I think I see where the communication was breaking down.
An ideal reference plane is infinite. You never come anywhere near the edges, so you don’t need to worry about boundary conditions.
In the real world, infinity doesn’t exist, and you make compromises. The Sierra team is used to talking about which compromises are acceptable.
A reference “plane” is just a polygon fill/copper pour, though it might extend to the entire size of the board on some layers, and it will probably have a few holes in it, for vias if nothing else. As long as your return current stays “far” from the edges/holes, this will be good enough.
On the other hand, you also want each component (or at least each type of noisy or noise-sensitive component) to have its own ground plane, and if they’re going to talk to each other, your signals will have to cross that gap between planes somehow.
Figure 2.6 shows a track just leaping over the gap. The DC return current will eventually find a way, but the high speed AC stuff (including what happens when a signal switches on or off) will hit a wall and angrily radiate in all directions, so that you have crosstalk/interference and fail emissions testing. Figure 2.7 moves the “plane” boundary a little so that isn’t a problem.
But if you must cross the gap, because you’re talking to some component on the other side?
Usually, the least bad option is to just combine the ground planes beneath the two components, so the trace never leaves its plane.
Usually, the next least bad option is to join the two “planes” at a single “point”, and to route the signals over that join point. (You can look up star grounding, but be aware that much of what you’ll find was written assuming audio or 1980s tech, so the unwritten assumptions will be different from the unwritten assumptions for anything labeled High Speed or digital, let alone RF or microwave.)
If you can’t do star grounding either, then you can link the planes with something less than a direct electrical connection; capacitors are a common choice.
Going from the top layer to the bottom layer, and therefore switching to an entirely different layer for the reference plane, is fairly common. If that new reference plane is a power plane instead of a ground plane, then you don’t want to actually connect them and create a short-circuit. Linking them with a capacitor is sometimes a good enough compromise, but it is still a compromise, so the experts are hesitant to recommend it.
The same applies if your ground “planes” are both power planes, but at different voltage levels, or even if they’re both ground planes that you’re just isolating from each other, like your analog ground vs digital ground example. A capacitor may be the least bad compromise, but it is worth first taking another look at avoiding the plane split entirely.