Logic model

Why this page exists

A logic model is the plainest way to say what a product is supposed to do and how anyone would know. Ours is public so that a parent, a teacher or a researcher can hold us to it. It is the first requirement of ESSA Tier IV ("demonstrates a rationale"), and it is where every later study starts.

Inputs

• A knowledge-component map: 279 skills across phonemic awareness, letters, phonics, heart words and word parts for Pre-K to Grade 2, with prerequisites, in the order most structured-literacy programs teach them.
• A decodability checker that reads every word against the sounds a child has been taught, so nothing a child reads is beyond what has been taught.
• Lessons built from a fixed content library and checked against a schema before a child sees them. No AI model runs during a lesson.
• A learner model (Bayesian knowledge tracing with capped guess and slip) that counts first attempts as opportunities and treats a hinted answer as help, not success.
• Adults: a parent or caregiver for Pre-K and kindergarten sessions, a teacher where TaleTykes is used at school.

Activities (what a child does)

1. An explicit lesson on the next skill: I do, we do (with fading support), you do (no hint before the first try), then a short story that uses the skill, then building words from sounds.
2. A mastery check when the model is confident: six items, three taught words, one untaught word and two made-up words, with no hints.
3. Books decodable with the child's taught set, three on the shelf a day: new books written for the child up to the plan's monthly allowance, and checked library books the rest of the time, with two quick words on the book's skill at the end.
4. Spaced review: a skill comes back after 1, 3, 7, 21 and 60 days, and a review a week after the check decides whether it is verified.
5. Four sessions of 15 to 20 minutes a week, about 60 practice opportunities. The plan ends on its own.

Short-term outcomes (we measure these ourselves)

• Learning curves per skill: error rate on first attempts falls with opportunity. A skill whose curve does not fall is flagged and fixed.
• Verified skills per instructional hour (the verified learning rate).
• Retention: probe pass rates at 7, 21 and 60 days.
• Transfer: the gap between taught and untaught words on checks stays small.
• Hint dependency and bottom-out rates fall over a child's first months.

Medium-term outcomes (someone else measures these)

• Words correct per minute on grade-level passages, scored by a person, next to published ranges.
• A standard decoding measure given by someone who is not us.
• Teacher judgment of reading in class.

We have not run that study yet. The pilot protocol on this site says how we will.

Assumptions and what would break them

• The sequence is right. If children stall at the same place, the sequence, not the child, is suspect; the learning-curve flags are built to catch it.
• Co-use happens for the youngest children. Sprout lessons need a grown-up in the room; the plan assumes it and says so.
• Practice measures track real reading. Only the outside measure can confirm it; until then we call them practice measures and nothing more.

What this model does not claim

It does not claim TaleTykes improves reading outcomes, that any child learns sooner than with another program, or any gain in standardized units. Those are claims for a study to make, and our claims policy blocks them in code until one has.