What Is a Brilliant Move in Chess? The Gates a Move Has to Pass

A brilliant move in chess is a sacrifice that the engine also rates as the best move, that leaves you no worse than equal afterwards, and that was not already winning without it. Those are our four gates, with the numbers behind them, plus what separates Great from Best, what a Missed Win is, and where the error labels begin.

These are our gates, in full, as they run today: deterministic rules applied to engine numbers, so the same game analyzed twice gets the same labels.

What has to be true before a move can be called brilliant?

Four things at once, checked in this order, after the move has already cleared the error ladder described further down.

1. It gives material. A static exchange evaluation on the position before the move must show you offering at least two pawns' worth on the square you moved to (an exchange, a minor piece, a rook, a queen), and the evaluation may drop by no more than half a pawn. Since August 2026 there is a second way in: the move leaves at least two pawns' worth newly capturable somewhere else on the board, and a real engine search shows that taking it costs the opponent at least a pawn's worth. Both paths are described in the next section.
2. It is the engine's best move. Either the top line of the search, or within half a point of winning chances of it. A sacrifice that is second best is not brilliant, however pretty.
3. You are not worse after it. Your evaluation after the move, from your side of the board, must be at least equal. A sound attacking sacrifice that leaves you slightly worse is graded Best, not Brilliant.
4. You were not already winning without it. We look at the engine's second-best line. If that alternative was already worth seven pawns or more for you, or was itself a forced mate, the sacrifice is technique rather than the move that wins the game, and it is graded Best.

A move that passes all four is Brilliant. Fail any one and it drops to the next label it qualifies for, usually Best.

Why isn't every sacrifice brilliant?

Because "gave material" is the cheapest of the four gates to pass, and it is the one players notice. Three of the gates exist to filter it.

The material bar is two pawns' worth on purpose: every gambit hangs a pawn, and grading each one Brilliant would empty the label of meaning.

The best-move gate is the one that catches most romantic sacrifices. If the engine's first choice was a quiet move that keeps everything, and your sacrifice is the second line more than half a point of winning chances behind it, you played a good move and gave up material you did not need to. That is Good or Excellent, not Brilliant.

The "already winning" gate is keyed to the alternative, not to the position before the move. Before August 2026 we suppressed Brilliant whenever the position before the sacrifice was already better than two pawns for the mover. That hid exactly the moves the label is for: a sacrifice at a big evaluation where every other move drops back to equality is the only way to convert, and the label should say so. Now the question is whether the second-best line was also crushing. If your sacrifice wins and the next-best move is roughly level, the sacrifice is the game and it is Brilliant. If the next-best move was already worth a queen, any decent move wins, and the sacrifice is Best.

The collateral path added in August 2026 closes a gap in how sacrifices are priced. A static exchange evaluation only looks at the square the piece landed on, so a move that left a bishop hanging across the board while doing something more important was invisible to it. The fix was a search, not a looser rule: the material must be newly en prise because of this move, and every capture that wins it is probed by the engine against the position the opponent would actually reach. If the search cannot show at least a pawn's worth of punishment for taking, the move is a loose piece, not a sacrifice, and it fails closed.

When does a mating sacrifice skip the gates?

When the move delivers checkmate and gave material to do it, it is Brilliant with no further checks: no best-move requirement, no floor on the position after, no "already winning" guard. Mate is the best possible outcome by definition, so a different move that also mates cannot demote it, and a mate delivered by a piece that could have been taken is the canonical brilliancy. And because smothered mates happen in cramped positions with few legal moves, the old rule that a Brilliant move had to be found among many legal options is gone.

One deliberate exclusion: the collateral path does not apply to mating moves. After mate the opponent has no legal replies, so "material left capturable elsewhere" is an illusion on that board. A mating brilliancy must give material on the mating square itself.

What separates Great from Best?

Best is the engine's top move, or an equal alternative within half a point of winning chances. Great is Best under pressure, and it needs three things: the move is the engine's top choice, it is the only good move, and you were in trouble before you played it.

"Only good move" is measured on the gap between the engine's first and second lines from your side: at least twelve points of winning chances, or at least two pawns of evaluation. The second test exists because winning chances saturate in lost positions. Between minus six and minus nine pawns the curve barely moves, so a winning-chances gap alone made Great unreachable in exactly the positions where an only-move save matters.

"In trouble" means your evaluation before the move was worse than half a pawn against you. Finding the only move from a comfortable position is Best; finding it from a bad one is Great.

Why are the label thresholds in winning chances, not centipawns?

Because a pawn is not worth the same everywhere: one pawn lost at equality moves the game a great deal, one pawn lost at plus eight moves nothing. So the error bands are set on the change in winning chances across your move, from your side of the board, on a zero-to-hundred scale converted from centipawns with the Lichess curve and clamped at ten pawns either way. How that number is made is the eval bar explainer's job; the accuracy statistic built on it is covered in what a good accuracy is.

Every eval is stored with the side it belongs to, because a gate such as "not worse after the move" is meaningless unless the number is from the mover's point of view. That sounds obvious and was once the cause of a real regression.

Two details on the search behind every label. It is graded at a minimum depth of 20 with an eight-million-node ceiling, for the reasons in the analysis depth post. And it always runs with two principal variations, because Brilliant and Great both read the second-best line. A single-line search cannot award either label honestly.

Where do Inaccuracy, Mistake and Blunder start?

The error labels are bands on winning chances lost by the move you played. They are the same for every player and every rating:

The narrow Inaccuracy band is intentional: it is a label for a slip that is nearly a mistake, not a catch-all for every move that was not best. The bands were fitted once, in July 2026, against a labelled reference set using a pre-registered objective, and have been fixed since; we do not tune them per player. And because the error ladder runs before the sacrifice gates, a move that loses more than twenty points can never be Brilliant, whatever it gave up.

What is a Missed Win?

A Missed Win is a move that was not the best move, lost between 5.25 and 5.75 points of winning chances, and did one of three things: gave a clear advantage of at least seven tenths of a pawn back to rough equality, slipped while you were still winning by more than two pawns, or let a forced mate go while falling below a clearly winning evaluation. It outranks Mistake and Inaccuracy in the ladder, and never outranks Blunder.

The label exists because "you were winning and now you are equal" is a different lesson from "you made a mistake". The moves themselves are similar in size; the situations are not. The conversion post is about the habit behind them.

FAQ

Can a move be brilliant if you were already winning?

Yes, if the second-best move was not. The guard looks at the engine's alternative line, not at the position before the move. A sacrifice from plus five where every other move drops to equality is the only way to convert and is graded Brilliant. If the alternative was already worth seven pawns or more, the sacrifice is Best.

Why did the engine call my queen sacrifice a blunder?

Because the error ladder runs first. Any move that loses more than twenty points of winning chances is a Blunder before the sacrifice gates are even consulted. The label describes what the move did to your winning chances, not how brave it was.

Does a brilliant move need to be a queen sacrifice?

No. The bar is two pawns' worth of material on the square you moved to, so an exchange sacrifice qualifies and a single pawn does not. Since August 2026 a move can also qualify by leaving two pawns' worth capturable elsewhere, provided the engine shows that taking it costs the opponent at least a pawn.

Does analysis depth change which moves get the brilliant label?

It can, which is why we fix the minimum. The best-move gate and the "already winning" gate both read engine lines, and lines change with depth. A brilliancy awarded at depth 12 and withdrawn at depth 20 was never verified, so our minimum is depth 20 with two lines searched; the depth post has the measurements.

Why is Great so rare?

Because it needs three conditions that seldom coincide: you were already worse than half a pawn, the engine's best move was at least twelve points of winning chances or two pawns better than the next line, and you found it. A player who collects several across a set of games is defending well in bad positions, which is worth knowing about yourself.

Labels are only useful when you know the rule behind them. Analyze your games free and every Brilliant, Great, Missed Win and Blunder in your games comes with the gate it passed or failed.