SUGGESTION 1 (D. Jones): Library of folds.

Firstly it is essential that a record be made of the proteins/domains that
are found in the servers' fold libraries. Clearly if a fold is not present
in a fold library then this must be noted in the assessment. Of course
participants have the opportunity to ensure that the appropriate folds
are present given that the answers are already known (the fact that the
folds are now known is of course the major fly in the ointment for CAFASP1).

SUGGESTION 2 (D. Jones): Assess by structural similarity.

Secondly, I think you are going to have a hard time assigning some of the
SCOP numbers to the partial similarity targets. It was by no means clear
how SCOP would end up assigning some of the
fold identifiers to some of the targets when they are eventually added to
SCOP. My feeling is that for an assessment of automated methods you should
use an automated assessment! I would suggest making the assessment very
simple by saying that a correct match is anything which produces a Dali
Z-score of over 4.0 (or perhaps 6.0 to be safe). After all, if a structure
comparison method cannot find a significant similarity then how can a
fold recognition method be expected to - particularly an entirely
automated method. I would certainly be _very_ happy if I could reproduce the
results of Dali with a fold recognitio method. The day I can reproduce such 
results I will book a flight to Stockholm and start preparing
my Nobel Prize speech!