Study Garden

The research

The research behind Study Garden

Each finding below changed how Study Garden works. For each one, you'll see what the study found, what we built because of it, where to find it and where its evidence stops.

No study has tested Study Garden itself. We haven't measured its effect on test scores yet, and we'll share what we learn.

When you're stuck

Answers on tap can backfire

In a study of nearly 1,000 high schoolers, chatbot answers raised practice scores but lowered exam scores. Hints instead of answers avoided that.

What we built
Your tutor gives hints and explanations, and a full solution if you ask. An answer you give after help is saved as practice, never as passed.
Find it
Master TutorA hint, please
Where the evidence stops
One study, in high school math. The hint version avoided the harm but didn't raise exam scores. Study Garden's tutor wasn't part of it.

A tutor off to the side gets little use

In a two-year trial in 18 Tennessee middle schools, students had an AI tutor beside their math practice. The typical student messaged it on only a third of the days they practiced. A review of learning software found that students often don't use the help on offer well, though using it well goes with learning more.

What we built
Master Tutor is the first screen you see. Today's suggestions are already on it, and so is the box to ask it.
Find it
Master Tutor
Where the evidence stops
That trial's tutor added little over the same practice without AI, and its authors point to how rarely it was used. No study has tested putting the tutor first.

Checking each step works better than checking answers

In a review of decades of studies, tutors that checked each step of the thinking came close to human tutors. Tutors that checked only final answers did far less. And in a 2026 trial, AI help worked best right after a mistake.

What we built
When you're stuck, your tutor finds the concept you're missing, one small idea inside the topic. It asks concept checks, short questions on just that concept, until you pass 3 different ones. Then you go back to the exercise.
Find it
Master TutorConcept check
Where the evidence stops
Concept checks are our way of checking steps, and no study has tested them. Tutors that checked even smaller steps did no better. 3 checks is our starting rule, not a measured one.

Moving up a level

Doing well in practice isn't the same as learning

A review of decades of memory and movement research found that how well practice goes is often a poor guide to lasting learning. People take it as one anyway.

What we built
Your record keeps the two apart. Practice with help is saved and shown, but only an exercise you pass on your own, first try, counts toward a checkpoint.
Find it
My growth
Where the evidence stops
Help matters too, so it's always there. One first try is a noisy measure on its own, which is why a checkpoint takes 3 of 4.

Showing it before moving on helps, most for students who struggle

A review of 108 controlled studies found that mastery learning, where students move on only after showing they've learned, raised exam results. The gains were largest for weaker students.

What we built
Each topic has three levels: Easy, Medium and Difficult. A checkpoint is 3 of your next 4 new exercises at your level, on your own, first try. Passing it moves you up a level.
Find it
My growthCheckpoint
Where the evidence stops
A review of school programs found little effect on standardized tests. In a 2026 trial, requiring three right in a row didn't by itself improve learning a week later. No study set 3 of 4; it's our rule.

Passing once isn't knowing

In a 2026 trial, requiring three right in a row didn't by itself improve learning on a later test. In another study, rereading won after five minutes, but testing yourself won after two days.

What we built
A concept is secure only after you pass 3 different concept checks on your own, across at least two sessions. After 30 days without one, it shows Review due.
Find it
My growthConcepts
Where the evidence stops
One later success is still one observation, so secure never means learned for good.

A few answers can show where to start

With students in grades 9 and 10, a quick test of algebra and geometry matched full tests closely, in a fraction of the time. In urban India, software that taught each middle schooler at their own level raised math scores in four and a half months.

What we built
On a new topic, your tutor picks a starting level from your levels in related topics. Over your first 3 answers on your own, it moves that level up or down.
Find it
My growthFinding your level
Where the evidence stops
Three answers carry little information, so the starting level is a first guess that your work corrects. The program in India ran in after-school centers with staff.

Prizes you expect can lower interest, and feedback can raise it

A review of 128 experiments found that expected rewards lowered later interest in a task, most when given just for taking part. Positive feedback raised it. Prizes tended to hurt children more than college students.

What we built
Practice celebrates only what you pass on your own. A level-up card says what you passed and when, never how many tries it took.
Find it
My growthShare
Where the evidence stops
Even rewards for doing well lowered interest a little. Study Garden also counts showing up: each day you practice grows your practice garden, beside your level-ups.

Remembering it later

Remembering beats rereading

A review of ten study techniques rated two the most useful: testing yourself, and spreading practice over days. Rereading rated low. A review of 222 classroom studies found that quizzes raised achievement.

What we built
Recall cards ask you to write what you remember before you see the answer. How it felt sets when the card comes back, later each time you remember it.
Find it
Master TutorToday
Where the evidence stops
The benefit can be smaller for complex material, and seeing the answer after each try matters. So every card shows its answer once you've tried.

Feeling you've learned it is a poor guide

In college physics, students in active classes learned more but felt they learned less. In another study, students who reread felt surer they'd remember, yet remembered less a week later.

What we built
Test myself is a short quiz or an essay you do on your own, with your tutor closed. It shows what you know without help.
Find it
Test myself
Where the evidence stops
A test teaches most when you see what you missed afterward. Test myself isn't ready for every study yet.

Your study plan

The best gap between reviews grows with the time until the test

More than 1,350 people learned facts, reviewed them after a gap, then took a test up to a year later. The best gap grew with the time left until the test.

What we built
Your study plan schedules a topic's next level-up 7 days after its last one, and other topics fill the days between. When the test is near, the wait gets shorter.
Find it
Master TutorYour path to the test
Where the evidence stops
The study taught facts. Methods for solving problems may space differently, and the 7 days are our choice.

People plan too hopefully

Students guessing when they'd finish their thesis were too hopeful, even in their worst-case guess. Across 27 sets of practice data, students started in very different places but learned at similar rates.

What we built
Your path to the test counts the minutes each level actually took you. Until 3 level-ups are measured, it uses a typical 30 minutes a level, and says so.
Find it
Master TutorYour path to the test
Where the evidence stops
Minutes are estimates from your saved activity. The similar rates were counted per practice question, while your plan counts minutes.

A goal for today works better than one far away

Children who struggled with arithmetic progressed faster with a small goal each session than with one distant goal, which showed no effect. Too many options make choosing harder when you're unsure what you want.

What we built
On Master Tutor, Today shows at most 3 suggestions for today, above your path to the test.
Find it
Master TutorToday
Where the evidence stops
One small study of children, from 1981. Too many options don't always make choosing harder; it depends on the situation.

Lightning mode

Timed practice helps, once you know the method

A federal review of 27 studies of math help for struggling students rated timed practice as strong evidence for fluency. Fluency means knowing facts and steps well enough to use them quickly. It advises short rounds on what students already know, beating your own score, and keeping progress private.

What we built
Lightning mode opens only after you pass a topic's Difficult checkpoint with no clock. Its new Difficult exercises each come with a target time. Only you see your times, unless you share your card.
Find it
My growthLightning mode
Where the evidence stops
The studies were in elementary school, and timing was one part of larger programs.

Pressure hurts the students most ready to do well

Under pressure, adults with the most working memory, how much you can hold in mind at once, did worse on the hardest math problems. In grades 1 and 2, math anxiety went with lower scores, most for children with strong working memory. Across 238 studies, test anxiety went with lower results, most in the middle grades.

What we built
In Lightning mode the clock stays on screen and never stops you. Your tutor never tells you to hurry, and praises your method, never you.
Find it
Lightning mode
Where the evidence stops
None of these studies tested what a tutor says.

A quick mistake and a slow one are different mistakes

In large sets of answers and their times, from arithmetic to chess, the errors in quick answers differed clearly from the errors in slow ones.

What we built
In Lightning mode, feedback follows your result and your time. A quick miss is probably a slip, so your tutor suggests reading the question again. A slow miss gets the step taught again.
Find it
Lightning mode
Where the evidence stops
The study described answers and times. It didn't test feedback.

Harder problems reward more time

In a large test of adults on computers, taking longer went with better problem solving, and more so on harder problems.

What we built
Your tutor says the test's time is an average, so some questions take longer. If you pass over the time, it asks how you did it and shows a shorter way beside yours.
Find it
Lightning mode
Where the evidence stops
The study tested adults on computer tasks. Students on test questions may differ.

Finding what's new

People find new features by running into them

Researchers interviewed 18 programmers and asked 76 people at work to keep a diary. Coming across a tool in the software they used was the most common way to find one.

What we built
What's new is there when you look, in the app's menu and on the home page, and never pops up. Each new feature's entry names the buttons to tap, and Try it opens it.
Find it
What's new
Where the evidence stops
The study looked at people using software at work. Hearing about a tool from someone else was rarer, but worked well.