided play beat direct instruction on early maths \(*g* = 0.24\), shape knowledge \(0.63\) and task switching \(0.40\), and free play on spatial vocabulary \(0.93\); no differences on other key outcomes. Studies defined guided play very differently. - Zosh, J. M., et al. \(2018\). Accessing the inaccessible: Redefining play as a spectrum. Frontiers in Psychology 9.[doi:10.3389/fpsyg.2018.01124]()Found laterPlay as a spectrum from free play through guided play to games. The game sits at the free end. - Hirsh-Pasek, K., Zosh, J. M., Golinkoff, R. M., Gray, J. H., Robb, M. B., & Kaufman, J. \(2015\). Putting education in "educational" apps: Lessons from the science of learning. Psychological Science in the Public Interest 16\(1\), 3–34.[doi:10.1177/1529100615569721]()Found laterFour pillars \(active, engaged, meaningful, socially interactive\) within a supported learning goal. Fits in part: social interaction is weak for a child alone, and the learning goal is deliberately absent. - Rutten, N., van Joolingen, W. R., & van der Veen, J. T. \(2012\). The learning effects of computer simulations in science education. Computers & Education 58\(1\), 136–153.[doi:10.1016/j.compedu.2011.07.017]()Found laterRobust evidence that simulations can enhance traditional instruction, especially lab activities; most of that research ignored teacher support and the lesson around the simulation. Fits "on screen, then for real"; pushes back on a simulation as the only teacher. - Resnick, M., & Silverman, B. \(2005\). Some reflections on designing construction kits for kids. IDC '05, 117–122.[doi:10.1145/1109540.1109556]()Found laterDesign principles, including "low floor and wide walls", "make powerful ideas salient, not forced" and "support many paths, many styles". The game resembles them; it wasn't based on them. ### Rewards and motivation - Lepper, M. R., Greene, D., & Nisbett, R. E. \(1973\). Undermining children's intrinsic interest with extrinsic reward: A test of the "overjustification" hypothesis. Journal of Personality and Social Psychology 28\(1\), 129–137.[doi:10.1037/h0035519]()Found laterPreschoolers who drew to get an expected award later showed less interest than those given an unexpected award or none. The game's no-rewards rule predates our reading of it. - Deci, E. L., Koestner, R., & Ryan, R. M. \(1999\). A meta-analytic review of experiments examining the effects of extrinsic rewards on intrinsic motivation. Psychological Bulletin 125\(6\), 627–668.[doi:10.1037/0033-2909.125.6.627]()Found later128 studies. Engagement-, completion- and performance-contingent rewards undermined free-choice motivation \(*d* = −0.40, −0.36, −0.28\); tangible rewards were worse for children. Positive feedback enhanced it \(*d* = \+0.33\), less so for children. - Cameron, J., Banko, K. M., & Pierce, W. D. \(2001\). Pervasive negative effects of rewards on intrinsic motivation: The myth continues. The Behavior Analyst 24\(1\), 1–44.[doi:10.1007/BF03392017]()*Pushes back.* Rewards in general don't harm motivation; negative effects appear on high-interest tasks when rewards are tangible, expected and loosely tied to performance. Tempers "rewards undermine". ### Screens, ads and families - Meyer, M., Adkins, V., Yuan, N., Weeks, H. M., Chang, Y.-J., & Radesky, J. \(2019\). Advertising in young children's apps: A content analysis. Journal of Developmental & Behavioral Pediatrics 40\(1\), 32–39.[doi:10.1097/DBP.0000000000000622]()Informed129 of 135 apps for ages 5 and under had ads: all of the free ones, 88 % of the paid ones; apps labelled educational were no better. Read through press coverage in the prior-art review; behind the game's no-ads rule. It counts ads; it doesn't measure harm. - Radesky, J., Hiniker, A., McLaren, C., Akgun, E., Schaller, A., Weeks, H. M., Campbell, S., & Gearhardt, A. N. \(2022\). Prevalence and characteristics of manipulative design in mobile applications used by children. JAMA Network Open 5\(6\), e2217641.[doi:10.1001/jamanetworkopen.2022.17641]()Informed160 children aged 3 to 5 and the 133 apps they used: only 26 apps \(19.6 %\) had no manipulative features. Read through a press release while the business rules were written; it shaped those rules, not the play. - Yogman, M., Garner, A., Hutchinson, J., Hirsh-Pasek, K., Golinkoff, R. M., et al. \(2018\). The power of play: A pediatric role in enhancing development in young children. Pediatrics 142\(3\).[doi:10.1542/peds.2018-2058]()Found laterDevelopmentally appropriate play with parents and peers builds the skills behind executive function. It is about play with people, not about apps. - Lillard, A. S., Lerner, M. D., Hopkins, E. J., Dore, R. A., Smith, E. D., & Palmquist, C. M. \(2013\). The impact of pretend play on children's development: A review of the evidence. Psychological Bulletin 139\(1\), 1–34.[doi:10.1037/a0029321]()*Pushes back.* Existing evidence does not support strong causal claims for pretend play's unique importance. Tempers every "play builds X" sentence, ours included. - Brod, G. \(2021\). Generative learning: Which strategies for what age? Educational Psychology Review 33\(4\), 1295–1318.[doi:10.1007/s10648-020-09571-9]()*Pushes back.* Predicting and testing seem to work from lower elementary school; drawing and self-questioning look largely ineffective until secondary school. This is drawing to learn content, not art play, but it rules out "drawing helps them learn the science". - Takeuchi, L., & Stevens, R. \(2011\). The new coviewing: Designing for learning through joint media engagement. Joan Ganz Cooney Center and LIFE Center.[joanganzcooneycenter.org/wp-content/uploads/2011/12/jgc\_coviewing\_desktop.pdf]()Found laterDesigning media for children and adults to use together. Fits the website's kitchen versions; nothing in the game pulls a grown-up in during play. - World Health Organization \(2019\). Guidelines on physical activity, sedentary behaviour and sleep for children under 5 years of age.[who.int/publications/i/item/9789241550536]()Found laterAt ages 2 to 4, sedentary screen time of no more than an hour, less being better; at age 1, not recommended. Covers under-fives only. The game's limit was not derived from it. - Munzer, T., et al. \(2026\). Digital ecosystems, children, and adolescents: Policy statement. Pediatrics 157\(2\).[doi:10.1542/peds.2025-075320]()Found laterMedia use can't be judged by screen limits alone; engagement-first and commerce-first design is the concern. Read for context, not used for design, and it endorses no product. ### Product precedents \(not research\) - Earlier creative apps and sandbox games for children, among them Kid Pix, Tux Paint and falling-sand games such as Noita and Sandspiel, reviewed on 16 October 2026.Applied*Applied directly.* The actual source of the sandbox: no rules, no failure, no wrong move; destruction as play; simple per-cell rules that give emergent stories; an undo a child can trust. Taken from press, reviews and parents' reports. These are product lessons, not evidence that anything teaches.