Comment by aidenn0

I admit I got nerd-sniped here, but the table for floats[1] suggests that 10000.0 be represented as a float32. However, isn't it exactly representable as 0x70e2 in float16[2]? There are only 10 significant bits to the mantissa (including the implicit 1), while float16 has 11 so there's even an extra bit to spare.

1: https://cborbook.com/part_1/practical_introduction_to_cbor.h...

2: i.e. 1.220703125×2¹³