Learning Methods
Curriculum & Levels
AI & Language Tools

THE SHORT ANSWER
Level suggestions come from targeted practice, reviewed every three sessions.
Your Dioma level selects the curriculum topics you practice. After every three completed sessions, Dioma reviews the practice that was aimed at specific topics: how much of your level you have attempted, and how your recent results went. When both support a move, it suggests one step up or down. You can accept, defer, or set your level yourself. A review can also end with no change.
Suppose you have a bad session. The prompts ask for the kind of sentence you usually manage, and this time you just had an off day. One session shouldn’t interrupt your progress.
Your level has one job in the way Dioma works: it selects the curriculum topics your practice is drawn from. If the level moves too quickly, your practice changes direction every time you have an off day, and you never stay with a set of topics long enough to get better at them. If it moves too slowly, you can spend a long time on material that is too easy to teach you much, or too hard for the corrections to help. The review has to tell an off day from a level that doesn’t fit.
That requires thresholds: how accurate is accurate enough, and how much practice is enough. But those thresholds only help if the evidence behind them is reliable.
A level review needs more than an accuracy score
First, the review needs to distinguish what you were asked to practice from what else appeared in your answer. When an exercise asks you to practice a particular rule, your answer provides direct evidence about that topic. A full spoken or written answer also uses other topics along the way. Those results can tell us something about your language, but they answer a different question from whether you can use the rule the exercise was designed to test.
Second, the accuracy measure needs to detect improvement. If we build a test practice history with a known improvement in it, Dioma needs to be able to pick up on it.
Third, the review needs to account for the grader. An AI model grades your answers, and different models can grade the same answers differently. Your accuracy score can change even when your answers haven’t, so a change in the score does not necessarily mean a change in your ability.
Only practice aimed at a topic counts toward a level move
Dioma records, with every topic result, whether the exercise was aimed at that topic and which model graded it. The level review reads the targeted results.
Say an exercise asks you to practice the Spanish subjunctive after expressions of doubt. Your result on the subjunctive goes to the level review. If you also use object pronouns in the same answer, that result still shows on your topic map, but the level review leaves it out.
A teacher deciding whether to move a student up a class works in a similar way. She doesn’t go by a conversation she overheard in the hallway, where the student chose what to say. She goes by the work that’s core to your level, because that work asked for the thing she needs to see.
Scenarios count in the same way as regular exercises, because each Scenario is built around a topic. Open Practice is treated differently for now. There you choose a freer response to a prompt, and a grade on that kind of answer doesn’t mean the same thing as a grade on an exercise built around one rule, even when the same model does the grading. Until we have shown that the two can be compared, Open Practice informs your topic evidence and carries no weight in the accuracy the level review reads.
The accuracy side of the review reads only results graded by the current grading model. Practice you did before a change of grader still counts toward how much of your level you have covered, while the older grades stay out of the accuracy reading. The point is that a change of grader on our side shouldn’t show up as a change in your level. The testing behind the corrections themselves is in how Dioma chooses corrections.
A move needs breadth and recent results to agree
With the evidence sorted out, the review asks two questions. Dioma divides each level into smaller steps called sublevels, and both questions are about the one you are on. How much of this sublevel have you been asked to practice? And how did your most recent targeted results go?
Strong results on a handful of topics show that you are good at those topics, and say little about the rest of the sublevel. Wide coverage with mixed results shows that you have seen the material and may need more time with it. So a suggestion to move up needs both: enough of the sublevel attempted, and strong recent results.
Then the review waits to see the pattern twice. Once you are past your first reviews, the evidence has to support a move at one review and again at the following one before Dioma suggests it.
Your first reviews can correct a starting level sooner
You choose your starting level, and it’s hard to choose precisely. If you picked one well above or below where your speaking and writing are, waiting for a pattern to repeat would keep you on the wrong material longer than necessary.
So the first reviews after you start are treated as calibration. Dioma can suggest a move from a smaller amount of targeted evidence, with a minimum it still has to reach, and without waiting for a second review to agree.
Keeping you at the same level can be the right result
Go back to the bad session. Its results join your recent targeted results, and if the others were sound, the review sees a learner at the right level who had a hard day. No suggestion appears. Your plan is still rebuilt from your results, as it is every three sessions, so the topics you practice can change while your level stays where it is.
The same logic runs the other way. Low results across a sublevel you have practiced widely, repeated at the next review, lead to a suggestion to move down.
Dioma holds off on a move down in two situations. It won’t suggest one soon after you have moved up, so you get time with the new material. And once you are past your first reviews, it won’t suggest one when you are returning from a long break, so that time away doesn’t count against you. Every suggestion is one sublevel at a time, within the curriculum for your language.
Every week we check real learner data to make sure it’s working as intended
The investigation behind this rule ran on existing user data, tests and simulations. They showed us how the readings behave on practice histories we built, and they can’t show how learners do after a real suggestion. So every week I review the real readings myself, outside the product. It is a different review from the one that reads your practice every three sessions, and its job is to check the rules.
I ask the same questions each time. Are learners accepting the offers or declining them? How do they do in the reviews after moving up? How often does a move get reversed? How long does a learner wait for a first offer? Do the accuracy readings look alike across languages, and across regular exercises, Scenarios and Open Practice? And is the evidence being recorded correctly to begin with?
That last question comes first whenever something looks wrong. An odd result can mean the rule is asking for the wrong amount of evidence, or it can mean the evidence was recorded wrongly, and those call for different fixes.
When the results do point at the rule, a change gets tested before anyone lives with it. Every review saves the readings it used and what it decided, so a proposed change can be replayed against those saved reviews to see which past decisions would have come out differently. The replay can’t show how those learners would have done on practice they never received, so I keep a change small: one setting, adjusted when there is enough evidence, with the reason written down. The weeks that follow show whether it helped.
The suggestion is an offer, but you decide
When a review does support a move, Dioma offers it and gives the reason. If you accept, your practice moves to the topics of the new sublevel. If you defer, you keep practicing where you are, and the reviews continue every three sessions. You can also change your level yourself in Settings at any time, with or without a suggestion.
We made it an offer because you know things the review can’t see, such as how your last conversation with a tutor went and how the sessions have felt. My advice is to compare the suggestion with your own experience of practice. If the sessions have felt comfortable for a while and the offer is to move up, take it. If the offer is to move down and the sessions have felt like a struggle, the sublevel below is likely to give you corrections you can use, and a later review can suggest moving back up. If the reason doesn’t match what you have noticed, defer, and let the next few sessions add evidence.
Keep in mind what the level describes: the part of Dioma’s curriculum that fits your practice now. A move changes what you practice next, and a formal exam remains the way to certify a CEFR level.
Whichever way you decide, the next three sessions are the next evidence.
See how Dioma builds your practice
Speaking and writing practice guided by your results and an educator-written curriculum.
Explore the method
Frequently asked questions
Can I change my Dioma level myself?
Yes. You can set your level in Settings at any time, whether or not Dioma has suggested a change. The reviews continue every three sessions afterward.
How does a long break affect Dioma's level review?
Once you are past your first reviews, Dioma won't suggest moving down when you are returning from a long break. The targeted practice you did before the break still counts toward how much of your level you have covered.
What does my Dioma level tell me about my CEFR level?
Your Dioma level tells you which part of Dioma's curriculum fits your practice. A level suggestion reads your targeted speaking and writing results inside Dioma, so it can't stand in for a full CEFR assessment or an exam certificate.
Related Articles






