quoting the principle of operations
For signed binary integers, the leftmost bit represents the sign, which is followed by the numeric field. Positive numbers are represented in true binary notation with the sign bit set to zero. When the value is zero, all bits are zeros, including the sign bit. Negative numbers are represented in two's-complement binary notation with a one in the sign-bit position.
all this irrelevant of the number of bits used to represent the number
so for the case in question, interpreting the FFFFADCC as a negative 21044 seems correct
adding a semibyte simply adds confusion
just an analogy, but for DFSORT BI
implies unsigned FI
check the syncsort documentation for something similar