Every sacrifice looks brave. Only some of them work. The difference comes down to what you get back, how forced it is, and whether you had a choice at all.
Giving up material on purpose is the most admired thing in chess and the most commonly misjudged. Club players remember the sacrifices that worked and forget the ones that quietly lost a piece for nothing. Engines are unsentimental about the difference, and their rules are worth borrowing.
The word covers two very different moves. In a real sacrifice you hand over material and receive something concrete: a mating attack, a passed pawn that costs a piece to stop, a king dragged into the open, a pin that wins the material straight back. In the other kind you give up material for a feeling, usually described afterwards as pressure or initiative, and the position quietly falls apart three moves later.
The practical test is whether you can name the return, and name it as a line rather than an adjective. "I get a rook and the h-file with check" is a return. "It looked dangerous for him" is not.
If your sacrifice does not fit one of these, be suspicious of it.
This is the distinction that catches people out, and it catches engines out too. If your bishop is attacked and cannot be saved, taking a pawn with it before it dies is not a sacrifice. You are not giving up a bishop; the bishop was gone either way. You are collecting a pawn on the way out, which is good technique and no bravery at all.
Chess literature calls this a desperado. It looks exactly like a brilliant move on the board, because a piece walks into a capture, and a naive test will score it as a sacrifice. The honest way to judge it is to ask what your alternatives were worth. If every other move also loses the piece, giving it up cannot be the price you paid for anything.
The same logic applies to trapped pieces. A queen with no squares that takes a rook before it falls has lost you less material than any other move, which makes it a good decision and not a sacrifice.
Most review systems reserve their brilliant label for a move that is both a genuine material sacrifice and still the best move in the position. That combination is rare, and the conditions strip out nearly everything players expect to qualify:
What is left is a small set: you gave up real material, the engine agrees it was the best move, and you are still clearly better afterwards. In ordinary club games that happens rarely, which is the point of the label.
Long before engines, writers reserved the word brilliancy for something stricter still: a move that is winning, a real sacrifice, and quiet. Quiet means no check and no capture. A checking sacrifice at least announces itself, and a capture is a natural thing to look at. A quiet offer of a piece is the one most players never even consider, which is why those moves have kept their reputation for a century.
They are genuinely rare. Scanning thousands of club rapid games turns up only a handful, and most players will go a long time between them. If you ever play one, it is worth keeping.
You cannot run an engine mid-game, so use the questions the engine is effectively asking:
Statmate's game analysis labels a move as a sacrifice only when the material is genuinely given up, ignoring recaptures, even trades, and pieces that were lost whatever you played. Above the ordinary brilliant label it adds two rarer tiers for the quiet, engine-hidden sacrifices that read as brilliancies to a human, and a brilliants log collects the ones from your own history in one place.
Every flagged move opens on the board with the engine's lines beside it, so you can see the compensation rather than taking the badge at face value.
A sacrifice is only a sacrifice when you had a choice and you got something concrete back. Name the return as a line, check what happens when it is declined, and remember that grabbing material with a doomed piece is good play rather than a brilliancy.