Details on rounding are not specified w.r.t. breaking ties, and possibly differ between languages.

Updating tests for Crystal was tricky because default rounding mode for round changed from half-up to half-even, and fixed tests expected values different than returned by reference solution.

The final test should have fixed tests

JS :

• assertSimilar makes the output harder to parse vs the alternatives
• there is no reason not to precode a complicated return object just like you preload dict's

JS :

• rounding doesn't solve float errors, you can still be off by +/- 0.001 in the final result and have to submit a few times
• don't round -OR- intelligently allow for rounded comparisons

JS :

• no example calculation

When you are using concepts which are not trivially known, like unit conversions, m/micro prefixes, it is really necessary to give a calculation unless you want the kata to be about figuring all of that out vs coding (most people won't hence the likely few solutions).