I see three different usages:

• Explicitly using std::numericl_limits<type>::quiet_NaN()

• "NaN" appears to link to MR::NaN in core/types.h, and is explicitly of default_type, ie. double-precision

• "NAN" for me claims to link to <math.h>:

  # if __GNUC_PREREQ (3, 3)
  #  define NAN (__builtin_nanf (""))
  

  , which seems to be single-precision

Would be better to be consistent.

• Choose which syntax we want to standardise

  • Always use std::numeric_limits?
  • Permit NaN but preclude NAN?

• Augment check_syntax to check for non-conformant usage

• Modify code base to conform