Cone tips do not point at the correct positions

In the following, each of two vectors ends at the start of the next one. Tweaking the length of the cones in any way suited to examining them visually, they clearly do not point at the base of their successor cone as (naively?) expected. At least if we expect that the bases of the cones are centered to the start point of the vector it is representing.

import plotly.graph_objs as go
from numpy import array

points = [
    [array([-0.03620968,  0.24876811,  5.51279187]), array([1.39049627, 0.57429448, 4.67921793])],
    [array([1.39049627, 0.57429448, 4.67921793]), array([2.02424265, -0.15235655,  2.63707954])],
    [array([ 2.02424265, -0.15235655,  2.63707954]), array([ 3.68335955, -1.14431782,  1.09909298])]
]

cone_traces = []

for pair in points:
        start_point = pair[0]
        end_point = pair[1]
        vector = end_point - start_point

        # a partial workaround to scale the vectors more or less to their actual length, which
        # plottly does not allow fully controlling: https://github.com/plotly/plotly.js/issues/3613
        scaled_vector = vector * 1.5

        start_x, start_y, start_z = start_point[0], start_point[1], start_point[2]
        end_x, end_y, end_z = end_point[0], end_point[1], end_point[2]

        trace = go.Cone(
            x=[start_x],
            y=[start_y],
            z=[start_z],
            u=[scaled_vector[0]],
            v=[scaled_vector[1]],
            w=[scaled_vector[2]])
        cone_traces.append(trace)

layout = go.Layout(

    scene=dict(
        camera=dict(

            # set the virtual point of view to point to the origin
            center=dict(x=0, y=0, z=0),

            # set the virtual point of view to be positioned parallel to the depth axis, at a distance
            # sufficently away from the vectors to see them all; although plotly won't fully respect the
            # angle this eye position should make, https://stackoverflow.com/q/76q921229/1509695,
            # it will be very very close to that, and the chart user can nudge the tilt to
            # perfect the viewing angle.
            eye=dict(x=0, y=0, z=-3))))

fig = go.Figure(data=cone_traces, layout=layout)
fig.show()

Reproduce the supplied Python go.Cone example, varying cone length and camera settings to confirm the positional discrepancy. Trace the corresponding plotly.js 3D cone rendering behavior and add a regression check for cone bases aligning with the supplied vector start points; done means the cones point to the expected successor positions.