chessrank.org
Blog

The Hunt for Brilliant Moves: Why Spotting a Brilliancy Is One of the Hardest Things in Chess Software

The Brilliant move — the Genius move, as we call it at chessrank.org — is the famous Brilliant that chess.com made every player crave, and the single most exciting thing a game review can show you. It is the moment the review tells you: you didn't just play well, you played something beautiful — something brilliant!

It is also, quietly, one of the hardest things in all of chess software to detect. In this post we lift the lid on how chessrank.org hunts for Brilliant moves, we are honest about the fact that we under-call them on purpose, and we show why this innocent little badge is really a research-grade problem in disguise.

A needle in a haystack

The first thing that makes this hard is simple: brilliancies are rare. In a normal game almost every move is “fine”. A genuine, beautiful sacrifice that the board is begging you not to play shows up a tiny fraction of the time.

<1 in 500
moves is a true brilliancy
5
recognised types of Brilliant
engine searches per candidate

When the thing you are looking for is that rare, the maths turns brutal: even a detector that is right most of the time will, across thousands of ordinary moves, produce a flood of false Brilliant badges. Getting it right means being extraordinarily picky.

What actually makes a move brilliant?

A Brilliant move is, almost always, a sacrifice the opponent should not accept: you hand over material — a knight, a bishop, a rook, even your queen — and the board that looks like a blunder is actually winning. Take the piece, and you fall apart. There is also one special case: an underpromotion, where turning a pawn into a knight or rook (instead of the “obvious” queen) is the killer.

BrilliantQueen sacrificeBrilliantRook sacrificeBrilliantBishop sacrificeBrilliantKnight sacrificeBrilliantUnderpromotion

Sounds simple. It is not. The catch is the word should: the sacrifice only counts if the engine confirms that accepting it is a real mistake. That single word is where the whole problem lives.

Why this is genuinely hard

To be sure a move is Brilliant, our engine has to walk it through a gauntlet of tests — and a real brilliancy is what survives all of them at once:

Every move in the game100%
Offers up real material~4%
...and it's (near-)best~2%
...and they should refuse it~1%
...and taking it backfiresBrilliant

Confirming that last step — that the offered piece is poisoned — means asking the engine a second question for every candidate: “what happens if they take it?” And distinguishing a sound sacrifice from a piece that merely looks loose (but is defended by a tactic three moves deep) is exactly the kind of thing that is easy for a strong human to feel and painfully hard to pin down in code.

An honest aside
Formally separating a “true” sacrifice from an ordinary loose piece, proving the sacrifice is sound, and doing it fast enough to run in your browser — that is genuinely the kind of problem you could build an entire master's thesis around. We are not exaggerating: perceived move “brilliance” is an active little corner of academic research.

Our philosophy: quiet confidence

Here is where we will be completely straight with you. Compared with the badge that popularised the concept, chessrank.org is stricter. We deliberately under-detect. Where a more generous detector casts a wide net — catching more brilliancies, but also splashing Brilliant onto moves that make experienced players raise an eyebrow — our engine holds a much higher bar.

A generous detector — casts a wide netmore Brilliant, some questionable
chessrank.org — quiet confidencefewer Brilliant, each one earned

Why choose to catch fewer? Because a Brilliant is only special if it means something. If everything is brilliant, nothing is. We would much rather stay silent than be wrong. So when chessrank.org does hand you a Brilliant, it has passed every test above — the move genuinely offered material, it was genuinely the right call, and the engine genuinely confirmed the trap. In other words: our model is more sure of itself. It says less, and it is right about what it says.

The honest scorecard

So yes — today we still miss a good number of the flashier brilliancies that a wider net would scoop up, especially quiet, deep, “deflection” sacrifices where the hanging piece is nowhere near the move you just played. Chasing those without drowning you in false positives is the hard, ongoing work. We would rather earn your trust one badge at a time than inflate the count.

Every improvement we make is measured the same brutal way: did we catch more real brilliancies without the badge starting to lie? That balance — recall versus trust — is the whole game, and it is why the next few percent are the hardest.

Go collect your own

The best part: brilliancies aren't a spectator sport. Every game you review and every bot you beat is a chance to find one of your own.

Game AnalysisReview your games — every Brilliant you played is waitingPlay BotsBeat the ladder and hunt for brilliancies liveOpponent PrepIts Game Review hunts for Brilliant too — scout an opponentBrilliant LeaderboardSee — and collect — every Brilliant you find

Play something beautiful. When it is truly brilliant, we'll be the first to tell you — and we'll mean it.