Two readers paste the same article. One started Spanish last week; the other reads B2 novels on the tram. Blend both texts at 30% and — until now — the word selection logic treated them identically: useful, mid-frequency vocabulary, evenly spread. Good defaults, but level-blind. The beginner met words years ahead of them; the advanced reader met casa and comer, words they'd known since week two.
One question instead of a placement test#
We considered the standard options. A CEFR picker (A1–C2) assumes you know what "B1" means — most people don't. A placement test is friction nobody asked for. So we ask the question the way a person would:
| Your answer | Roughly means | What the blend targets |
|---|---|---|
| Just starting | A1 | The core few hundred words |
| I know the basics | A2 | Everyday vocabulary beyond survival words |
| I can hold conversations | B1 | Mid-frequency words, opinions, narration |
| I'm comfortable reading | B2+ | The long tail — precise, abstract, idiomatic |
You'll see it once during onboarding, and again only if you switch to learning a different language — because your level is per language: comfortable in Spanish says nothing about your Japanese.
If you already know what A2 actually means or where B2 begins, the mapping will feel familiar — but the app never makes you translate yourself into the framework.
The i+1 idea: why "slightly too hard" is exactly right#
The theory here is one of the oldest and most-cited in second-language acquisition: Stephen Krashen's comprehensible input hypothesis — learners acquire language from input pitched at i+1, one notch beyond their current competence i. Too far below and there's nothing to acquire; too far above and there's nothing to anchor to. Blending is arguably the purest delivery mechanism for i+1 ever built: the surrounding sentence in your language is the anchor, and the blended word is the +1.
Level-aware selection turns that from philosophy into arithmetic. Every word in our vocabulary graph carries a real-usage frequency rank; your level maps to a depth in that ranking — your frontier. Selection now peaks just past your frontier and falls away in both directions: far-too-easy words score low (for a B2 reader, casa stops being interesting), far-too-hard words score low (an A1 learner won't retain academic abstractions), and the band just beyond you scores highest.
This composes cleanly with mastery-aware blending: mastery handles the words the app has measured you know; the frontier handles the thousands it hasn't — the vocabulary an intermediate learner walks in with but never saved.
Your answer is a starting point, not a label#
Self-assessment is imperfect — people undersell and oversell themselves. So the level question only steers while the app has nothing better. As your saved words and spaced-repetition history accumulate, LingoBlend measures how deep into the frequency ranking your demonstrated vocabulary actually reaches, and the measured frontier takes over from the declared one — in both directions, and continuously, because six months from now you won't be the level you are today and nobody remembers to update a setting.
And if the blend ever feels mistuned, the direct control is right there: your level sits in Personal Details, editable per language, any time.