In this report:
https://docs.jboss.org/optaplanner/blog/benchmark/2015-09-30/curriculumcourse/#problemBenchmark_comp01

It's says for tabu search comp01:

Best
0hard/-5soft
Worst
0hard/-57soft

But if we look at the graph single benchmark score distribution for the soft score, then tabu search isn't drawn to go down to -57 soft.