The horizontal X-axis of a folding funnel is RMSD, a measure of how far apart the atoms in two solutions are. The vertical Y-axis, is "free energy", which is related to the score in Foldit. Free energy is a negative number, the more negative, the better.
In a good folding funnel, the dots form a narrow "neck" pointing to the lower left corner of the plot. This means that different folds are tending to converge on the same solution. Higher up on the graph, solutions with more free energy can be scattered farther from the reference solution, forming the wide cone of the funnel.
In a bad folding funnel, there might be multiple "fingers" extending down toward the X-axis, showing solutions that are as good or better than the reference solution.
In Foldit, folding funnels are used to evaluate results in different ways.
One type of comparison involves proteins which have a known solution or native available. Most Foldit revisiting puzzles have one or more native solutions available in the Protein Data Bank, usually as part of a paper published in a scientific journal.
The folding funnel for a protein with a known native might show the difference between all player solutions and the native. The expected result is a classic funnel shape, with a neck pointing to the lower left of the plot.
Another type of comparison involves player-designed proteins. For these folding funnels, a protein designed by a player could be used as the reference, and the results from either Rosetta@Home simulations or other players used as comparisons.
For player designs, a folding funnel with a classic funnel shape indicates a successful design. Other shapes indicate problems, such the protein having alternate low-energy states.
The Partition Tournament early results shows folding funnels for 20 player designs. The results are from the 2018 Protein Design Partition Tournament, which challenged players to seek new folds for other players' designs. Some players' designs had survived relatively well, but other designs had one or more decoy states, folds with low free energy, but a different shape than the intended design.
The low-energy "decoy" states are also referred to as "holes" in the "energy landscape" of the protein. The more holes, and the deeper the holes, the less likely a protein is to fold up on its own, a key goal for Foldit designs.