It's stated in the documentation, but I find this behavior to be rather wart-y, and it led to some bugs in statsmodels. Why treat x array different from x sequence of lists?

For an array it iterates over the first axis

import numpy as np
import matplotlib.pyplot as plt

x = np.random.normal(size=(2,1000))
fig, ax = plt.subplots()

ax.boxplot(x.T)

for lists it iterates over the zero axis

fig, ax = plt.subplots()
ax.boxplot([x[0,:],x[1,:]])

Why the distinction? I'd expect them to be the same (i.e., the list gets an np.asarray call) and it seems inconsistent that they're not.