I don’t have experience with ab-initio but here’s my understanding:
When no native structure is available you just use the lowest scoring structure as a surrogate for the native structure when building your plot.
The “funnel” shape is really only intended to demonstrate that favorable scoring models tend to adopt a similar pose. Due to this fact an arbitrary low scoring model to demonstrate this (lowest is typically chosen by convention).
I went and grabbed a couple of resources to confirm my understanding:
The ab-initio tutorial here confirmed the thing about using the lowest scoring model as a surrogate for the native structure: http://www.meilerlab.org/index.php/rosetta-tutorials
Paragraph 2 of this discusses the importance of the funnel shape in a score vs rmsd plot: https://new.rosettacommons.org/docs/latest/getting_started/Analyzing-Results
.
