Fixed-point math, for any static or dynamic precision in any base.

Differences from floating-point:

• exact results, except where explicitly rounded
• supports rounding in any base, such as decimal
• simpler interface (no NaN, infinities)
• better performance
• more compact storage for some value ranges
• unnormalized, so sigfigs of result can be determined by sigfigs of inputs

Differences from bignums:

• if you know the "shape" of your inputs/calculations, the statically typed fixed-points can be more efficient
• XeN is optimized for values that fit in a relatively small range

Dynamic precision? Isn't that floating point?

Technically yes, but the Xe API is much simpler than typical floating point; it's designed for applications like working with a collection of values that are known to be fixed-precision decimal of the same type, but the number of decimal places won't be known until runtime.