S1: Evotuning pre-training dataset curation

TODO.

S2:  development and application of ranking-error function