Hi @janvv that sounds right, I think the integrator assumes linear interpolation between sample points.

A unit impulse input is well-
approximated by a unit-area short-duration pulse.

But the trapezoidal approximation of an impulse that starts at T=0 has an area of only 1/2.

@sawyerbfuller is exactly right. If you would like to see this explicitly, you can plot the inputs:

import control as ct
import numpy as np

system = ct.tf([5], [900,60,1,0])
t = np.arange(0, 60*12)

U = np.zeros_like(t)
U[0] = 10
resp1 = ct.forced_response(system, T=t, U=U)

# shift input one sample later
U= np.zeros_like(t)
U[1] = 10
resp2 = ct.forced_response(system, T=t, U=U)

plt.plot(t[0:20], resp1.inputs[0:20], label='forced @ t=0')
plt.plot(t[0:20], resp2.inputs[0:20], label='forced @ t=0')
plt.legend()

Also, note that the first response is not the same as an impulse response for a continuous time system. If you plot the impulse response for this system, it won't match either of these traces.