Well, having never used the Floating Point routines before ... you still knew the value went in AARG instead of Aint ... that's worth something.
I guess it's easier when you know how it works.

A floating point value consists of 4 bytes (32-bits).
The highest byte is the exponent.

Looking at the list of variables in FP1832L.bas, the only way to put 4 bytes in a row with the exponent lining up with AEXP is to start at AARGB2 ($1d).

Code:
aint    var     long $1c	' Long access to AARG
AARGB3  var     aint.byte0	' $1c
AARGB2  var     aint.byte1	' $1d
AARGB1  var     aint.byte2	' $1e
AARGB0  var     aint.byte3	' $1f
AEXP    var     byte $20
Microchip saves RAM space by using the same memory for Floats and Integers.
But when you have an integer value, the LONG variable uses AARGB3-AARGB0 (aint).

I believe the routines were intended to be used by starting with integers, converting to floats, and doing the math from there.
They were never intended to accept previously encoded floating-point values, whether IEEE or Microchip format.
Hence that ability is not documented.

Undocumented features are way more fun than the documented ones.