[vbuterin]

> For those who don't know, never ever use floats for things like money.

This is normally a very good rule to follow. The reason is that floating point is binary, whereas cash is decimal, so even an innocent number like 0.4 has no exact representation in binary (it's actually an infinite tail: 0.011001100110011...).

Here, however, I'm using floats NOT to store decimals; every single value that I store is an integer. There is absolutely no danger in using floating point numbers to store relatively small integers like 1253251126; the issue only arises in the context of very large numbers (specifically, those above 2^53) and decimals. If anyone can come up with a remotely realistic series of integer manipulations that will cause an inaccuracy in Javascript where all values always stay below 2^50.9, I will certainly abandon my choice of moving to integers at once; otherwise, I see no problem.

[shalmanese]

Still seems risky to me. There are some numeric calculations where input and output values are small but intermediate representations might be many times larger. It's not certain that all the people working after you are going to be aware of the assumptions intrinsic in your code.