To tolerance irregularity you have a couple of options. 

1) First, for Standard surfaces in a typical single-pass layout, you can use the TIRR tolerance operand which will automatically convert a standard surface to an irregular surface, then apply spherical and astigmatism.

2) Alternatively, you can manually convert the standard surfaces to irregular surfaces yourself.  Then use the TPAR operand to set the range of variation applied to the Zs, Za, Zc & Angle parameters for each surface.  Assuming the corresponding error in fringes is measured in double-pass (e.g., Newton’s ring test), then the “Z” values can be converted to fringes by multiplying by (2/lambda).  Likewise, fringe values can be converted into Z values.  For example, if you want a half-fringe of spherical plus a half-fringe of astigmatism (for a total of one fringe of irregularity), then Zs = Za = 0.5*(lambda/2).  Note that the Z values are expressed in terms of lens units (typically mm).  So, let’s say that lambda = 5e-04 mm, and the total irregularity for any given surface can vary randomly between 0.5-fringe and 1.0-fringe, then the tolerance operands for that surface (here it’s surface 9) would look like this:

More generally, you can include coma if you think that’s appropriate, and vary the relative contribution of all three surface error types. 

Since you have a double-pass system, this second approach, in which the Irregular surface type is explicitly used, is probably the best way to go.  That way you can tolerance the forward-pass surfaces and use pick-ups to ensure the reverse path properly replicates the surface perturbations applied in the forward direction.