How Rookahead grades a move
The rules behind every verdict in a Rookahead review, from Best to Blunder, Clutch and Legendary, shown on Morphy’s Opera Game.
On this page
Every move in a Rookahead review gets one of eleven verdicts. They come from a short set of rules we wrote ourselves and keep in a public rulebook, so when a verdict surprises you, you can check why. This post walks through those rules using the most famous short game there is: Paul Morphy against the Duke of Brunswick and Count Isouard, played in a box at the Paris Opera in 1858. You can open the whole game with its review and step through it as you read.
Winning chance
Engines measure a position in pawns: +1 means White is about a pawn better. That number doesn’t say much on its own, because a pawn matters far more in a level game than when you’re already a rook up. So Rookahead turns the evaluation into a winning chance between 0% and 100%, using the same curve as Lichess and the eval bar. A level game is 50%. About +1 is 59%, +3 is 75%, and a forced mate is 100%.
Each move is then measured by how much winning chance it gives up compared with the engine’s best move in that position. We call that the loss, and it decides the verdict:
| Loss | Verdict |
|---|---|
| the engine’s own move, or a checkmate | Best |
| any other move, under 2% | Strong |
| 2–5% | Solid |
| 5–10% | Slip |
| 10–20% | Mistake |
| 20% or more | Blunder |
From a level game, drifting to −0.3 is Solid, about a pawn is a Slip, two pawns a Mistake and a piece a Blunder. The same drop costs less once the game is decided: going from +8 to +5 is only a Slip, because you’re still winning easily.
Only the engine’s own move is Best. Two good moves are rarely worth exactly the same, and an engine’s numbers wobble by a tenth of a pawn or so from one search to the next, so a move that “came out just as good” gets Strong.
Clutch: the only move
Three verdicts can replace the one from the table. The first is Clutch (), for the engine’s top move when it’s the only move in the position.
“The only move” means every other move throws away at least 10% of your winning chance. Here the next best kept White about a pawn better instead of two and a half.
It doesn’t have to be hard to find. 7.Qb3 is a quiet queen move that only works because of two threats at once, and Morphy found two more like it later in the game, 12.O-O-O and 14.Rd1, which are Clutch too. But an obvious move counts just as well when nothing else works.
Everyone would play 10.cxb4, and it’s still Clutch: what counts is that nothing else worked. So is taking a queen that has just promoted with check, when every other move gets you mated.
Two moves are never Clutch, however necessary: taking back what your opponent just took, which is simply the game going on, and the checkmating move itself, which is Best.
Legendary: a sacrifice that works
Legendary () is always a sacrifice. However strong a move is, if it only trades evenly it can’t be Legendary. The sacrifice also has to be sound (no more than about 0.4 of a pawn worse than the engine’s best move) and leave you level or better. And you can’t already be completely winning before it, since throwing material away in a won game isn’t much of a feat.
After 10…cxb5 11.Bxb5+, White has given a knight for two pawns and Black can’t get the pieces out in time.
There’s a second road to Legendary: a sacrifice that leads to the fastest mate, at least two moves quicker than anything else. That’s how the game ends.
Not every sacrifice in the game makes it. 13.Rxd7, a rook for a knight, comes out Best. Giving up the exchange has to be the only good move as well to be Legendary, and at that point White had other ways to win.
Oversight, Theory and Forced
Oversight () is for a mistake that answers a mistake. If your opponent blunders and your reply gives the gift straight back, landing you about where you were before their error, the move is an Oversight instead of a Mistake or Blunder. It’s still a chance thrown away, so it gets a verdict of its own.
Theory () marks the opening moves that are in Rookahead’s list of named openings, as long as every move before them was too. A book move that would be a Mistake or worse gets that verdict instead, and theory ends there. Forced () is for the only legal move.
When we change the rules
These rules were tuned on real games, and they’ll keep changing as people send us games where a verdict looks wrong. Each change goes into the rulebook first. Rookahead keeps the engine’s work for every review you’ve saved, so when a rule changes, your old reviews are graded again in the background, usually without running the engine at all. Changes to the rules are listed in What’s new.
If you think a verdict is wrong, press Disagree? on the move, or bring the game to Discord. That’s how most of these rules got their current shape.