Say you solve a puzzle that Lichess rates 1,800 on its puzzle scale. Does that make you a 1,800 Lichess rapid player? It does not, and the reasons are worth spelling out, because they shape how we are building AccelaStudy® AI Chess, which is still in development.
What a puzzle rating measures
Every puzzle in the Lichess puzzle database carries a difficulty number. It comes from how people fare against that puzzle, and it moves as more people try it. The number describes the puzzle. On Lichess, solving puzzles also gives you a puzzle rating of your own, and that is a statement about your puzzle results.
A game rating is a different statement. It comes from the games you played against other players, and in the same pool it goes up or down with each result.
Why the two numbers drift apart
Solving a puzzle and playing a game ask for different things.
- A puzzle tells you a tactic exists. In a game nobody says "there is a combination here," so half the work is noticing that you should be looking.
- A game rewards the absence of mistakes. One loose piece in move twenty can undo an hour of good play. A puzzle never asks whether you saw the threat three moves ago.
- A game uses the whole of chess. Openings, quiet middlegame decisions, endgames, and the clock all count, and a set of tactical puzzles touches only some of that.
None of this makes puzzles a poor way to train. They are one of the best. It only means that the number at the end of a puzzle session is a result about puzzles.
Rating lists are not interchangeable either
A rating is only meaningful inside the list that produced it. Lichess uses a Glicko-2 system, and FIDE uses Elo. Within Lichess alone, rapid, blitz, and classical are separate pools. There is no fixed conversion between any two of them, so "1,600" on its own tells you very little.
That is why every target we show carries a pool: 1,600 Lichess rapid, never a bare 1,600 or a bare "Elo." In the current plan, our rating targets use Lichess standard rapid, and other pools will need their own rules before we offer them.
Three records, kept apart
We are building the product around three separate records, and they never share a label.
- Learning progress. What you have practiced and learned, and what still needs another look.
- AVIAN's readiness estimate. AVIAN is AccelaStudy AI's adaptive learning engine. An estimate of how close you are to a target is planned, and it will not appear until the underlying model is calibrated against real, independently verified ratings.
- A verified playing rating. A rating recorded by the rating provider, for an account that is yours, in the pool you chose.
The product never assigns a rating, and it never turns a puzzle rating into a playing rating. If you tell us a rating yourself, it will be labeled self-reported.
What reaching a target means
A course called Toward 1,600 is a path, and the path does not end when the lessons do. Graduating at a target is planned to require demonstrated play in that pool plus the course's own learning requirements. Any rule about how many recent rated games are enough is a candidate we are still validating, so we do not treat it as final.
We also cannot promise a rating gain, or a date for one. What we can promise is that the target is only marked reached when your play in that pool shows it.
Where this stands
Today the development clients let you choose a target, save it, and practice puzzles that match a module of the plan. Lessons, opening study, game review, rated play, and AVIAN's recommendations are planned and not built.
The puzzles come from the Lichess puzzle database, released under CC0. AccelaStudy AI Chess is not affiliated with or endorsed by Lichess.