I have created a bug
MDL-11860 on this problem which is related to the rounding limits.
I have tested it with 34013 but also with 30000 and 40000. In the last two values everything goes well
However you can get a work-around if you set {a} to be between say min 200 and max 400 and you write your question this way
{a}00 cm =___km
and change the formula
{a}/1000
I have tested it.
You should also correlate your tolerance limits to more reflect the values you accept. If you stay in the 10000-90000 cm your answers will be between 0,100 and 0,900 so your tolerance limits should be 0.001 nominal. In this case everything is OK.
I have tested it.
So this is less a bug than a good knowledge of the interactions between the tolerance and the rounding. At least it should be more documented ...
Pierre
P.S. the 13 in 34013 creates conflict between the rounding process (number of decimals) and the grading limits .