The science

Every text in First Steps is checked three ways, and we only trust it when all three agree.

1. 1The model scores it.Our learned model of English development reads the text's structure — its tenses, its clause depth, its vocabulary — and places it at Pre-A1.
2. 2The benchmarks confirm it. We check the same text against the public standards the field already uses — the CEFR descriptors, the Cambridge Young Learners structure, Oxford's reading levels. It lands below Starters, where we said it would.
3. 3The intuition agrees. Read it aloud to a five-year-old and it plainly fits. When the maths, the benchmark, and the ear all say the same thing, we ship it.

The structural features that separate a child's first English (Pre-A1) from the first formal exam band (A1) are concrete and countable:

First Steps opens with a phonics pre-band built on a small, deliberate set of fifty-four words. Forty of them are anchored to frequency norms — the words a young child actually meets first, in the order the evidence says they meet them. The other fourteen are curated social words: please, sorry, mine, again — the language of being a small person among other people, which the frequency lists under-weight but childhood does not.

Beyond the phonics set, each episode carries a handful of target words drawn from the Cambridge Pre-A1 vocabulary. None of them is a checkbox to be ticked when “done.” Every word is a tracked position: how reliably the child recognises it, says it, and reaches for it unprompted — measured each time it recurs, and never assumed because it was once seen.

We seed the geometry from CEFR. It is the international standard for describing how well someone uses a language — the Common European Framework of Reference — and it gives us a trustworthy first shape for the space a learner moves through.

But we are CEFR-seeded, not CEFR-bound. A band is not a box a child is in or out of; it is a region of a continuous space, and a child sits at a point within it. As learner trajectories accumulate — thousands of children moving through the same material — the data deforms the geometry, pulling the bands into the shape English acquisition actually has rather than the shape a committee drew. The standard is the prior. The children are the ground truth.

A child joins with an age — a prior, not a verdict. It tells us where to start, not where they will stay. An optional few minutes of spoken placement sharpens that first guess; after that, every spoken response updates it.

From then on the next thing is chosen, not sequenced. There is no fixed playlist that every child walks in the same order. The model looks at where this child is and picks the episode, the speaking pane, the target word that does the most for them right now. A child graduates a band when their position has genuinely moved into the next region — not when they reach the end of a list.

Where a child sits is described across five dimensions: Accuracy, Pronunciation, Fluency, Vocabulary, and Content. Any one spoken response scores only the dimensions that apply to it — reading a line aloud measures pronunciation and fluency, and leaves vocabulary choice null rather than guessing it.

This is the difference between Outloud and an app that hands back a single percentage. A child is not 72% good at English. They might be confident in fluency, reaching in vocabulary, and still finding their feet in accuracy — and the right next session is the one that meets that exact shape. The map is invisible to the child. Its output — your child is making strong progress in pronunciation — is not.