1Password mini Not in Center of Display

That appears to have put the centering behavior back. The odd thing is that the 'center' of the display where 1Password shows up is actually significantly above the position where it needs to be dragged back to to restore the centering behavior. I would classify that as a bug or at least suboptimal UI. The "snap back to centering" region should at the very least span the region in between the actual center of the display and the slightly elevated position where it appears when centered.

Okay, I did a few more tests, and think I understand the behavior better. The snapping algorithm seems a bit finicky to me. For one, if the window snaps while dragging, then you keep on dragging, releasing the snap, then the cursor is almost always no longer under the same part of the window when dragging started.

I guess I expect it to behave more or less like graphics programs (Keynote, OmniGraffle, etc.) that have manual or automatic guides. This is how I would suggest implementing it:

Define a rectangle of cursor positions within which snapping occurs. (This would depend on where the cursor and window were at the beginning of the drag event.) If the cursor is within this rectangle, the window snaps to the center, otherwise it follows the cursor exactly. This may create a discoverability issue, where users won't realize that they can drag the window because it doesn't immediately move when they start dragging. To address this, I would add a rubber band-like effect, where the window moves immediately, but only moves half the vertical and horizontal distance (as defined by the dragging delta) from the center while the cursor is within the snapping rectangle. After leaving the rectangle, the window "snaps out" and follows the cursor exactly. Implementing it in this way would also have helped me realize when I initially dragged the window away from the centered position that there is a certain "gravity" around the center that you can drag back into if that's where you want the window to be. In this sort of implementation, I would expect "snapping in" to behave symmetrically, where it doesn't immediately go to the centered region, but is again based on half the delta of the cursor from the centered position. If the mouse was released while in the rectangle, then the window would snap to the exact center.

Does that make sense? It doesn't seem like this should require so much code.

Great. I did notice something else that appears to be strange. When Safari has a normal web page open Command-Option-\ opens mini centered. If it has a blank browser window, it opens under the menu item. In Chrome, it seems to always open under the menu item. In other applications, it seems to open centered. I'm not sure what the idea is behind these different behaviors.