Built on how procedural skills are actually formed, and on the one mechanism known to make memory hold. You set the daily time. The system does everything else.
Language learning isn’t a cost of money; it’s a cost of hours. We encourage you to try the 'vibe-based' AI tutors or the gamified apps first. If you believe a general-purpose AI can teach you what our specialized morphological, semantic, and phonemic engine can, please test that theory. But if you find yourself a year from now still unable to navigate the deeper architecture of the language, we will be here. We don’t sell games; we offer the shortest distance to proficiency. We don't say this to be provocative; we say it because we have seen too many students lose years to bad methodologies only to end up back at square one. We simply want you to reach your goals—regardless of the tool you choose.
Never Be Speechless Again→What you were never told
Not your teacher, not your textbook, not the app with the streak counter. And not out of dishonesty — most teachers try. They drill the sound a few times, watch the room fail to hear it, and settle for close enough, because there has never been a method that reliably gets an adult past that point. Close enough is where almost every learner stops. Not because the ability cannot be built. Because nobody was building it.
In Spanish
/ˈpe.ɾo/ · /ˈpe.ro/
These are not the same sound said with more effort. The first is a single tap of the tongue; the second is a trill, produced a different way entirely. To a Spanish speaker they are as unrelated as p and n.
'But' and 'dog' are different words. If you have been treating the difference as decorative emphasis, you have been choosing between them at random for years.
In Russian
/brat/ · /bratʲ/
Russian pairs almost every consonant with a palatalised twin. The final sound here is not a t followed by anything — it is a separate consonant, and Russian has roughly thirty of these pairs.
'Brother' and 'to take'. Half the consonant inventory of the language is a distinction most courses mention once, in a footnote, and never train.
In Japanese
/ha.ˈɕi/ · /ˈha.ɕi/
Both are hashi. The same two sounds in the same order. The only thing separating them is the pitch pattern across the syllables.
'Bridge' and 'chopsticks'. Japanese pitch accent is not optional stress or dramatic intonation; it functions like a missing vowel. Pronouncing a high-low drop versus a low-high rise alters the underlying architecture of the word. If you flatten your pitch, you force listeners to reconstruct your meaning entirely from context.
Each of these is one distinction, in one language. There are several hundred more, and the ones that matter most are the ones you cannot currently hear. That is not permanent. An adult can be trained to hear differences they were deaf to a year earlier — it is not a gift, it is thousands of deliberate comparisons between things that sound almost the same. Musicians do it with intervals. Master Sommeliers do it with a single grape variety. So that is what we do. Contrast after contrast, inside real sentences, until the distinction stops being something you strain for and becomes something you cannot un-hear. This is the part every method skips. It is the part we start from.
How this works
You can memorise every rule for riding a bicycle. You will still fall.
The brain keeps two kinds of memory. One holds facts you retrieve: dates, formulas, verb tables. Retrieval takes time and attention, and it can fail.
The other holds skills that run without retrieval. Walking. Typing. Playing an instrument. Even driving. No lookup happens, because there is nothing to look up. The ability is not stored. It is built.
In a fluent speaker, language lives in the second system. In a classroom, it is taught into the first.
That single mismatch explains most of what goes wrong. Nobody becomes fluent by knowing more about a language. They become fluent when retrieval stops being necessary.
The mechanism
A path that is not walked disappears. This is measurable, and the rate is known.
But if it is walked again just as it begins to fade, it comes back stronger, and it fades more slowly the next time. Repeat that on an expanding schedule and the interval between necessary repetitions grows from days to months to years.
This is spaced repetition. It is not new, it is not ours, and it is not in dispute. It has been studied for over a century and it is the most reliable finding in the science of memory.
So here is the question worth asking. If the mechanism is known, and the software implementing it is free, why do most people who use it still fail?
The scheduling was never the problem.
The unit was.
The card is wrong · one
Comprehension, production, spelling, pronunciation. Four abilities, four rates of acquisition, four rates of decay. A system that records a single score per word is recording the average of four numbers it never measured, and scheduling all four by it.
We hold them apart. Every word carries four independent states, each with its own review curve.
Above them, two things are tracked at once: the word itself, and every grammatical form of it you have actually met. Knowing a word does not mean knowing it in every form, and the system does not assume otherwise. Because forms are grouped into families rather than filed as unrelated entries, that produces something a flashcard cannot: it locates the failure. When something goes wrong, the answer is not 'you missed this card.' It is either that this particular word is weak, or that this grammatical form is weak wherever it appears — and those two problems need entirely different work.
Below all of it sit the individual sounds, ranked by how often they actually occur in the language, and by which other sounds you are likely to confuse them with.
Every word in the system carries its frequency rank, its full morphological family, its part of speech, its complete phonemic transcription, and its relationship to every word it can be confused with.
A portion of it is published openly. Not a summary written for this page — the same data the lessons are built from.
The card is wrong · two
Repetition of a single context builds a single association. It does not build meaning. Meaning is what remains once a word has been met across enough different situations that the thing they share becomes obvious without ever being stated — which is how every native speaker acquired every word they have. It is also why you cannot define one.
Ask yourself what 'almost' means and you will produce examples, not a definition — because examples are all it was ever made of.
So no exposure repeats. Each time a word comes back it arrives inside a different sentence, assembled at the moment of the session rather than drawn from a fixed deck.
The sentences are real. They come from a corpus of natural language produced by native speakers, for native speakers. Nothing a learner reads here was written by AI or invented to demonstrate a grammar point. What you see is what natives actually say.
And every sentence is selected against what you already know, so that almost all of it is familiar and very little of it is not. Comprehension is never the obstacle. The new element is.
The card is wrong · three
Every spaced repetition tool asks you to rate your own recall, and builds the entire schedule out of that rating. For comprehension this is reasonable. The meaning is put in front of you, and whether you had it is something you genuinely know. No measurement would improve on your own answer, so we ask you, and we use it.
For everything else the question is incoherent. A learner cannot hear a distinction they have not yet acquired — that is what not having acquired it means. Asking someone to score their own pronunciation returns their confidence, not their accuracy, and every interval after that is built on the wrong number.
So the line is drawn where your judgement is actually reliable. You rate comprehension. Pronunciation is measured. Spelling is checked character by character. Usage evaluation is in development.
This makes for uncomfortable sessions. You will occasionally close a lesson holding ten percent on a word you would have sworn you knew. Every other app is built to prevent you from ever seeing that number. What they hide from you is precisely what you needed.
Simulation results
We have not been shipping long enough to publish user outcomes, and we are not going to present our own testers' results as if they were yours. What we can publish is the simulation. It is not an estimate. A simulated learner was played out day by day through the scheduling code the application actually runs — the real interval ladder, the real four-schedules-per-word structure, the real caps and floors — and the words were counted at the end.
| Daily commitment | Words acquired in six months |
|---|---|
| 20min / day | 0–0 |
| 45min / day | 0–0 |
| 60min / day | 0–0 |
| 120min / day | 0–0 |
Read the first row and the last row together, because the relationship between them is the entire proposition. Six times the time invested returns roughly six times the result. There is no plateau in that table, no diminishing return, and no point at which additional effort stops converting. That is the trade, stated plainly. We are not offering a discount on the work. We are offering a method where the work is not wasted.
They are wide because the single largest unknown is how often a given learner finds a given word difficult. Rather than publish a confident midpoint we have published the spread.
Several assumptions in the model make it pessimistic on purpose. Every card is priced at the same duration, when in practice the second, third and fourth skill-check on a word you have just seen is faster than the first. A word you already know is only allowed to be retired at its very first appearance, when in the application it can be retired at any point. We were more comfortable publishing a number that is likely too low than one that is likely too high.
At four independently tracked abilities per word, five thousand words is 20,000 tracked skills. And because grammatical structures are introduced against vocabulary rather than taught as a separate syllabus, the full grammatical inventory of the language — equivalent to that required for a CEFR C2 rating or its equivalent elsewhere — is covered well before the five thousand mark is reached.
This what the system puts in front of you, and at what rate.
Your numbers
It measures nothing else, and everyone who has held a long one knows what it actually rewards. Repeat a lesson you already finished and the number goes up. Pick the easy unit at eleven at night and the number goes up. Learn nothing for a month and the number goes up. That is not a flaw in the design. Nothing about the metric asks you to get better, so the fastest way to satisfy it is not to. Five hundred days later you are no more capable than on day one, and the counter cannot tell the difference, and every application that shows you one knows this.
Because the system tracks acquisition at the level of the individual word, the individual grammatical form and the individual sound, it can show you something considerably more useful: what you have actually acquired, separated from what you have merely been exposed to. Words introduced. Words mastered. Grammatical forms under way and forms complete. Sounds you now reliably produce.
Three diagnostics, a few minutes, in the language you are learning. They will find something. That is what they are built to do.
Then the system does what it would do in an ordinary session: it takes what you missed and assembles a short lesson out of it. One minute will not teach you a language and it is not pretending to. It is there so you can see how perception, production and recall are actually built here, and decide for yourself whether you want more of it.
Why it stays survivable
Every spaced repetition system fails the same two ways. Review load compounds until the daily session is unaffordable, or the learner misses a stretch and returns to a backlog large enough to end the habit outright. Both are caused by the same thing: new material introduced at a rate set by a curriculum rather than by what the learner already owes.
Before anything new is admitted, our system accounts for the work already scheduled across the weeks ahead and determines what capacity remains. New material is never cut off. It is throttled to what your actual load can carry, and it opens back up as that load clears.
Two consequences follow. Working past your daily allowance does not let you race ahead: a simulated learner who ignored the budget entirely and studied as much as possible finished six months roughly 7 percent ahead of one who stopped on time. And disappearing does not punish you. A learner returning after 30 days away completes about 10 normal-length sessions and is back to full speed, without ever being shown the size of the backlog. If they would rather clear all of it in one sitting, nothing stops them. Reviews are never withheld.
Daily time held across 17 simulated learners of widely differing ability, against a 20 minute setting, with no per-user adjustment.
One trait, arrived at by different routes: an unwillingness to settle for approximate.
For
Diplomats, interpreters, aviation and military personnel, clinicians, executives operating across borders. People whose competence is assessed by someone other than themselves, and for whom approximately right is a failure.
You want to read the literature as it was written, with nothing standing in between. You want to follow the film without the subtitles you long ago stopped noticing you were reading. Nobody is grading you — which is precisely why the only standard that applies is the one you set, and you have set it high.
Still hesitating before a word you have met a thousand times. Still going blank when someone finds out you have been studying for years and asks you to say something. Still bracing for the moment the other person gives up and switches to your language. That is not a deficiency of yours. Nothing you were given was built to fix it.
Not for
This will feel cold. That is not a failing on your part. There are applications built for that and several of them are genuinely well made. This is not one of them.
The system will not do it. It reports what it measures.
No method will give you that, including this one. Anyone who tells you otherwise is selling something.
What comes next
Think about driving. In the beginning every action is deliberate. Check the mirror. Find the gear. Remember the indicator. Judge the gap. You are working through a checklist, and the checklist is using all of your attention.
Then at some point — no day you could name — the checklist disappears. You are not running through the steps faster. You have stopped running through them at all.
That transition is the same one separating a person who knows a language from a person who speaks it. And there is nothing language-specific about it. It is what happens to any skill built from many small components, each acquired at its own rate and forgotten at its own rate, once they have been practised in the right order and at the right spacing.
Language is the largest and least forgiving example. It is not the only one.
Two more are being built. Music, covering sight reading and instrumental technique. And chess, where the components are positional patterns rather than words, and where the distance between knowing a principle and playing it under a clock is exactly the distance this system exists to close.
Both arrive inside the same subscription, at no additional cost, held to everything written above. If you are subscribed while they are being built, you are the reason they are being built.
How to use
You set the daily time. The system determines everything else.
Free for thirty days. No card, no payment details, nothing to cancel — the trial ends by itself.
Written while the system was being designed, and now behind it — the architecture described on this page goes beyond what these documents cover, and several of the figures in them have been replaced by the simulation above. We have left them up rather than quietly removed them. They remain accurate on how acquisition works and why. They are incomplete on what we built.