Let n be the length of x, i.e. the number of observations.
h = 1L:
$$T_n = \frac{1}{\hat{\sigma}} \max_{1 \leq k \leq n} \left| \frac{1}{n^{3/2}} \sum_{i = 1}^k \sum_{j = k+1}^n (1_{\{x_i < x_j\}} - 0.5) \right|$$
h = 2L:
$$T_n = \frac{1}{\hat{\sigma}} \max_{1 \leq k \leq n} \left| \frac{1}{n^{3/2}} \sum_{i = 1}^k \sum_{j = k+1}^n (x_i - x_j) \right|$$
\(\hat{\sigma}\) is estimated by the square root of lrv. The denominator corresponds to that in the ordinary CUSUM change point test.
By default, kernel-based estimation is used.
If h = 1L, the default for distr is TRUE. If no block length is supplied, first the time series \(x\) is corrected for the estimated change point and Spearman's autocorrelation to lag 1 (\(\rho\)) is computed. Then the default bandwidth follows as
$$b_n = \max\left\{\left\lceil n^{0.25} \left( \frac{2\rho}{1 - \rho^2}\right)^{0.8} \right\rceil, 1\right\}.$$
Otherwise, the default for distr is FALSE and the default bandwidth is
$$b_n = \max\left\{\left\lceil n^{0.4} \left( \frac{2\rho}{1 - \rho^2}\right)^{1/3} \right\rceil, 1\right\}.$$