If Number is an integer, the result is a tight breal whose float bounds enclose the integer. If the integer's magnitude is small enough to be accurately representable with a float, those bounds will be identical and the breal will have zero width. For IEEE 754 double representation the threshold is 9007199254740992.
If Number is a rational, the result is a breal whose float bounds enclose the exact value of the rational.
If Number is a float, the result is a zero-width breal with both bounds identical to Number. No outward-rounding is done, because the system has no way of knowing how inexact the float argument is. It therefore makes the (unrealistic) assumption that the value is accurate. [To manually construct wider intervals from a float, use breal_from_bounds/3. Moreover, the parser can configured to read numeric constants directly as safely rounded breals; see syntax_option read_floats_as_breals]
If Number is a breal, the result is Number itself.
Note: The implementation may sometimes round conservatively and not give the tightest possible result.
% small integers and floats are assumed to be accurate:
?- Result is breal(25).
Result = 25.0__25.0
?- Result is breal(1.5).
Result = 1.5__1.5
% rationals are conservatively rounded:
?- Result is breal(3_4).
Result = 0.74999999999999989__0.75000000000000011
% identity operation on breals:
?- Result is breal(1.0__1.01).
Result = 1.0__1.01
% rounding with large integers:
?- Result is breal(9999999999999999).
Result = 9999999999999998.0__10000000000000002.0