Sign in
Can a free game review catch a brilliant move?
A is the rarest call in chess review: a sacrifice that looks like a blunder but wins the game.
Chess.com's Game Review badged one in each of 100 real games. We replayed every game through each free review tool below, ours included, and ranked them on what they caught.
Eliot· Updated July 20, 2026
Chess.comFlagged all 100 by definition. These are its calls.
Chessigma
Wintrchess
Chessiro
Chessitup
ChesskitCoach Froggy
White sacrifices the knight on c7 to lure the black queen away from defending f6. If Black accepts with Qxc7, White wins the f6 knight with Qxf6, leaving White clearly better. Declining the sacrifice leaves White a pawn up with a dominant position.
Catching brilliants is only half the exam. The other half is restraint: a reviewer that stamps brilliant on ordinary moves is handing out confetti, not information. Same 100 games for every tool. Right means it catches more of the 100 real brilliants, up means more of its badges are confirmed by chess.com's review.
Chessigma hands out one unconfirmed badge every 20 games. The other tools: one every 5 to 7.
Source: chess.com Game Review's stored labels over the same 100 games for every tool. The formal definition of over-calling (incorrect calls per 1,000 ordinary moves) is in the methodology below.
A brilliant looks like a mistake until the payoff lands, and that is exactly what a detector has to see through. First, the four checks every reviewer runs on a candidate move. Then the five shapes these sacrifices take, ranked by how often the reviewers miss them.
Every reviewer asks the same four questions, whether it knows it or not. Each question below shows the same pair: one real brilliant that passes all four, and one false brilliant that fails exactly there.
Could they have played it safe?
Coach Froggy
Real brilliantForty-five legal moves, no opening book left to copy, and the queen lands on the one square White's rook can take. Nobody made Black do this.
Coach Froggy
False brilliantRf8 gives up a rook two white pieces can capture, and it looks heroic. It was also Black's only legal move: they were in check. No choice, no credit.
Every brilliant move here is a sacrifice, and each one fits a shape. Some shapes clear the gates above at almost every reviewer. Others read like blunders until the payoff lands, and most tools grade them accordingly. Pick a shape to see why it is hard to find.
Coach Froggy
White ignores the threat to the knight on d4 to create an immediate checkmate threat. If Black takes the knight with Bxd4, White forces mate starting with Qh8. Even if Black declines the sacrifice, they cannot stop the incoming checkmate.
The piece given up isn't the one that moved.
The giveaway happens away from the move you are staring at, so it never registers as a sacrifice.
11% of brilliants
Indirect sacrifice tops the list, flagged by fewer than 2 of the 5 reviewers, on average. The rest miss it not for being a lesser move, but for where it falls at the gates above.
100 real chess.com games played between 2020 and 2026, across rapid, blitz, and bullet. The players who found these brilliants range from roughly 450 to 3200 rated. From those games we pinned the 100 brilliant moves shown above, out of 671 brilliants in the wider pull we collected.
The reference labels are the ones chess.com's Game Review actually stored for these games. We use chess.com because it is the reference most players see: when someone asks “was my move brilliant?”, chess.com's badge is almost always what they mean. It is a reference, not objective truth. Chess.com's own detector has changed over time and will change again, and a move it labels brilliant today might not be labeled brilliant next year.
Each game goes into every tool's own review, the same way a user would run it. Tools name the label differently, so we map each tool's top designation to a single canonical “brilliant”:
| Reviewer | Its own label | Counted as |
|---|---|---|
| Chess.com | “Brilliant” | the answer key |
| Chessigma | “Sigma” | brilliant |
| WintrChess | “Brilliant” | brilliant |
| Chessiro | “Genius” | brilliant |
| Chessitup | “Brilliant” | brilliant |
| Chesskit | “Brilliant” | brilliant |
A hit means the tool flags the exact same ply as brilliant.
The over-calling chart measures the reverse direction: of everything a tool flags brilliant, how much does chess.com's review not confirm? The formal rate is incorrect calls per 1,000 ordinary moves. For each tool, over every game it reviewed, we count every move chess.com's stored labels do not mark brilliant, then count how many of those the tool calls brilliant anyway. The denominator is every non-brilliant move, not only sacrifice-like ones, because it requires no judgment call. Current rates: Chessigma 0.7 per 1,000, the other tools 2.0 to 3.0 per 1,000.
We do not claim any of these moves is “objectively brilliant”. We also do not publish the internals of our own detector. This benchmark measures one thing: agreement with the badge players actually see.
Every board above links to the original game on chess.com, and every card has a Copy PGN button. Paste any of them into any review tool and watch what it says. The dataset behind this page is one JSON file: the 100 games as PGNs, each with the ply of the brilliant move.
