Why we generate every lesson one step past your level

Every lesson SayLocal generates sits one step past what you can already do. Not two steps, not a fixed deck everyone gets. That choice is the whole product, and it comes straight out of fifty years of language-acquisition research. Here is the case for it.

The problem with a deck everyone shares

Most apps teach the same sequence to every learner. Card 400 is card 400 whether it is your first week or your fifth year. The trouble is that a fixed sequence is almost never aimed at you. Half of it is review of things you already own, which wastes the one resource a learner cannot get back: attention. The other half jumps to material you have no footing for yet, so you memorize a shape without ever understanding it.

The fix is not more content. It is content aimed at the exact edge of what you can handle today, and re-aimed tomorrow once that edge has moved.

i+1: comprehensible input, one step past you

The clearest statement of that idea is Stephen Krashen's input hypothesis. Krashen argued that people acquire language when they understand input pitched just beyond their current level, a level he wrote as i+1, where i is what you can already handle and the +1 is the next reachable step.^[1] Input that is all i is comprehensible but teaches nothing new. Input far past i+1 is noise. The narrow band in between is where acquisition happens.

The same shape shows up in Lev Vygotsky's zone of proximal development: the gap between what a learner can do alone and what they can do with a small amount of support. Learning lands inside that zone, not below it and not far above it.^[2]

Mostly words you already own, plus one or two you do not. That is the sentence we ask the model to write, every time.

Why one step and not two

If a little challenge is good, why not pile it on? Because the kind of difficulty that helps is specific. Robert and Elizabeth Bjork call it a desirable difficulty: effort that feels hard in the moment but produces more durable learning, as opposed to difficulty that simply overwhelms.^[3] One new item in a sentence you otherwise understand is desirable. Five new items in a sentence you cannot parse is just frustration with extra steps. The +1 is deliberately small for the same reason a spotter adds one plate, not four.

How SayLocal works out your i

To pitch a sentence at i+1, the app has to know your i. So every time you read, listen, speak, or review, SayLocal updates a per-word strength model: which words you reliably produce, which you only recognize, which you keep missing, and which grammar patterns trip you up. That lexicon and your list of weak spots are the raw material the generator works from.

When it builds a lesson, it draws mostly from words you own, drops in one or two new items at the edge of your range, and biases the drill toward a weak spot you have been dodging. The sentence is generated for you, in the register of the place you are going, and it is different from the one your neighbor gets, because your i is different from theirs. A fixed-content app cannot do this. Neither can a single shared deck, however well made.

Keeping what you learn

Meeting a word once at i+1 gets it in the door. Keeping it takes two more things the research is unusually clear about. The first is spacing. In a synthesis of 184 articles and more than 300 experiments, Cepeda and colleagues found that spreading study across time beats cramming the same amount into one sitting, and that there is an optimal gap before each review.^[4] SayLocal schedules reviews on that principle, so a word resurfaces right as you are about to forget it rather than long after.

The second is retrieval. Roediger and Karpicke showed that testing yourself on material produces far better long-term retention than re-reading it the same number of times, even with no feedback.^[5] That is exactly what roleplay is: you have to pull the word out and use it, not just see it again. Recognition is cheap; production is what survives the trip.

What this means for you

You will not see a generic A1 deck. You will see sentences built from words you already know, carrying one new piece each, in the dialect you are about to need, scheduled so they come back before they fade and surfaced in conversations where you have to say them out loud. The longer you use it, the sharper the aim gets, because every answer you give tells the model a little more about where your edge actually is.