1.0E39
expected:<[1D]> but was:<[999U]>

Common caveats about floating point errors apply as usual: 1e39 / 1e36 == 999.9999999999999.

Doubles are not suitable for this task, or at least, numbers at the rounding boundary should not be tested.

There are no fixed tests for values bigger than or equal 10^43, which requires using the scientific notation rather than the engineer notation.

Reference solution is wrong: it expects values floor to the nearest integer instead of 3 significant figures if no prefixes exist, e.g for 29.553385285106668 it expects 29 instead of 29.5.

madlad