It teaches a child to count, to read, and to stay safe. It collects nothing .
A learning world of nine worlds and twenty six skills, mapped to public standards, that works with no signal and stores nothing about the child playing it.
Numeracy · Phonics · Time & money · Personal safety
Children's software asks you to choose between knowing nothing and giving everything.
Most learning apps for young children fall into one of two shapes. Either nobody can tell you what the child is actually being taught, because the content answers to a roadmap rather than to evidence. Or the app knows a great deal about your child, because accounts, cloud sync, and analytics were the easiest way to build it.
Both of those are decisions. Neither is a constraint. We built Urumuri to demonstrate that a children's product can be accountable about its content and incurious about its user at the same time.
Nine worlds, and a dragon the child raises by getting better.
- Twenty six skills across five levels. Counting, place value, coins, doubles, halves, shapes, addition and subtraction, clocks, phonics and sight words, and running a small stand. First grade at the floor, third grade at the ceiling, advancing per world so nothing jumps ahead of the child.
- Mapped to public standards. Every core skill carries its Common Core code, so a teacher, a tutor, or a parent can read the progress report against the same standards the school uses. No proprietary taxonomy to decode.
- Every question is read aloud. From recorded audio, not synthesized speech, because iOS silences speech synthesis inside installed home-screen apps. A reading world that cannot be heard is a reading world that cannot be answered.
- It works with no signal. The whole game runs offline after one online load. Built for a back seat, a waiting room, and a school without reliable wifi.
- Mastery is earned and never taken back. No timers anywhere. A wrong answer is never punished, it quietly returns later. A streak is never removed. A child cannot fall behind inside it.
- A grown-up's board behind a PIN. Skill-by-skill progress, a reading-versus-maths diagnostic, an answer-speed read that separates fluent from effortful, and a printable report for the child's teacher.
- It takes the child's name. The title becomes theirs and the companion greets them by it. Progress moves between devices by a transfer code rather than an account.
Research decides the content. Not the roadmap.
Three research passes. Each one changed the build.
Research that confirms what you already shipped is decoration. Ours is auditable, and it is written down, because each pass forced a change.
- Developmental learning. Validated an hour-first approach to teaching clocks, then found the most common misconception was missing from the question set entirely: children read the long hand as the hour. It was added as a deliberate distractor and now appears in roughly half of all clock questions.
- Financial literacy. Killed a question that had already shipped. It taught a savings rate no framework anywhere prescribes, and implicitly claimed to train self-control, which the largest simulated-economy trial measured at exactly zero effect. It was replaced with safe-place and goal-gap arithmetic, which are defensible.
- Business and entrepreneurship. Profit is permanently excluded. Children do not integrate buying and selling until around age eleven, so money is not yet a measure of value at six, and the software makes no pricing judgments at any level. Two thousand four hundred generated business questions were checked to confirm zero appear.
The limits are written down, and we publish them.
Urumuri may claim vocabulary, recognition, procedure, and arithmetic in context. It may not claim that it teaches patience, planning, delayed gratification, or an entrepreneurial mindset. Those claims are common in children's software and the evidence for them does not exist. Ours is written into the source, so it survives whoever builds the next feature.
We make no claims about learning outcomes. There is no efficacy study behind this product, and until there is one we will not imply otherwise.
It collects nothing. There are no accounts, no server, and no analytics. Progress lives in the browser on the child's own device, which is why the privacy obligations most children's software carries do not apply here. That is an architectural decision, not a policy promise, and it is the harder of the two to break.
Twenty seven thousand questions, generated and checked, before every release.
A learning game fails quietly. The screen looks correct, the child answers, and the logs read like success. So every deploy generates twenty seven thousand questions across every world and every level and asserts what must always hold: three distinct choices, the right answer among them, a teaching line, and a spoken line. The build fails on a single violation.
It earns its place. That sweep caught questions that could collapse to two choices at certain values, which turns a question into a coin flip. Since a skill is judged mastered on eight of the last ten answers, guessing could have registered as mastery on a report handed to a teacher. It was invisible in play and obvious to the sweep.