What Makes a Move "Brilliant"?
Why platforms disagree about move quality, and what the labels really mean
Play the same game through three different analysis tools and you will get three different sets of labels. A move flagged as brilliant on one platform is merely good on another and unremarkable on a third.
This is confusing, and it leads players to chase badges rather than understand positions. The labels are not measurements — they are editorial decisions about which engine findings are worth highlighting.
What the Engine Actually Gives You
An engine does not produce move labels. It produces evaluations: numbers describing how good a position is after each candidate move. Everything above that — brilliant, great, inaccuracy, blunder — is a layer built on top by the platform.
That layer works by comparing the evaluation before and after your move, then applying thresholds. Lose enough evaluation and it is a mistake; lose a lot and it is a blunder. The thresholds are choices, and different platforms choose differently.
How Different Platforms Label Moves
Chess.com popularised the brilliant move concept, and their system classifies moves on a scale from best down through good, inaccuracy, mistake and blunder, with brilliant reserved for a narrow category — typically a strong sacrifice that is not obvious.
Lichess takes a more conservative approach. Their analysis classifies moves as best, excellent, good, inaccuracy, mistake or blunder, without a brilliant tier at all. The reasoning is defensible: the distinction between a good move and a brilliant one is a judgement call, not something the engine tells you.
Neither approach is wrong. They are answering different questions — one is trying to make analysis engaging, the other is trying to avoid claiming more precision than exists.
Why the Same Move Gets Different Labels
Three things cause most disagreements. The first is search depth: a move that looks like a blunder at depth 15 may be revealed as strong at depth 30, so platforms analysing at different depths reach different conclusions.
The second is threshold placement. If one platform calls a 50-centipawn loss an inaccuracy and another calls it a mistake, identical play produces different-looking reports.
The third is the engine and network version. Stockfish evaluations shift between releases, sometimes noticeably in specific position types. Two platforms running different versions genuinely disagree about the position, not just the label.
How ChessOnyx Approaches It
Our move quality system uses the same underlying idea — comparing the evaluation before and after each move against thresholds — with a separate category for moves that are both strong and genuinely hard to find, rather than simply best.
The distinction we care about is between a move that is correct and a move that required seeing something. Recapturing a piece is often the best move available and requires nothing. A quiet move that wins because of a tactic four moves later is a different thing, and it is worth flagging differently.
We are also deliberately conservative about the top label. A badge that appears constantly stops carrying information, and inflated feedback is pleasant rather than useful.
What to Do With Move Labels
Use them as a map, not a verdict. Labels are good at telling you where to look — a cluster of mistakes around move 20 means something happened there worth understanding.
They are much worse at telling you why. That part requires you to sit with the position and work out what you missed. The label points at the moment; the understanding has to come from you.
And treat the top-tier labels as decoration rather than achievement. A game with no brilliant moves and no blunders is better chess than a game with one of each.