# How to take 3-point or more correlators into edge probability computation in surface code decoding?

In Google's work of repetition code (Exponential suppression of bit or phase errors with cyclic error correction), they use the method of Correlation Matrix to characterize the error events and decode. But they only describe the calculation of the correlators up to 2-point correlation.

How can we take 3-point and more calculations into account? And how can we decide the maximum size of correlators we need to calculate while it seems difficult to compute all the correlators when the code becomes larger.