REXX NUMERIC DIGITS and NUMERIC FUZZ precision and comparison behavior across implementations

REXX arithmetic precision is governed by NUMERIC DIGITS and equality tolerance by NUMERIC FUZZ. The documented feature is implementation sensitive, with default values and comparison semantics varying between IBM z/OS REXX, Regina, and ooRexx.

The goal is to establish a consistent understanding of how significant digit limits and fuzz tolerance affect numeric results and equality tests when scripts move between interpreters. Uncertainty remains about default settings and the exact rules for numeric comparison under different FUZZ values.

What are the documented default values for NUMERIC DIGITS and NUMERIC FUZZ in the target implementations? How does NUMERIC FUZZ define equality for values that differ within tolerance? Which implementation differences should be considered when relying on numeric comparison for portable REXX code?