Yeah I missed this originally, and stuff could be faster with this assumption without a full double parser. The fastest java solution dies some near branchless decoding for these