Ok so the right value for literal 0.9999 is really -0.00010000500033335.
Thank you to have raised the problem
Do you have a solution for that ? Like using another literal to get -0.00010000500033336 instead ?

Edit: What im saying in this part of the comment is wrong, check the comment below
Voile you are right, ln(0.9999) in this test is false.
But there is a reason...
According to Math.log:
-0.00010000500033334732
According to WolframAlpha: (All decimals are true)
-0.00010000500033335834
So the precision 14 version should be -0.00010000500033336 and not -0.00010000500033335
The Math.log function is false for mantissas near to one.
I think i will set this test to precision 13 only, i will check that.
Edit: -- End of the wrong part --
The newton's method is a good idea
I thought it would not have enough time to execute a million times, what is your advice as a 2 dan, for the kata to be interesting, should I leave the possibility of the Newton method or should i censor it?

Ok i will check that, random tests are 62 random generated bits from 2nd to 62 bit, sign are always positive and first bit is always one.

Thanks again

I patched it

Ok perfect, thank you

I think it was patched, can you tell me if it's effective?

Thanks for helping me find the vulnerabilities, don't hesitate if you have any advice,
I set module.children = undefined; is it sufficient ?

The english one doesn't have explicitly write the fast-converging formula (taylor series)
If you have a better link to propose give it to me

Do you think we should add precision?

In math null mean zero yes