Frequently Asked Questions

At the early levels, a level is defined by which letter-sound patterns are available to the story — not by an external score. We're building that sequence from the bottom up, so the upper levels are still shaped by sentence length and vocabulary while we get the foundational ones right.

Once a child knows /s/ /a/ /t/ /p/ /i/ /n/, they can read sat, pat, tip, nip, pit, sit, pin, tan — and nothing else. Every new pattern widens the vocabulary a story can draw on, which is what makes each level feel different from the last. That constraint is also why early decodable text is so hard to make interesting.

We deliberately don't map our levels explicitly to Lexile®, Fountas & Pinnell, or grade levels. Those are other people's frameworks, we haven't licensed them, and claiming alignment we can't back would be exactly the kind of borrowed authority we're trying to avoid.

Yarnia teaches a story's focus pattern, then gives your child unlimited text to practise it. It's built on the Science of Reading, and that text is written to work all five pillars: phonemic awareness, phonics, fluency, vocabulary, and comprehension.

Systematic phonics works when a pattern is explicitly introduced, modelled, and then practised under guidance. That is the shape of a Yarnia story: a lesson page introduces the story's focus pattern and demonstrates the letter-sound correspondence, and the story that follows is written to practise it — unlimited, on-sequence text at the point a specific child has actually reached. We then marry that with tailored phonic sounds: tap a word and it breaks into the sounds your child has already been taught, the way a parent reading alongside would sound it out rather than simply saying it.

What Yarnia can't do is hear your child read — and that one is deliberate. Some reading apps listen to a child read aloud and send that audio away to be scored. We don't, and we don't want to: your child's voice is theirs. The only place Yarnia uses the microphone at all is when your child speaks a story idea, and that audio becomes text on the device itself — never recorded, never stored, never sent anywhere. Yarnia sees only which words your child asks for help with. That signal is real and we use it, but it isn't the same as a person listening — which is why an adult in the loop still matters, and why we build to support teachers and parents rather than replace them.

Early reading starts with a teacher, parent or family member. The gap we set out to fill is giving children engaging text, focused to them, built on the Science of Reading, and made in a way that makes them both the creator and the consumer of their own work.

Partly, yes — but our intention is to enable teachers and parents to do what they do best: teach. We're here to support that work, not replace it.

Yarnia introduces the pattern, demonstrates the sound, provides examples, and gives your child a story built to practise it. That's phonics instruction, and we're not going to be coy about it. What we don't do is listen to them read — you or their teacher still catch the word they stumbled on.

Underneath, we build a level-based curriculum much like a teacher would and write every story against it. We then check the result rather than trusting it: every story is scored against the patterns your child has actually been taught, and rewritten if it drifts off-level, before you ever see it.

We also make the reading fun — your child picks the story topic, colours it, shares the book with family, reads other children's books, and prints their own to color off device and to take with them. Yarnie asks questions along the way. If they get stuck on a word, that's expected and welcomed — tapping it breaks the word into sounds they've already been taught, much like a parent or teacher would, so they sound it out inside their own story instead of being handed the answer.

Taps are the glue. We use them to signal where your child is strong or shaky within their level, and to aim the following story's focus there. The more they read, the better we can aim it.

What nobody has measured is Yarnia itself — whether this app moves reading scores, fluency, or comprehension. Until someone has, we'll tell you what we built and why, not what it will do for your child. You get enough of that from social media and the news. The why — backed by our interpretation of the science, and explained to you — is what matters to us.

Each tap is recorded with the word and where that word's pattern sits in the taught sequence — as an anonymous reading-practice event, with no audio, no voice, and nothing that identifies your child. The reading is deliberately careful: a single tap is never read as difficulty, because a first tap is usually curiosity. Difficulty is only inferred from repeated returns to the same word, weighed against whether the story was finished.

Because stories deliberately mix the current focus pattern with everything taught before it, taps are read against the whole sequence — so a child quietly losing hold of an earlier pattern while practising a newer one becomes visible. That record is designed to steer practice, not to judge: it pulls the next story's focus back to a pattern that needs support, and rotates practice forward once a pattern has had enough recent, low-help exposure in finished stories. It never changes your child's level — that stays your choice — and it is never shown to your child as a mistake or a score. Two honest limits: we don't claim to measure reading accuracy from taps, and a lack of taps is never treated as proof of mastery.

No — a tap is a learning act. Hearing a word pronounced while looking at its spelling is exactly how words get bonded into memory: the tap goes through the letters, not around them, which is the opposite of guessing from pictures.

Saying the whole word is also the best-evidenced kind of help: in the largest study of help types for young readers — nearly 190,000 trials in a reading tutor that listened to children read (Heiner, Beck & Mostow, 2004) — hearing the whole word helped as much as or more than anything else, while guessing from the surrounding sentence performed worst of all. The app never treats a tap as a mistake.

A phoneme is a unit of sound. A grapheme is the letter or group of letters that spells it. The relationship runs both ways, which is what makes English hard:

• /k/ is one phoneme with several spellings — c (cat), k (kit), ck (duck), ch (school). Four graphemes, one sound.
• ea is one grapheme with several sounds — /ee/ (beach), /e/ (bread), /ay/ (great). One spelling, three sounds.

The thing most people get wrong is that graphemes aren't letters. sh is one grapheme. igh is one grapheme. eigh in eight is one grapheme. So "ship" has four letters but three sounds: /sh/ /i/ /p/.

English has roughly 44 phonemes and something like 250 graphemes that spell them. That's why learning to read English is harder than Spanish or Finnish, where the mapping is much closer to one-to-one — and it's why the order items are introduced in matters so much.

Because the order isn't a preference — it follows how English actually builds.

A child can't read shed before they can read sh. They can't read digs before dig, because -s is an ending attached to a word they already have. Short vowels come before letter teams because a letter team is two letters agreeing to make one new sound, and that only makes sense once single letters are solid.

You can check this yourself against the phonemes and graphemes above. Each level adds the next thing that becomes readable, and nothing appears before the thing it depends on. That's a constraint of the language, not a choice we made — which is why it lines up with what your child's school is teaching, whichever programme they use.

What we won't do is borrow someone else's name to make that sound more official than it is.

Heart words are common words like said or was that are taught as wholes because part of them can't yet be sounded out with the patterns your child has learned. Often only part of the word is irregular — in the, the th is a taught pattern and only the vowel has to be learned by heart.

Yarnia treats a word as either sound-out-able or a heart word at each level, never both — and as new patterns are taught, words graduate off the heart list: like is a heart word early on, but once silent-e is taught it becomes a word your child can decode.

Decodable text is written so that the large majority of words use only the letter-sound patterns a child has actually been taught, with the rest coming from their heart words. Systematic phonics instruction is one of the best-evidenced findings in education research (the National Reading Panel in 2000, England's Rose Review in 2006, and Castles, Rastle & Nation's 2018 review all reach the same conclusion), and decodable text is the practice material built to match it — your child succeeds by sounding words out, not by guessing.

We'll be honest about the boundary: the evidence for phonics instruction is much stronger than the evidence for any particular practice text, so we don't claim decodable text is "proven" — we claim our text is decodability-controlled, and every story at a decoding level is checked word by word against the taught patterns before it's shown. A story that falls short is rewritten.

Every story at a decoding level runs through a deterministic scanner before it is shown: each word of the child-read text is checked against the patterns taught at that level plus the child's heart words, and the story has to land in its readability band and genuinely practise its declared focus pattern in at least three different words — character names and heart words never count. A story that misses gets exactly one targeted rewrite: the scanner names the exact failing words and the exact shortfall, only the story pages may change, and the better attempt is served with its true measured score stored — never an inflated one.

Those stored measurements are how generation improves. Our first baseline run showed the earliest level under strain — about one story in twenty fell below its band, mostly because of invented character names — and the fixes that followed (supplying example words for each pattern, keeping names out of the scoring) took the next run of 72 stories to zero below-band at every decoding level. The same record also tracks which patterns each finished story practised, building a practice history per child that is designed to guide what the next story teaches — always within the level you chose, never moving your child up or down.

Spark is pre-reading: letter sounds, first and last sounds, and rhyme, before blending. Rhyme introduces short-vowel words like cat and dog and word families. Fable adds letter teams like sh and ch and blended sounds like st and mp. Tale brings long vowels: silent-e words and vowel teams like ai and ee. Saga and Legend grow sentence length, word length, and vocabulary toward fluent reading. Each level includes everything below it, so the readable vocabulary only ever grows.

This is by design. Words become instantly recognizable through repeated successful sounding-out — reading researchers call this orthographic mapping (Linnea Ehri's work): each time your child decodes a word while looking at its spelling, the spelling and pronunciation bond a little more strongly in memory, until the word is recognized on sight. At the early levels a story can only draw on the words your child's patterns unlock, and repetition with one word changing is how that bonding happens. The simplicity is the design, not a limitation of the writing.

The sound-out feature only breaks up words your child has the tools to blend. Heart words and words that need a pattern not yet taught play as whole words instead — breaking go into g + o would quietly teach a pattern out of order, so the app refuses and says the whole word.

Playing the whole word is not a lesser answer. The largest study of help types for young readers — nearly 190,000 trials in a reading tutor that listened to children read (Heiner, Beck & Mostow, 2004) — found that hearing the whole word helped as much as or more than any other kind of help, and that guessing from the surrounding sentence helped least of all. So on the words Yarnia won't break up, your child is still getting the best-evidenced help there is.

Spark is for children who aren't blending words yet. It works on what comes first: letter sounds, hearing the first and last sound in a word, and rhyme — the sound-awareness skills that reading research treats as the foundation under phonics itself. Single words and big sounds rather than sentences; it's a read-together level.

The story is written from your child's idea, but the words it uses are constrained to their level's patterns and heart words. Once a child knows just six letter sounds they can read sat, pin, tip — and nothing else; every new pattern taught widens what a story can say. A dragon story at Rhyme level will still be about the dragon — told in words your child can actually read. Names and a few story words may sit outside the pattern set; those play as whole words when tapped.

Short vowels come before letter teams, letter teams before silent-e and vowel teams, because each one builds on the last — a child can't read shed before knowing sh. The order within a group (like which short vowel comes first) follows established teaching sequences; the order between groups is a constraint of how English is built. Yarnia's levels follow that build order, which is why they line up with what schools teach even though we don't borrow any program's name.

We don't trust the AI's word for it. Every story at a decoding level is scanned word by word against the patterns your child has been taught, plus their heart words. A story below the readability bar, or one that doesn't genuinely practise its focus pattern in at least three different words, gets one targeted rewrite — and the measured result is stored, never an inflated one. Character names and heart words never count as pattern practice.

Yes. Tap-to-speak in story reading mode reads any word aloud, so your child can keep moving through a story while that skill is still building.

There's a deeper reason too. A five-year-old's listening comprehension typically runs years ahead of what they can decode. Text that matches their decoding starves their language comprehension; text that matches their language comprehension can't be decoded. Reading together, with you taking the harder text, is how both strands grow at once.

Two reasons, one practical and one pedagogical.

The practical one: they're friendly for young readers, and every illustration becomes something your child can colour in. The illustration/colouring layer is a layer to increase engagement reading the content and viewing a story your child played a part in making.

The pedagogical one matters more. Levelled readers often pair predictable text with heavy picture support, which lets a child identify an unknown word by looking at the picture instead of the letters. That's the habit the research pushes back hardest against — it works until the pictures stop. Our intent is that at early reading levels illustrations stay decorative rather than informative: they set mood and provide reward, without giving away the words your child is meant to be decoding.

Stories pass safety checks at two points — we filter your child's idea before writing the story (blocking unsafe words and phrases), and we scan the finished story before it's shown. The AI is also instructed on every request to write only age-appropriate content.

Unsafe words and phrases are blocked before any story is created, and the idea is steered back to something kid-friendly. We are always working to improve our filters to block inappropriate story topics from being sent on to be written. If an unsafe phrase does get through, we ignore it and generate a child safe story that is again reviewed for safety.

We discard any personalization value that looks like a manipulation attempt rather than passing it along, and we re-check the AI's output before display instead of trusting it. If a story doesn't pass, it isn't shown.

Right now we are using Claude (Anthropic) but are testing and evaluating other LLMs including openly available ones that we might host ourselves.