I'm going to partially disagree. For purposes of discussion, I'll assume we're looking through a real airplane window (I don't think we are).
I agree with your thinking that we'd expect the camera to offset like that if angled forward. And that this would have been an easy mistake for a faker to make -- to forget to do that. What I'm a bit skeptical on is that, if the camera is angled to look forward, wouldn't that also result in a somewhat oval reflection? I think so, and I don't see that in the vid. I need to go rewatch it, maybe. Perhaps it's tilted forward slightly enough that it does not get oval enough to be noticable. Might have to go mess with a camera and a reflective surface a bit...
Another way to test it would be take stills from the video and try fitting a perfect circle to the lens's reflection with Photoshop. From a quick test, I think the reflection in the frame above is
slightly ellipsoid, shorter on the X axis compared to the Y.
But here's the thing — If you import a 737 model into some 3D software and position the camera just inside the fuselage, then with enough tweaking you can achieve a good match with the scene in the video (the lump just under the wingtip is the saucer, btw):
... which puts the camera here, the window just behind the wing, with a 70mm focal length:
Notice that the camera is angled
backward, toward the wingtip. It's rotated 13.6 degrees counterclockwise from perpendicular.
If the camera was pointing straight out the window, i.e. zero degrees from perpendicular, then the lens's reflection would be perfectly circular, and the reflection would be in the center of the frame. But then the wing's position in the frame would shift to here:
Clearly that's no longer a very good match. Can we fix this by moving the camera back a few rows? Well let's try ...
Now the angle of the wing is all wrong. And if the camera was rotated
past perpendicular, i.e. pointing forward as the lens's reflection suggests to me is happening, this problem would be even worse.
In fact there's very little wiggle room in the camera placement. Because it's a fairly long focal length, any changes in camera position or rotation affect the scene quite a lot. And there's no way to line everything up without having the camera pointing backward along the sweep of the wing.
So I'm asserting the following things here:
* The camera that filmed the view of the airplane wing was angled backward (based on reconstruction of the scene)
* The camera whose lens we see reflected in the video was angled forward (as its reflection is offset to the left of the frame)
* Therefore the camera that produced the final video is
not the camera filming the view from the plane
If this is correct, it would be good evidence for the idea that a pre-recorded, CGI enhanced scene was filmed off a screen through some scratchy plastic — as
@Mick West demonstrated
several months ago.
I've attached my Blender project if anyone wants to repeat the experiment. The camera's Y location and Z rotation are the things you need to fiddle with, Numpad 0 (or View->Viewpoint->Camera) snaps the viewport to what the camera sees. The screengrab from the video can be faded in/out under the camera's Data Properties tab, the alpha slider under Background Images.
This is the 737 model used:
https://sketchfab.com/3d-models/boeing737-259de41be0c7410e86b7e79be33d6b3a