Puzzle Training serves positions matched to your current puzzle rating and adjusts that rating after every attempt, so the difficulty follows your level without you having to choose it. Puzzles come from the Lichess open database — over five million positions taken from real games, sorted into rating bands from 200 to 3199.
Ratings use Glicko-2, the same system Lichess applies to its puzzles. Every puzzle carries a rating, and the amount you gain or lose depends on the gap between that rating and yours: solving a puzzle rated well above you is worth more than solving an easy one, and missing an easy one costs more than missing a hard one.
Glicko-2 also tracks a rating deviation — a measure of how confident the system is in your number. When you start, deviation is high and your rating moves in large steps. As you solve more, deviation falls and the rating settles. This is why the first twenty or so puzzles can swing several hundred points and later ones move by ten or twenty.
Your puzzle rating is not the same thing as a game rating, and comparing them is misleading. A puzzle tells you a tactic exists in the position; a game does not. Most players hold a puzzle rating meaningfully higher than their playing rating, and that gap is normal rather than a sign of anything.
The most common mistake in rated training is playing the first move to see what happens. Puzzles with multi-move solutions will accept a plausible first move and then present a reply you had not considered, and correcting from there trains guessing rather than calculation.
A more useful method: before touching a piece, identify what is loose or exposed in the position — an undefended piece, a king with no escape squares, two pieces on a line. Then look only at forcing moves, meaning checks, captures and direct threats, because the solution is nearly always one of them. Work the line to its end, including the opponent's best defence, and only then play it.
When you get one wrong, find out what you missed before continuing. In most cases it is not a pattern you did not know — it is a defensive resource you did not look for.
Rated training builds calculation depth. It pairs well with two other modes that train different things: Puzzle Storm for recognition speed under time pressure, and Woodpecker training, which uses deliberate repetition of a fixed set to move patterns into instant recall.
If you want to train the specific errors you make rather than tactics in general, mistake training builds puzzles from your own games. To see which motifs cost you the most rating across recent games, run a Profile Report, or review a single game move by move with Game Report.