What is your answer?
Same as yours, given that scenario.
But I'm not looking for a point. I'm looking for a line.
And I'm not depending on alignment of points, solely. I'm also using perspective lines.
In your jellybean example. I don't have to be aligned with the jellybeans to find the horizon LINE. I can be off to the side, and if the line they do make is parallel to the plane of the ground, then I know the plane of my eyeline is in the same plane as the jellybeans.
But it's more than that. Because if I'm using multiple lines of perspectives, i.e. other alignments of jellybeans, then I don't have to be inline with any of them. I just see where all of those lines are converging; and not to a point but to a plane.
That's what I consider to be the strength of this method: I'm using both a leveling plane and perspective lines, hoping that each will back the other up. It takes some precision, no doubt. But I'm not trying to measure "dip." I'm just trying to detect whether it exists or not.
I can (and will, time permitting) demonstrate in a controlled environment how changing different parameters of the setup effect the observation. Things like aligning the sight line away from the cube center, both vertically and horizontally. Changing the location of the camera. Changing orientation of the camera (pitch, roll & yaw). Changing orientation of the cube (pitch & roll).
I concede it's very hard, using those 1 1/4" vinyl tubes to ensure a level sighting. Not that the water won't be level (because it will be, a priori); but because the camera height is hard to match. It's difficult even if I move the camera well off center laterally so that the tubes line up. Which is why I like the idea of using a basin of some sort to provide a liquid surface unaffected by glass or plastic optical effects. I may not ditch the tubes, since they're already there. Might as well keep them. But adding an "infinity pool" of a sort would remove the deficiency those present.
I do still hope to resolve the question about how we can know we're seeing the true horizon. I could see this island today, at a distance of nearly 20 miles:
...but I couldn't see San Clemente Island 60-some miles on the bearing I took today's picture ¹.
At 100', the globe earth calculation of visual horizon is about 12 1/4 miles with no refraction. With refraction? Maybe 14 miles.
I don't know how to calculate a distance to the horizon if assuming a flat earth, but how far away should the vanishing line be from a height of 100'? More than 20 miles? Even if I could have seen San Clemente 60 miles away, I wouldn't be seeing it's shore. I'd only be seeing its higher elevation. But there'd be a clear line of horizon. If I'm able to see the Islas Coronado but not San Clemente, does that mean I should consider it a no-go for observation?
You need to tell me because I don't know.
¹ 66 miles, for the record. (Had to check)