Here is a post-finalization note for SRFI 144 and associated change to the sample implementation.  The note reads:

+It is recommended that implementors return <code>+nan.0</code>

+(if the implementation supports that number)

+from <code>flonum</code> when the argument is a non-real number,

+rather than signaling an error.

+This behavior is permitted but not required by the current specification.

The definition of `flonum` in the implementation has been changed. 

However, the other functions have not been: thus `(flsin z)`

where z is not a real number still throws an error.

--