The Evaluator

How LobbyChess Grades a Move

This article explains how the LobbyChess move evaluator works, the philosophy behind it, and a gallery of famous positions judged by the same standard now running on live games.

LobbyChess Move Evaluator

Why Ratings Don't Define Brilliance

The LobbyChess evaluator has expanded beyond the analyze game feature. It runs as you play, in the live commentary beside the board (featured in Stockfish mode only). Post game you can request a detailed evaluation. It's now incorporated into the Sunday Column, where every replayable board has an evaluation bar and a badge annotation on each move.

User rating does not change the evaluation.

A brilliant move is a truth about the position whether you are 800 or 2800.
Style and Accuracy

Dual Approach

Every move gets scored on two independent criteria. This allows a stylish move and an accurate move to each be measured on their own terms.

The grade

How much accuracy the move costs in the evaluation you gave up compared to the best alternative. This drives accuracy percentage.

The badge

The story layer. It lives in a separate field.

MissYou were handed a real chance because of an opponent's mistake. It could be a win, the only move that holds a draw, a forced mate, or something similar, and let it slip.
!GreatThe uniquely strong, non-obvious move. One right answer, and it was not obvious.
!!BrilliantThe move creates a forced mate, or sacrifices real material soundly.
LegendaryThe spectacle tier. A high piece parked where enemy pawns can legally take it.

Precedence when more than one could apply: Legendary, then Brilliant, then Great. A Miss is separate, and only ever lands on a move that was not best. Every badge also carries its reason in plain words, so a thrown away draw reads as "Blunder. Missed Draw." to offer further insight at a glance.

The Grading Ladder
Best  Excellent  Good ✓Inaccuracy ?!Mistake ?Blunder ??

Here are the principles.

The core idea of a Great Badge

Criticality and Difficulty

A Great means two things: there was exactly one right answer, and it was not obvious. The first half is easy to measure, there would exist a gap between the engine's best move and the second best. The second part we start from "there was only one move," then subtract the moves anyone would find anyway: recaptures, free grabs, saving a piece that was already attacked, the only legal reply, and some other factors.

Here is the position that makes the point. What would you call it?

Your call before you scroll:
Best
Bd7 · Black to move
The engine agrees this is the top move. Evaluation +0.63, 3.43 ahead of the runner-up. It cost 0.02 in accuracy.

Black is in check from Qc6, which also attacks the undefended a8 rook. The c8 bishop looks attacked too, but it is defended by the rook and the queen, so the rook is the only real worry. Bd7 handles it: the bishop blocks the check, attacks the queen, and clears c8 so the d8 queen also guards a8, and the rook is safe. Saving a loose rook is the first or second thing a player looks for, so Bd7 is obvious, Best, not a Great.

The bishop move is the only move that holds, so it is critical. But you find it naturally: you block the check with the first piece you look at, and you hit the queen while you do it and save the rook! Critical, yes. Difficult, no. So it gets graded Best. Now compare to another position where the only move that works is tricky:

! Great find.
Rcc6 · Black to move
The engine agrees this is the top move. Evaluation +1.34, 3.80 ahead of the runner-up. It cost 0.00 in accuracy.

Black is in check from Qb5. Two rooks can block on c6, but only the c8 rook works: the a6 rook is tied to defending the a4 knight, so Rac6 would drop it. Seeing that just one of two natural blocks holds is a real find, so it earns a Great.

! Great find.
Kf8 · Black to move
The engine agrees this is the top move. Evaluation -0.07, 1.75 ahead of the runner-up. It cost 0.00 in accuracy.

Nothing is in check and nothing looks like it is hanging, yet only one move keeps Black level. Every natural bishop retreat costs a pawn; the quiet king step to f8 is the single move that holds. A do-nothing king move is easy to overlook, so it is a Great.

What makes the LobbyChess algorithm different

We only grade chess truth

Some grading systems tune the badge to your rating, so the same move earns a medal for a beginner and nothing for a master, because the badge is really guessing "impressive for a player of your level." That is a reasonable thing to measure. It is just a different philosophy.

Our badge depends only on the board and the move. The brilliancy you found at 900 is still a brilliancy when you are 2500. Take a look at this example:

! Great find.

It was the only move that held the position, so a system that rewards the only good move calls it a Great.

There is a consistency payoff, too. A game with a run of strong play, a rating-scaled grader might hand out five brilliant medals. We hand out four. It must meet criteria for addition: is it a forced mate, a real sacrifice, a genuine, non-obvious move.

See For Yourself

3 Tiers of Amazing Moves

These are the badges worth celebrating, and they come in three rising levels. A Great is the uniquely strong, non-obvious move. A Brilliant creates a forced mate or soundly sacrifices real material. A Legendary, the rarest of the three, is a valuable piece parked defiantly where enemy pawns can take it. Press Play on any board to see the move that earns each one.

! Great find.
Kf8 · Black to move
The engine agrees this is the top move. Evaluation -0.07, 1.75 ahead of the runner-up. It cost 0.00 in accuracy.

Nothing is in check and nothing looks like it is hanging, yet only one move keeps Black level. Every natural bishop retreat costs a pawn; the quiet king step to f8 is the single move that holds. A do-nothing king move is easy to overlook, so it is a Great.

!! Brilliant.
Qb8+ · White to move
The engine agrees this is the top move. Evaluation forces mate in 2, the line forces mate. It cost 0.00 in accuracy.

Morphy's Opera Game, Paris 1858. 16.Qb8+ throws the queen in front of the rook; after ...Nxb8, 17.Rd8 is mate. A queen sacrifice that forces mate is a Brilliant.

Legendary.
Qg3 · Black to move
The engine agrees this is the top move. Evaluation +5.16, 0.39 ahead of the runner-up. It cost 0.00 in accuracy.

Marshall vs Levitsky, 1912, the famous 'shower of gold coins.' The queen steps to g3 where both the f2 and h2 pawns attack it, refuses to be taken (capturing loses), and is still the best move on the board. A great piece planted defiantly in pawn fire is the Legendary tier.

The Special Miss Badge

Miss is the one badge that is not praise. It is the only badge a less-than-best move can earn, and it marks a real chance let slip: a win, the only drawing move, or a forced mate. It always names what was missed, shown as a two-layer verdict like "Blunder. Missed Draw." Press Play to see one.

Blunder Missed Draw
Kg8 · White to move
The engine's own top pick was Kg7, not the move played. Evaluation +0.00, 5.25 ahead of the runner-up. It cost 5.00 in accuracy.

The mirror of the Reti Great, one square off. 1.Kg8 walks away from both pawns instead of toward them, throwing away the only drawing move (Kg7), so the game is lost. It reads in two layers: Blunder for the result, Missed Draw for the exact chance let slip. Same board as the Reti Kg7 Great, opposite verdict.

The whole game

A game is a story, not forty snapshots

Everything above grades a single position. But a mating attack is one idea, not five, and we do not award five badges stapled to one combination. After a game finishes, a second pass reads the whole thing and keeps a single anchor badge per forced mate, the move that commits to the attack, and quietly demotes the routine follow-ups. Step through these and watch the live badges collapse to one on review.

Morphy's Opera Game: one Brilliant, not three badges

Starting position
#MoveLive badgeReviewed
1Qb8+Brilliant !!Brilliant !!
2Nxb8--
3Rd8#Great !-

The queen sacrifice 16.Qb8+ creates the mate and keeps the Brilliant. A single-position view also flags the forced finish 17.Rd8# as a Great (it is the only move), so review would show two badges on one three-move attack. Stage 2 keeps the anchor and demotes the routine mate to its plain grade.

Smothered mate: the queen sacrifice anchors the episode

Starting position
#MoveLive badgeReviewed
1Qg8+Brilliant !!Brilliant !!
2Rxg8--
3Nf7#Great !-

The textbook 1.Qg8+ hands the queen to the rook so the knight can deliver Nf7#. The sacrifice is the Brilliant anchor; the forced knight mate, a provisional Great on its own, is demoted.

A routine mate in 3: one anchor on the initiating check

Starting position
#MoveLive badgeReviewed
1Bxd6--
2Qe3+Great !Great !
3Bc5--
4Qxc5+Great !-
5Kb7--
6Qb6#Great !-

No sacrifice here, just a forced mate where every White check reads as a unique 'only move' Great on its own. Stage 2 keeps ONE, and the anchor is the FIRST forcing move (Qe3+), the move that commits to the mating attack, not the check closest to delivery. The mate-distance tiebreak that once pulled the badge to the end of the run was the bug the mate corpus caught.

The guardrails

What keeps a badge worth something

A dead-drawn fortress in LobbyChess only collects a Great under special circumstances: when one move holds the draw and every other move loses.

Deeper Game Review

Coming Soon

Some of the deepest hidden brilliancies still slip past in the moment. The most famous example on this page, the queen brashly stepping into a wall of pawns, is so deep in calculation that it sits ninth on the engine's list until you search very deep, so at live speed it does not yet read as near best. It also shows deep human calculation can still be stronger than a 3200+ ELO machine.

Soon we'll be allowing the grader to show more of its work: the concrete refutation behind a Miss, the full exchange behind a sacrifice, and an Explain button that walks you through why. Until then, enjoy the new Lobbychess Review Game.

LobbyChess Evaluator is Live!