How we make the data

This page explains where the numbers on each opening page come from, which games we count and what we do not claim.

Data from Lichess (CC0) and Stockfish 19. Text written with AI from this data and checked automatically against it.

Sources

We use the open Lichess database, released under a CC0 license. From it we take every standard game played on the site in September 2026: that is where the club player numbers and the strong online player numbers come from. For over-the-board chess we use the live broadcasts of in-person tournaments that Lichess publishes, covering 12 months, from October 2025 to September 2026. Stockfish 19 evaluates the positions. Opening names come from an open catalog of names and moves. So one position, for example with a white pawn on e4 and a black pawn on c5, collects its club player games, its strong player games and its evaluation.

Club players

These are the players we write for: 900 to 1700. Whether the game starts with a pawn to e4 or to d4 makes no difference: we filter by who plays and at what time control. A game counts if it is standard chess, starts from the initial position, has a result and gives each player a base time of 3 minutes or more. Both players must also be rated between 900 and 1700 on Lichess. That is a Lichess rating, not a FIDE rating, and the two do not match one to one. If your club rating sits around 1300, your Lichess rating may be different.

Rating bands

We split club players into four bands by the average of the two ratings: 900 to 1100, 1100 to 1300, 1300 to 1500 and 1500 to 1700. Each band includes its lower limit and excludes its upper one, and a game with an average of 1700 goes into the last band. This lets you see whether an idea, such as bringing the knight to f3 before moving the c2 pawn, gets more or less common as the rating goes up.

Strong players

For comparison we count two more groups. Online, Lichess games from the same month where both players are rated 2200 or higher. Over the board, broadcast tournament games where both players are rated 2300 or higher. These groups are a reference: they show how far more experienced players handle the same position, for example with a pawn on d5. They are not a target you have to copy move by move.

Reading the numbers

Club player figures are given as natural frequencies, such as "3 out of 10 games".

Only the short summary on each page uses a percentage, with the number of games, the source and the month, so it can be quoted.

A side's score is the average of its results: a win counts 1, a draw counts half and a loss counts 0. If White scores 6 out of 10 after putting a bishop on c4, White wins more than it loses with that idea among club players.

How we count positions

We read the opening 12 moves for each side of every game and merge them into a tree. Two move orders that reach the same position count as one: whether the bishop reaches c4 before or after the knight reaches f3, the final position is the same and its games are added together. A move enters the tree if at least 20 club player games or at least 5 strong player games reached it. Below those amounts it does not appear in the data, because with so few games the result depends too much on chance.

The engine

Stockfish 19 evaluates each position at depth 20, with a single main line. Evaluations are given in centipawns: 100 centipawns are roughly one pawn. For example, giving away the e5 pawn for nothing usually costs about that much. An evaluation at that depth is an estimate, not a final verdict. In sharp tactical positions it can change if the engine searches longer.

How we mark mistakes

For both sides, we go through the positions club players reach most often. A move goes on the list if it is played often in that position and loses evaluation compared with the best move. Depending on how much it loses, we call it an inaccuracy if it loses less than 100 centipawns, a mistake if it loses 100 or more, and a blunder if the loss is much larger. Leaving the f7 pawn undefended when the opponent can win it is a typical example of a mistake.

For each mistake we show the refutation, what you lose and how club players score with that mistake and with the best move, when both have enough games. If the score barely changes, we say so: between 900 and 1700 the opponent often does not punish the mistake, even though the evaluation drops. For example, if a pawn is left hanging on d5, the evaluation drops even when the opponent does not take it.

A trap is a blunder that the opponent punishes within 6 half-moves and that ends in mate or in a material gain of 3 points or more, such as a minor piece. If a pinned knight on f6 falls within those 6 half-moves, it counts as a trap.

When there is enough data

Not every opening is played often enough. We grade each page by how often its position is played by club players and by strong players. If almost nobody between 900 and 1700 reaches a position, the page says so and explains what you will face instead, for example what club players actually answer when a pawn arrives on d4. Pages with little data stay online, but we do not ask search engines to index them until there are more games.

How we use AI

A language model writes the texts from each page's data: moves, games, scores, evaluations and mistakes. Its instructions tell it not to add numbers, games, plans or variation names that are not in that data. The Spanish text and the English text are each written from the data, not translated from each other. That is why a page about the pawn on e4 may explain the same point in different words in each language.

Then a program checks each text against its page's data. It checks that every number is in that data, that every move mentioned is one the page is allowed to name, that length and style follow the rules, and that the text is not too close to another page. If a text fails these checks, the page is not offered to search engines. The checks cover numbers and moves. They cannot confirm every nuance of an explanation, such as why you should keep an eye on the f7 square.

What we do not claim

We do not say an opening is better or worse for everyone.

Club player results show how players rated 900 to 1700 on Lichess do. They do not tell you what the position is worth.

We do not say a Lichess rating equals a FIDE rating.

We do not say the Stockfish 19 evaluation at depth 20 is the final word.

We do not say why a move scores better. Stronger players may choose it, rather than the move itself giving an advantage.

We do not give numbers for positions with very few games.

Open data

The per-opening figures we use can be downloaded from the open data page.

Each page shows the month of its data, and when the data changes, the numbers are recalculated.