The extra zero at the end comes from how np.bincount and np.searchsorted are used together.

bins = np.linspace(0, 1, n_bins + 1) creates n_bins + 1 edges

np.searchsorted(bins[1:-1], y_prob) returns bin indices from 0 to n_bins - 1.

When you call np.bincount(..., minlength=len(bins)), you force the result to have length n_bins + 1, even though the valid bin indices stop at n_bins - 1.

That’s why you get an array like [3, 2, 4, 0] — the last element is just padding (no samples fall into a non-existent “fourth” bin).

The implementation likely uses len(bins) for consistency across bin_true, bin_sums, and bin_total, especially if cumulative or edge-based plotting is done later. You can safely ignore the last zero; only the first n_bins elements represent meaningful bins.