Evidence
What we claim, what we check, and what we don’t know yet.
Every early-reader book is checked by code against the sounds a child has learned. We don’t claim it improves reading yet. Here’s what backs each sentence.
We say
- Every early-reader book is checked by code against the sounds that child has been taught, before they see it
- Books are checked for readability, safety and shape before any child sees them
- No child PII reaches any AI vendor
- Reading speed is shown against published norms, as a range, never a verdict
We don’t say
- That TaleTykes improves reading, or any effect claim, until an outside study says so
- That home practice readings equal a benchmark
- That a font or a setting treats any condition; spacing is a preference we offer first
- Anything about other kids to a child under 13
Research documents
Six documents, kept current
Each one says what question it answers, what we read, and what changed in the product because of it.
- Logic modelIf a child does this, we expect that: inputs, activities, the outcomes we measure ourselves, and the ones only an outside test can show.Updated 2026-11-03
- Pre-registration templateThe fields every TaleTykes study and every in-product experiment fills in before it starts, in the order the Open Science Framework and the Registry of Efficacy and Effectiveness Studies ask for them.Updated 2026-11-02
- Knowledge-component validation report (template)The ten sections our learning-science staff fill in for a skill whose learning curve does not behave, from the data to the decision.Updated 2026-11-02
- Proof-of-teaching pilot protocolHow the first outside study of TaleTykes will be run: who, how many, what is measured by whom, and what counts as success or failure.Updated 2026-11-02
- Read-aloud scoring calibration protocolThe gates automatic speech recognition must pass before it scores a child's reading, including a bias audit across accents, ages and recording conditions.Updated 2026-11-02
- Claims policyWhat TaleTykes may and may not say about learning at each evidence tier, who signs off, and how the policy is enforced in code.Updated 2026-11-02
Claims registry
Every sentence on this site that makes a claim, and its source
The build fails if a page uses a claim pattern that isn’t allowed.
| Claim | Where it appears | Backed by | Status |
|---|---|---|---|
| Every word checked first | Home, For families | Decodability checker · test suite | Checked by code |
| Three books on the shelf every day | Home, For families, How it works, Pricing | Shelf job at each child's midnight: library books first, then re-reads · DB tests | Product fact |
| 15 or 25 new books a month | Home, For families, Pricing | Plan allowance in plans.ts, counted per family; the shelf asks for them · the book maker that writes them is being built | Being built |
| No child PII reaches any AI vendor | Home, Schools, Privacy | assertNoPii() on every prompt · audit | Checked by code |
| Reading speed vs typical range | Parent fluency view | Hasbrouck & Tindal (2017) | Published source |
| Follows the order most structured-literacy programs teach in | For schools | Logic model · skill catalog | Order only, not materials |
| Improves reading outcomes | — | — | Not claimed |
- We do not say that anything here has been shown to work by a study of TaleTykes, because no such study has been completed.
- We do not state multiples, comparisons with other programs, or gains in standardized units we did not measure.
- We do not say that TaleTykes identifies or helps with any named condition. Schools run screening; we are not a screener.
- Every public statement about learning is checked against a claims list in our code; a sentence that is not on the list fails the build.
The full policy is in Claims policy.
Fluency norms
How we show reading speed
Words correct per minute from home readings, shown against the typical range for the child’s grade. A range, never a verdict.
Maya · words correct per minute · Grade 1
Logic model
If a child does this, we expect that
Inputs, activities, the outcomes we measure ourselves, and the ones only an outside test can show.
- 1
Explicit lessons in sequence
I do, we do, you do, then a check with made-up words, then a story that uses the sound. Four short sessions a week.
- 2
Verified mastery, not a guess
A skill is verified only after a 6-word check with no hints and a review a week later. Two bad reviews send it back to practice.
- 3
What an outside test would show
Words read correctly per minute and a standard decoding measure, given by someone who is not us. We have not run that study yet.
The full logic model, with assumptions and what would break them, is published here.
Research basis
What each part rests on
Each practice, how strong the evidence is, and the caveat: the evidence is for the practice, not yet for our version.
| Practice | Evidence for the practice | Source | In TaleTykes |
|---|---|---|---|
| Systematic phonics in a set order | Strong† | National Reading Panel (2000); What Works Clearinghouse foundational-skills practice guide (2016) | The 279-skill sequence follows the common order; a lesson teaches one skill at a time. |
| Decodable text matched to taught sounds | Moderate† | WWC practice guide (2016), recommendation 3 | Every book and passage is checked by code against the child's taught set before they see it. |
| Explicit instruction with modeling and fading support | Strong† | Archer & Hughes (2011); WWC practice guides across subjects | I do, we do with fading, you do with no hint before the first try. |
| Spaced review and retrieval practice | Strong† | Cepeda et al. (2006); Roediger & Karpicke (2006) | Reviews at 1, 3, 7, 21 and 60 days; retention probes at 7, 21 and 60 days. |
| Mastery learning with transfer items | Moderate† | Kulik, Kulik & Bangert-Drowns (1990); Pane et al. (2014) | A six-item check with an untaught word and two made-up words; practice continues until it passes. |
| Error-specific feedback and hint ladders | Moderate† | Koedinger & Aleven (2007); Shute (2008) | Attention cue, strategy, partial model, full model; the full model leaves the item ungraded. |
† Strength of evidence for the practice in the literature, as summarized by the sources named. None of it is evidence about TaleTykes.
Measurement quality
What our numbers are, and are not
Everything a child does here is practice. It shows what’s working this week, not a diagnosis or a grade.
- Verified skillsCheck passed with no hints, two made-up words right, then a review a week later.per child, per skill
- Learning curvesError rate by opportunity on first attempts, per skill, across children.from 200 children
- RetentionProbe pass rates 7, 21 and 60 days after a check.per skill
- Reading speedWords correct per minute from a read-aloud, next to published grade ranges, only if you turn recording on.a check-in, not a test
- Not reportedRisk labels, reading levels as ages, rankings between children, and anything that sounds like a medical finding.never
Learning curves, not testimonials
We fit a standard learning-curve model to each skill. A skill whose curve does not fall is flagged and fixed. Null results stay on the page.
Experiments only among defensible choices
We compare lesson settings we would defend either way, never withholding teaching, only with consent, pre-registered, and we keep the results whatever they show.
Read-aloud scoring has gates
Speech recognition scores a reading only when it agrees with human raters within set limits across accents and ages. Until then it is off.
Subgroup numbers appear only for groups of 10 or more. Nothing here identifies a child.
Safeguards and dosage
What keeps the lessons honest
Rules that are checked by code, not by promise.
No lesson from a model
Lessons are built from a fixed content library and checked against a schema. No AI model runs during a lesson, and children never type to one.
Nothing unchecked reaches a child
Every book passes decodability, readability, safety and shape checks before a child sees it.
Recommended dosage
Four sessions of 15 to 20 minutes a week, about 60 practice opportunities. The plan ends on its own; there is nothing to keep going for.
No streaks, no pressure
Reading days are counted, never lost. Children see no numbers in Sprout and are never compared with each other.
Where we stand
How we will know if it works
ESSA Tier IV: a logic model and research-based practices, with a study planned. Nothing above that until an outside evaluation says so.
Tier IV, today
This page and the logic model are public. That is the whole claim.
Planned pilot
A pre-registered comparison in partner schools with an outside reading measure. The protocol is below; results will be posted whatever they are.
Studies completed
None yet.
Advisory board
Being seated. Names and affiliations will be listed here.
We test one change per book, randomized, and measure the next reading check. Results are published here when a study finishes.
Changelog
Changes to this page
- 2026-11-02 · Evidence page, logic model and the six research documents published. Tier IV stated.
Questions about the evidence?
Write to research@taletykes.com. We answer researchers, teachers and parents the same way.