Well I lost my own bet... by one frame.
15/16 spotdodges caught with the same frame setup. The one time I missed by one frame (14th repetition), I caught him out of his buffered spot dodge at the end without even counting the frames.
Here's the frame setup I used exactly.
Frame | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29
Dedede | 1st SD Frame | 2nd (Invinc. start) | 3rd | 4th | etc | | | | | | | | | | | | | | Buffer SD | | Invinc. end | | | | | | | 1st SD Frame |
MK | Crouch | | | | | | | | | | | | | | | | | | | | | | Input Dtilt* | | | | | Hit D3's Foot
*This is the first frame of the input; in frame-advance the input would be held while frozen on frame 22 and about to advance to 23.
**This entire table assumes Dolphin's default 2-frame delay, which is why there's 2 extra frames between MK's Dtilt input and it hitting.
I input MK's Dtilt on the same frame of D3's spotdodge every single time and hit him with 94% accuracy.
Also I tested a couple of times, watching for the input delay, and got the same delay for both D3 and MK twice in a row.
Hypothesis:
When an input is made, the "random" amount of lag becomes set for all subsequent inputs made for the duration of the first input's animation. This way, even though the amount of lag at any given moment may seem random, someone reacting to an animation before it ends will experience the same lag that the animation first experienced.
(Ignoring the 2 frame delay within Dolphin)
Example: D3 inputs a spot-dodge and it comes out after 1 frame of lag. The number of frames from the input to the end of the animation is 28 (input frame > no input frame 1 animation starts > 27 frame animation). For that entire 28 frame window, all inputs made experience the same 1 frame of lag. MK starts his Dtilt on frame 24 (input frame > no input frame 1, animation starts > no input frame 2 > no input frame 3 > no input frame 4, dtilt hitbox) and it connects on frame 28; D3 is hit.
This is probably incorrect and would need more testing but the current data upholds this hypothesis fairly well.