Soundness

This repertoire statistic measures the objective strength of your repertoire: how much engine evaluation you give away, along the paths opponents are likely to actually take against you.

For each likely path through your repertoire, we measure the avoidable evaluation concession — how much worse your position gets, according to the engine, compared to the best you could have kept. This considers the whole prepared path rather than grading each move in isolation, and gives extra weight to risky critical paths, so a serious concession in one important line can't hide behind lots of quiet positions.

Unlike Effectiveness and Learnability, soundness is graded on an absolute scale that means the same thing at every rating: a path concession around 140 centipawns (roughly "playing a very dubious gambit everywhere") maps to a score of 1, and keeping the concession under about 10 centipawns maps to 99. The tooltip in the app shows the actual path concession the score is derived from.

A high score means the repertoire consistently keeps a healthy objective position. A lower score isn't automatically bad — some gambits deliberately concede evaluation in exchange for practical chances, and that can be a perfectly reasonable trade. That's exactly why soundness is separate from effectiveness: effectiveness is based on how positions score in real games, soundness on what the engine thinks of them. Both are useful, and they often disagree in interesting ways.