1) ScoreUtils.extractLevelDoubles() ignores Score.getInitScore(). Start reading from SolverBenchmarkResult#accumulateResults() which does:

        standardDeviationDoubles = StatisticUtils.determineStandardDeviationDoubles(singleBenchmarkResultList, averageScore, getSuccessCount());

2) The string value of standard deviation doesn't always make sense, for example for the nurse rostering sprint benchmark:

Standard Deviation 	sprint01 	sprint02
0.0E0/1.0 	0hard/-59soft  0 	0hard/-61soft  0
0.0E0/0.0E0 	0hard/-64soft  1 	0hard/-64soft  1

What does that trailing 0 mean? 0.0E0/0.0E0

3) Can we add the hard and soft level strings from ScoreDefinition.getLevelLabels() ?

4) Better yet, can we calculate the standard deviation as a Score object somehow? Then we 'd get the levels formatting etc out-of-the-box? It's hard though as we probably need to use doubles to calculate the standard deviation (underflow, overlfow, rounding, ...).